A Comprehensive Review of Advanced Sensor Technologies for Fire Detection with a Focus on Gasistor-Based Sensors

Abstract

Early fire detection plays a crucial role in minimizing harm to human life, buildings, and the environment. Traditional fire detection systems struggle with detection in dynamic or complex situations due to slow response and false alarms. Conventional systems are based on smoke, heat, and gas sensors, which often trigger alarms when a fire is in full swing. In order to overcome this, a promising approach is the development of memristor-based gas sensors, known as gasistors, which offer a lightweight design, fast response/recovery, and efficient miniaturization. Recent studies on gasistor-based sensors have demonstrated ultrafast response times as low as 1–2 s, with detection limits reaching sub-ppm levels for gases such as CO, NH₃, and NO₂. Enhanced designs incorporating memristive switching and 2D materials have achieved a sensitivity exceeding 90% and stable operation across a wide temperature range (room temperature to 250 °C). This review highlights key factors in early fire detection, focusing on advanced sensors and their integration with IoT for faster, and more reliable alerts. Here, we introduce gasistor technology, which shows high sensitivity to fire-related gases and operates through conduction filament (CF) mechanisms, enabling its low power consumption, compact size, and rapid recovery. When integrated with machine learning and artificial intelligence, this technology offers a promising direction for future advancements in next-generation early fire detection systems.

1. Introduction

Fire is one of the most catastrophic hazards worldwide, causing thousands of fatalities each year and leading to significant property and environmental damage. For instance, fire-related incidents in India alone accounted for an average of 16,714 deaths per year between 2014 and 2018, with even more widespread repercussions observed worldwide [1]. While fire plays a vital role in human civilization and industrial processes, its uncontrolled manifestation poses critical safety concerns, particularly in modern infrastructure composed of complex and often highly flammable synthetic materials [2]. Fire detection systems have, therefore, evolved as an indispensable aspect of building safety, industrial monitoring, and public protection. Traditional fire sensing methods primarily rely on the detection of smoke, heat, or visible flame signatures through ionization, photoelectric detectors, and thermal detectors [3]. However, these systems often demonstrate delayed response times, inability to detect early-stage fires, and susceptibility to false alarms-particularly in environments where visibility is poor or where sensor performance is affected by dust, humidity, or aerosol interference [4,5]. It is essential to distinguish between sensors and detectors in fire safety systems. A sensor is a device that measures physical or chemical parameters, such as temperature, gas concentration, or light intensity. A detector, on the other hand, interprets signals from one or more sensors and makes decisions such as triggering alarms or initiating emergency protocols [6]. This distinction ensures clarity in describing fire detection mechanisms throughout this review. Conventional detectors, such as non–dispersive infrared (NDIR) and electrochemical gas sensors, are generally reactive rather than predictive; they are typically triggered after combustion is well underway. Moreover, many such systems lack adaptability in dynamic conditions and fail to distinguish between nuisance signals and genuine fire hazards. Key fire parameters, such as the heat release rate (HRR), gas emissions (e.g., CO, CO₂, NO_x, and HCN), and smoke composition during various fire stages (smoldering vs. flaming) are not adequately addressed by these traditional systems [7,8].

Recent developments in material science, the IoT, machine learning (ML), and microelectronics have ushered in a new era of intelligent and integrated fire detection systems. These modern systems utilize multisensor fusion to analyze chemical, thermal, optical, and visual indicators of fire, while employing intelligent algorithms to enhance detection accuracy and minimize false positives. Among the emerging technologies, gasistor-based fire sensors–which leverage a CF mechanism within memristive structures–have gained significant attention. Compared to traditional metal oxide semiconductor (MOS) sensors, gasistors offer rapid response and recovery, high selectivity, low power operation, and compact scalability, making them particularly well-suited for embedded and IoT-integrated fire detection platforms [1,7,9,10]. In recent years, gas sensors utilizing memristor technology have gained considerable attention for environmental monitoring, particularly with the incorporation of advanced materials, such as two-dimensional structures and heterostructures. Recent advancements in memristor-based gas sensors have laid a strong foundation for the emergence of gasistors in fire detection applications. Earlier studies have explored both simulation and experimental approaches, particularly focusing on TiO₂-based memristive devices, which demonstrated notable sensitivity and response rates to gases, such as H₂, ethanol, ammonia, and C₂H₆. By integrating gas-sensing and memory functionalities into a single device, gasistors enable the detection and recording of early-stage fire indicators, such as toxic gases released during combustion. Gasistor-based systems show pronounced sensitivity to early-stage fire gases, such as CO and NO_x, which are often released during the pyrolysis and smoldering stages, before the onset of open flames. For instance, a TiO₂-based gasistor exhibited a response of 164.2 to 1 ppm of NH₃, with the response and recovery times under 1 s [11]. Additionally, the incorporation of materials like HfO₂ has been shown to enhance device stability and reproducibility, crucial for reliable fire detection [12]. The low power consumption and compact size of gasistors make them suitable for integration into Internet of Things (IoT) platforms, facilitating real-time monitoring and early warning systems in fire safety applications. Their unique electrical characteristics also support nonlinear, memory-based sensing, enabling synergistic integration with AI-driven analytics and smart monitoring networks for real-time fire risk prediction [13]. Additionally, their low energy demand and flexible design make them viable for use in portable, wearable, and embedded applications across industrial and residential domains.

Beyond gasistors, other emerging sensor platforms, such as quantum dot (QD)-based and biosensor-based fire detectors are drawing attention. QD-based fire sensors, particularly those using PbS colloidal quantum dots, offer high responsivity in the near-infrared range, enabling early flame detection even under ambient lighting. Their compatibility with silicon and solution processability supports scalable, low-cost fabrication [14]. Similarly, biosensors show promise due to their molecular specificity and rapid biochemical response, potentially allowing for the detection of combustion byproducts, such as VOCs or gas-phase biomarkers. However, both technologies currently face challenges related to long-term stability, environmental sensitivity, and limited validation in real-world fire scenarios.

Despite notable progress in recent years, a comprehensive review that critically evaluates gasistor-based fire sensors in the broader context of advanced fire sensing technologies remains lacking. The current literature often isolates discussions of individual sensor types or focuses on specific methods, such as vision-based fire detection (VFD) or infrared flame sensing [4]. To address this gap, the present review systematically examines the state-of-the-art fire detection technologies, emphasizing the advantages, challenges, and future prospects of gasistor-integrated platforms.

This paper begins by discussing the core indicators of fire smoke, thermal energy, and gaseous emissions, as well as the associated detection techniques. It then focuses on the design, functionality, and performance metrics of gasistor-based sensors, including their selectivity, sensitivity, response time, reversibility, stability, and energy efficiency. Finally, it explores AI- and IoT-enhanced integration, highlights current challenges, and proposes future research directions to promote the development of next-generation, smart fire detection systems.

2. Fundamentals of Fire Detection and Sensor Technology

2.1. Fire Characteristics and Indicators

The primary indicators of fire characteristics typically include smoke, thermal energy release, and the emission of various gases. These elements play a crucial role in understanding fire dynamics, enabling the detection, assessment, and management of fire incidents. Smoke provides early visual and chemical signals, while thermal energy release reflects the fire’s intensity and progression through its different stages. Additionally, the release of gases, like carbon monoxide, carbon dioxide, and other combustion byproducts, offers vital information about the fire’s combustion efficiency and material composition. The following sections provide a detailed analysis of each of these indicators.

2.1.1. Smoke

Smoke is one of the earliest indicators of fire and plays a critical role in fire detection systems. It is a complex combination of the gases, liquid droplets, and solid particles produced during combustion [15]. The composition and characteristics of smoke vary depending on the materials being burned and the environmental conditions [16]. For example, modern building materials, such as plastics, release hazardous chemicals like hydrogen cyanide, hydrochloric acid, and volatile organic compounds (VOCs) during combustion, posing significant health risks to respiratory and cardiovascular systems. Smoke contains particulate matter of varying sizes, with fine particles (PM2.5) being especially dangerous due to their ability to penetrate deep into the lungs. These particles can lead to serious health issues, including respiratory diseases, cardiovascular problems, cognitive impairments, and low birth weights [17]. Furthermore, smoke’s behavior within structures is influenced by architectural features and environmental conditions. In confined spaces, like tunnels, high-temperature smoke oscillates between walls, creating multiple high-temperature zones that complicate evacuation planning and response strategies. Beyond health implications, smoke has environmental consequences, as large-scale fires can disperse smoke over vast distances, impacting regions far from the fire source. For instance, smoke from Australian bushfires has been detected over Antarctica, contributing to air pollution and accelerating ice melt due to the deposition of light-absorbing particles.

The early detection of smoke is crucial for effective fire suppression, successful firefighting, and improved survival rates. Techniques such as optical methods and visual surveillance systems are used for smoke detection [18]. Optical smoke detectors utilize the interaction of light or electromagnetic radiation with smoke particles to identify the presence of smoke. These detectors are highly sensitive to flaming fires, operate across a wide temperature range, and can detect particles between 1 to 10 μm. However, smaller particles and interference from dust and aerosols may reduce their effectiveness and lead to false alarms. Visual surveillance systems, on the other hand, rely on imaging technologies to monitor changes in the environment caused by smoke [19]. Despite the challenges, smoke detection systems play a crucial role in minimizing the impact of fires on human health and safety, while also helping to reduce their environmental effects. Understanding the characteristics of smoke and improving detection technologies remains essential for enhancing fire safety measures.

2.1.2. Thermal Energy

Thermal energy generated during combustion serves as a fundamental parameter for detecting fires and understanding their progression. The release of significant heat during fire incidents provides critical insights into the fire’s intensity, growth stage, and potential spread. The real-time monitoring and analysis of thermal energy are necessary for early fire detection and the deployment of effective suppression strategies. Fires generally advance through four distinct stages: initial growth, fire growth, full development, and decay. Each stage is marked by unique thermal patterns and rates of heat release, as shown in Figure 1 [20]. The accurate assessment of these stages is crucial for understanding fire dynamics and mitigating their effects.

A primary metric for evaluating thermal energy in fire scenarios is the heat release rate (HRR), which quantifies the energy output per unit time during combustion [21]. The HRR is a key parameter that directly impacts a fire’s intensity, growth, and spread. It is measured in kilowatts (kW) or megawatts (MW). It serves as a cornerstone for fire detection, suppression strategies, and hazard assessment. The HRR is determined by several factors, including the type and properties of the fuel involved, the availability of oxygen, and the environmental conditions during combustion.

The chemical composition and heat of combustion of fuels significantly impact the HRR. Hydrocarbon-based fuels, for instance, typically exhibit higher heats of combustion compared to cellulosic materials, resulting in a greater HRR. Additionally, the physical properties of the fuel, such as size, shape, and surface area, play a vital role in determining combustion rates. Smaller particles with larger surface areas combust more rapidly, leading to higher HRR values. The availability of oxygen is another critical factor; fires in well-ventilated environments often achieve higher HRRs due to efficient oxidation processes. Conversely, oxygen-limited conditions can result in incomplete combustion and a reduced HRR. The HRR is further influenced by environmental factors, including the ambient temperature and humidity [21]. Elevated ambient temperatures can preheat the fuel, lowering the ignition energy requirement and potentially increasing the HRR, whereas a high moisture content in the fuel can absorb heat and reduce the HRR.

Effectively managing and controlling the HRR is essential for successful fire suppression and mitigation. Several strategies can be employed to limit the HRR and reduce the intensity of fires. Fuel management, which involves minimizing the availability of flammable materials, can significantly decrease the HRR potential in a given area. Utilizing construction and manufacturing materials with lower combustibility can inherently limit the HRR during a fire event. Regulating airflow to the fire is another effective approach, as reducing oxygen availability can lower the HRR. This principle underlies suppression systems that seal the fire environment to starve it of oxygen.

Heat absorption methods, such as the application of water, are also widely employed to reduce the HRR. Water absorbs significant amounts of heat, lowering the temperature of both the fire and surrounding materials. This effect slows down the combustion process. Similarly, chemical suppressants can disrupt the chemical reactions of combustion and absorb heat, thereby decreasing the HRR [21].

Understanding the factors influencing the HRR and implementing targeted strategies to manage it are critical components of fire management and suppression efforts. A fire intensity can be controlled by addressing key variables, such as fuel properties, oxygen availability, and thermal absorption. This helps reduce the risk of rapid fire spread and improves the effectiveness of firefighting measures. This approach underscores the importance of HRR as a central parameter in fire detection and suppression systems [21].

2.1.3. Gas Emissions

Modern buildings contain a wide variety of materials, and their combustion releases various toxic products, including aerosols and gases. Even materials that are not burning can undergo thermal decomposition or pyrolysis, releasing additional hazardous emissions. These combustion products pose significant health risks to occupants and emergency personnel, which can be categorized as follows:

Irritants: Gases and particles can irritate the respiratory system and hinder escape. At high concentrations, they may lead to incapacitation or death.

Asphyxiants: Gases that affect the central nervous system, causing disorientation, loss of consciousness, and potentially death.

Thermal Effects: Skin and respiratory tract burns, along with elevated body temperature (hyperthermia).

The impact of these hazards depends on factors such as the exposure duration, gas concentration, and individual health conditions. Eye irritation and smoke can further delay escape, increasing the exposure time and risks. Fires also produce toxic emissions during pyrolysis, defined as the heat-induced decomposition of materials, and combustion, a rapid oxidation process generating heat and light. The increased use of polymeric materials in modern products has led to higher volatile emissions in fires. These materials release volatile compounds during thermal decomposition, even before combustion begins. Smoldering fires, which occur in porous or densely packed materials, generate different products than open-flame fires. At lower temperatures (around 400 °C), smoldering combines pyrolysis and oxidation, producing a high CO/CO₂ ratio with carbon monoxide (CO) as a primary toxicant. Due to the slow evolution, low temperatures, and less dense smoke, smoldering fires can be particularly dangerous for sleeping occupants, often leading to fatalities from asphyxia. Fires that smolder for over 30 min before detection are known to cause more casualties than rapid-flame fires [3].

Table 1. Possible combustion products depending on the combustion material [22,23].

Test-Fire	Combustion Material	Combustion Products
TF1	Open beech-wood fire (with ethanol)	CO2, CO, H2, H2O, NO, NO2, methane, acetylene, ethane, ethene, styrol, chlorobenzene, ethanol, etc.
TF2	Smoldering beech-wood fire	CO, H2, H2O, CO2, NO, HCl, methane, ethane, ethene, aromatic hydrocarbons (benzene, toluol, xylene, styrol, chlorobenzene), acrolein, formaldehyde, formic acid, acetic acid, etc.
TF3	Glowing smoldering cotton	CO, H2, NO, CO2, H2O, methane, acetylene, ethane, ethene, benzene, acetaldehyde, formaldehyde, ethanol, etc.
TF4	Open polyurethane-foam fire	CO2, CO, H2O, NO, NO2, N2O, HCN, NH3, amines, ethane, styrol, acetone, etc.
TF5	Open n-heptane fire (3% toluol)	CO2, CO, H2, H2O, NO, NO2, ethane, ethene, styrol, chlorobenzene, ethanol, octane-n, hexane, etc.
TF6	Open ethanol/alcohol fire (liquid)	CO2, CO, H2, H2O, NO, ethane, styrol, ethanol, etc.

, below, lists the toxic gases that can be produced by different combustion materials, as summarized by O. Linden and J.T. Meyer.

3. Main Toxicants for Fire Emissions

Fires typically release a variety of gas particles, depending on the material being burned. Among these, carbon oxides (CO₂ and CO) and nitrogen oxides (NO and NO₂) are frequently generated. Notably, certain gas particles are generated in consistent patterns, regardless of the nature of the fire or the combustion environment. This indicates that the early and accurate detection of these common combustion products could enable the development of a universal response system applicable to all types of fires.

3.1. Carbon Dioxide

Smoke is a mixture of solid and liquid particles suspended in the air, along with gases produced during pyrolysis or combustion. In flaming fires with good ventilation, most of the carbon from the burning materials is converted into carbon dioxide (CO₂). This transformation continues even in post-flashover fires, where a significant amount of carbon is still converted into CO₂. Consequently, the CO₂ yield is often used to estimate the burning rate of materials when it is not feasible to directly measure the mass loss [24].

3.2. Carbon Monoxide

While both carbon dioxide (CO₂) and carbon monoxide (CO) are produced in fires, carbon monoxide presents a far greater hazard to human health. CO is generated in both types of combustion, though its production rate in smoldering fires is relatively slow. However, smoldering fires lack sufficient mixing with ambient air, allowing dangerous concentrations of CO to accumulate near the ignition source within approximately 10 min. It may take 1 to 3 h for lethal concentrations of CO to spread throughout the room, by which time, smoldering may either subside or transition into flaming combustion. In flaming fires, CO production occurs in the gas phase. During combustion, fuel vapors or carbon-based decomposition products interact with oxygen in a series of complex reactions to produce CO. This CO is further oxidized into CO₂, depending on the availability of oxygen in the local environment. A limited oxygen supply—caused by either a reduction in the oxygen concentration or restricted airflow—can disrupt this process, leading to increased CO formation. The formation of CO is closely related to the fuel-to-air equivalence ratio [25].

3.3. Hydrogen Cyanide

The production of hydrogen cyanide (HCN) in fires depends on both the material composition and the temperature. HCN is produced when nitrogen-containing polymers decompose, either during smoldering fires or through pyrolysis in flaming fires. It can also form during the flaming combustion of these materials. However, there is no evidence to suggest that toxicologically significant amounts of HCN are formed through the fixation of nitrogen from the air. Unlike carbon monoxide (CO), there is a lack of comprehensive studies to enable accurate quantitative predictions of HCN generation in fires. When sufficient oxygen is available, nitrogen-containing materials may also produce nitrogen oxides (NO_x). Additionally, HCN has been observed to oxidize into NOx in scenarios where flames extend beyond a flashover room and continue burning outside the doorway [3].

3.4. Halogen Acids

Polymer systems that contain halogen elements, such as fluorine, chlorine, or bromine, produce halogen acids, including hydrogen fluoride (HF), hydrogen chloride (HCl), and hydrogen bromide (HBr), during thermal decomposition. The formation of these acids is primarily dependent on the material composition and occurs during the pyrolysis phase of combustion, even without the presence of flames. Since the production efficiency of halogen acids is nearly complete, high yields are expected in fires. For instance, polyvinyl chloride (PVC) releases HCl at temperatures ranging from 225 °C to 275 °C (437 °F to 527 °F) [26]. In the early stages of PVC decomposition, HCl may be released in amounts exceeding stoichiometric predictions due to the low combustion of carbon in the material. However, PVC with a high calcium carbonate content shows reduced HCl yields. This occurs because the chlorine reacts with calcium to form calcium chloride, a stable solid at typical fire temperatures [27]. In addition, at temperatures below 100 °C (212 °F) and in the presence of adsorptive surfaces or water droplets, the concentration of halogen acids decreases rapidly [28].

3.5. Nitrogen Oxides

The FTIR analysis of fire effluents has shown that nitrogen oxides are primarily present as nitric oxide (NO). This gas remains stable at the low concentrations and temperatures typically encountered by humans during fire events. Nitric oxide is also found in tobacco smoke and motor vehicle emissions. Additionally, nitrogen dioxide (NO₂) may be present. NO₂ is highly water-soluble and acts as an acidic irritant with significantly higher toxicity compared to NO, posing serious health risks [29].

The composition of the burning materials can result in the generation of additional toxic substances in fire smoke. For example, phosphorus-based fire retardants may generate phosphoric acid aerosols, while sulfur-containing polymers can release sulfur oxides. However, there is no dependable method available to predict the yield of these particular toxic byproducts.

4. Key Requirements for Fire Detectors

In recent years, fire detectors have advanced considerably, integrating cutting-edge technologies that enhance accuracy and minimize false alarms [30]. Modern fire detectors utilize scattering methods, such as forward and backward scattering, combined with heat detection to identify smoke. To minimize false alarms caused by dust or steam vapors, multiple wavelengths, including infrared and blue light, are used, enhancing the accuracy of fire detection [31].

There are various types of fires, including those involving ethanol, that produce non-smoking flames, making them challenging for traditional fire detectors to detect. To address this challenge, the integration of gas sensors with varying wavelengths proves useful in detecting combustible gases like carbon monoxide (CO), nitrogen dioxide (NO₂), and even non-smoking fires like ethanol [32]. These advanced sensors offer promising solutions for early fire detection while minimizing false alarms [2].

Moreover, integrating advanced algorithms with multi-component sensor systems enables the more-precise identification of fire types and faster response times. By analyzing changes in smoke, temperature, and gas concentrations, these systems can adapt to different environmental conditions, reducing the impact of signal fluctuations. As fire detection technology continues to improve, the focus is shifting toward minimizing false alarms, enhancing detection accuracy, and ensuring quicker response times: crucial factors for both safety and cost-efficiency in high-risk environments, such as in aircraft or industrial settings [33,34].

The sensors used in high-risk environments must meet several key requirements to ensure efficiency and reliability. They should be low-cost to enable mass production and widespread accessibility, while also incorporating a self-check feature to maintain functionality without manual intervention [35,36]. With low power consumption (less than 1 mW), these sensors should operate on a battery backup [37,38]. Their compact design allows for easy integration into existing housing and sockets in residential environments. The sensors must be highly sensitive to hazardous gases, such as carbon monoxide (CO), carbon dioxide (CO₂), and nitrogen dioxide (NO₂), while avoiding interference from substances like humidity or cooking fumes to reduce false alarms [39]. Lastly, long-term stability is essential, ensuring consistent performance over a decade and minimizing the need for maintenance or replacement [40]. In real-world fire detection scenarios, sensors are expected to perform reliably over extended periods, despite exposure to temperature fluctuations, humidity, and air contaminants. Long-term instability can lead to signal drift, reduced sensitivity, or increased false alarm rates. For example, metal oxide semiconductor (MOS) sensors are particularly susceptible to humidity-induced baseline shifts. Water molecules adsorbed on the sensor surface can block reaction sites, leading to the under-measurement of target gas concentrations and baseline drift over time [41]. Additionally, variations in ambient temperature can affect MOS sensor performance, with different sensor types exhibiting varying responses to temperature changes [42]. Electrochemical sensors also face degradation over time due to factors like electrolyte evaporation and electrode passivation, impacting their long-term stability [2]. Reports in the literature suggest that sensor coatings, calibration mechanisms, and encapsulation strategies can improve durability. For instance, the use of MoS₂/graphene oxide composites in humidity sensors has shown enhanced stability and sensitivity [43]. However, standardized, long-term validation data remains limited. Addressing these stability concerns through material engineering and robust testing protocols is essential for transitioning laboratory-grade sensors to dependable fire safety solutions in the field [2]. These factors collectively enhance the safety, dependability, and fire prevention capabilities of residential fire detectors.

4.1. Sensitivity and Selectivity

Fire sensors detect the presence of fire by monitoring environmental changes, such as smoke, heat, or flames. Sensors can be categorized into two main types: high-sensitivity sensors and low-sensitivity sensors. High-sensitivity sensors, as shown by Lee et al. [44], are capable of detecting fires in their early stages. They fabricated metal oxide nanocolumns of NiO, SnO₂, WO₃, and In₂O₃ using glancing angle deposition (GLAD). These sensors can detect gases like HCl, CO, and VOCs, which are released during the decomposition of polyvinyl chloride (PVC). Among the materials tested, SnO₂ exhibited the highest sensitivity at 200 °C, with a maximum response of 2.1. In contrast, NiO showed significantly higher sensitivity at 350 °C, as shown by its response of 577.1 in Figure 2.

Li et al. [45] also fabricated a fire detection system using flame-retardant cellulose paper incorporated with graphene oxide (GO) and two-dimensional titanium carbide (Ti₃C₂, MXene). Cellulose paper is first immersed in a phytic acid (PA) solution. Next, varying ratios of graphene oxide (GO) are combined with MXene to create a homogeneous suspension, resulting in MGO. Using a dip-coating method, the PA is immersed in the MGO suspension to produce PA@MGOy. The PA@MGOy system is highly sensitive to gases, like carbon monoxide (CO) and volatile organic compounds (VOCs), emitted during combustion, detecting these gases within 2 s at 250 °C, making it ideal for early fire detection. Figure 3 shows how the electrical resistance of the sensors changes at various temperatures. The PA@MGO₅ exhibits the fastest response time of just 2 s at 250 °C, demonstrating its exceptional sensitivity for early fire detection.

To detect fires in their early stages, highly sensitive fire sensors are designed to identify even minute amounts of smoke or gas emissions, making them ideal for environments where early detection is crucial, such as residential buildings, hospitals, and data centers. However, due to their high sensitivity, these sensors are more susceptible to false alarms caused by non-fire-related particles, like dust or steam. On the other hand, low-sensitivity fire sensors require higher concentrations of smoke or gases to detect a fire, leading to a longer detection time. These sensors are more suitable for environments such as industrial settings, kitchens, or workshops, where minimizing false alarms is a priority. In these scenarios, low-sensitivity sensors are less likely to respond to non-fire-related particles [3].

Figure 4 shows the relationship between the sensor input and time. A short response time indicates that the sensor can detect fires at an early stage, signaling high sensitivity. In contrast, a longer response time indicates a lower sensitivity, meaning the sensor requires higher concentrations of smoke or gases to detect a fire [46].

The environment contains various gases such as oxygen, nitrogen, argon, and carbon dioxide. Selectivity refers to a sensor’s ability to distinguish accurately between the gases released during a fire and those that are naturally occurring in the environment. Sensors typically have limited selectivity, often responding to non-fire gases and leading to false alarms [47]. To reduce these false alarms, different sensing materials are utilized to better distinguish between gases.

Table 2. Sensing materials capable of detecting gases emitted during fire.

Fire Gases	Types of Sensing Materials to Detect Gas	Reference
CO	SnO2, WO3, In2O3, ZnO, α–Fe2O3, NiO, Co3O4, Cr2O3	[48,49,50,51,52,53,54,55]
CO2	SnO2, ZnO, TiO2, La2O3, CdO, CeO2, In2O3, WO3	[56]
NOx	ZnO, In2O3, SnO2, ZnCr2O3, WO3/YSZ/Pt, Pt–WO3/TiO2, yttria-stabilized zirconia	[57]
HCN	CuO, Ni(OH)2, AiN nanotube, Boron sheet	[58]
SOx	Metal Sulfides, stabilized zirconia (YSZ)/metal sulphates	[59,60]
HCl	Nd–SrCeO3, polyaniline copolymer nanocomposite thin films	[61,62]

presents a range of sensing materials capable of detecting gases emitted during a fire.

Achieving the right balance between sensitivity and selectivity is crucial for ensuring prompt fire detection while minimizing the risk of false alarms.

4.2. Response and Recovery Time

The response time of fire sensors refers to how quickly the sensor can detect fire or hazardous conditions, such as smoke, heat, or gases, and trigger an alarm. Ideally, fire detectors should have a short response time to identify a fire in its early stages. However, there are several factors that need to be considered. As shown in Figure 4, a shorter response time increases the likelihood of false alarms, whereas a longer response time helps minimize false alarms. Therefore, optimizing the response time is critical for the effectiveness of fire detectors. Tan et al. [63] utilize carbon nanotubes (CNTs) doped with iron oxide (Fe₂O₃) to detect acetone gas, a common volatile organic compound (VOC) found in fire gases. As shown in Figure 5, the response time of the sensor varies at different temperatures. The sensor performs optimally at a sintering temperature of 250 °C, achieving a response time of approximately 4 s.

Recovery time refers to the duration it takes for a fire detection system to reset itself after the fire has been addressed. As shown in Figure 6, once the device reaches its maximum conductance, it requires a certain amount of time to return to its baseline state. This period is known as the recovery time [64]. Tan et al. also calculated the recovery time of the CNT/Fe₂O₃ sensor. As shown in Figure 5, the sensor’s recovery time is between 8 and 14 s across all the tested temperatures.

To enhance the clarity and comparative understanding of different transducer materials used in fire-related gas detection, we have provided a dedicated table (

Table 3. Comparative overview of transducer materials used in fire-related gas detection, detailing sensing materials, target gases, typical response and recovery times, and associated operational constraints.

Sensing Material	Target Gas	Response Time (s)	Recovery time (s)	Constraints/Limitations	Ref
SnO2	CO, C2H6, VOCs	10–30	20–40	Requires high temperature (~200 °C); affected by humidity.	[65]
TiO2	NH3, CO	1–2	5–10	Sensitive to humidity; can operate at RT with nanostructures.	[66]
In2O3	NO2, CO	~15	~25	Moderate selectivity may require doping for enhanced performance.	[47]
MXene	CO, VOCs	<2	<5	Prone to oxidation; requires stabilization.	[67]
NiO	H2, CO	3–8	10–20	Operates at higher temperature (~300–350 °C); moderate stability.	[68]
WO3	NO2	~ 30	~60	Good sensitivity but slower recovery; temp-dependent.	[69]
Graphene/GO/rGO	NH3, VOCs, CO	<5	<10	Highly sensitive; humidity can impact response.	[70]
Ta2O5	NO, C2H6	<1	>300	Excellent sensitivity but slow natural recovery; enhanced by bias.	[71]

) summarizing key sensor characteristics. This includes the sensing material, corresponding target gases, typical response and recovery times, and notable operational constraints. Each material exhibits distinct advantages and limitations based on its physicochemical properties and operating conditions. For instance, metal oxides, like SnO₂ and NiO, offer high sensitivity to combustion gases, such as CO and H₂, but typically require elevated temperatures for optimal performance. In contrast, emerging materials, such as MXene and TiO₂-based nanostructures, enable room temperature operation with rapid response, although their long-term stability or environmental susceptibility may pose challenges. The tabulated comparison offers a concise reference for evaluating material specific trade-offs and supports the selection of suitable sensing platforms for next-generation fire detection systems.

5. Conventional Sensor Technologies for Fire Detection

Fire detection technologies have evolved significantly over time, with some methods still in use, while others have become obsolete. These systems rely on various sensors, including smoke, heat, gas, and flame detectors, each designed to convert physical parameters into electronic signals for processing and alarm activation [3,44,72]. Smoke detectors, operating on ionization or photoelectric principles, are particularly effective in identifying fires during the early smoldering stages [73]. Thermal sensors, among the earliest automated devices, are available in fixed-temperature, rate-of-rise, and line-type designs, with the latter utilizing heat-sensitive insulated cables. Despite their simplicity and affordability, these sensors activate only after reaching predefined temperature thresholds. Flame detectors, known for their reliability in energy and transportation applications, require a direct line of sight and are comparatively costly. Gas sensors, particularly those based on semiconductor metal oxides, have gained prominence due to their sensitivity, compact size, and affordability, though stability issues remain a challenge. Emerging technologies, such as graphene oxide-based sensors, show promise with rapid response times, while advances in the IoT, wireless sensor networks, and deep learning algorithms offer opportunities for enhancing detection accuracy. Visual techniques for detecting smoke and flames are increasingly favored for their swift response and low error rates, complementing non-visual approaches, like infrared and microwave-radiometer-based systems, which are capable of detecting fires through barriers. These advancements underline the critical role of multisensor fusion in modern fire detection systems.

5.1. Smoke Detectors

Smoke sensors play a critical role in fire detection by identifying the airborne particulates produced during combustion [15]. These devices leverage various techniques, including optical and ionization methods, to measure parameters such as the smoke concentration, size distribution, and volume fraction [74,75]. Optical sensors detect smoke by measuring the scattering or absorption of light by smoke particles [18], while ionization sensors rely on changes in electrical currents caused by smoke particles interacting with ions in an ionization chamber [76]. Advanced systems often combine multiple sensors to improve their sensitivity and the reduce false alarms caused by non-combustion sources, like dust and water vapor [77,78]. Innovations, such as integrating visual surveillance or employing algorithms like convolutional neural networks (CNNs), have further enhanced the accuracy and applicability of smoke detection systems in diverse environments [79].

5.2. Types of Smoke Detection Techniques

5.2.1. Optical Smoke Detection

Optical smoke detectors measure the scattering or obscuration of light by smoke particles [15]. They are sensitive to the larger particles produced by smoldering fires and are commonly used in residential and commercial settings [76]. These detectors use a light emitter and a photo detector, triggering an alarm when scattered light reaches a predefined threshold, as shown in Figure 7. Advanced optical methods incorporate multi-angle scattering data and multi-wavelength illumination to reduce the false alarms caused by dust or aerosols [80].

Light scattering by particles depends on their size, shape, refractive index, wavelength, and scattering angle [1]. A multi-sensing approach improves smoke detector sensitivity and reduces false alarms. Spectral analysis in two stages—data acquisition and fire detection—enhances detection accuracy [81]. Dual-polarized channels adjust the sensitivity for white and black smoke. Signal patterns from fires and non-fire sources aid in minimizing false alarms [80], while particle size distribution analysis further refines detection. A system using optical components provides data on burning materials [82].

5.2.2. Ionization Smoke Detection

Ionization smoke detectors rely on the ionization of air molecules using a radioactive source, typically Americium–241 [76]. The presence of smoke particles alters the ion flow, reducing the current and activating the alarm [74]. These detectors are more responsive to the smaller particles generated by flaming fires but are less commonly used today due to the complexities of handling radioactive materials [76]. Table 4 shows a comparison of fire detection techniques, based on their principles, applications, advantages, and limitations.

5.2.3. Photoelectric Smoke Detectors

Multisensing Approaches for Enhanced Detection

To improve the accuracy and reliability of smoke detection, multisensor systems are employed [75,83]. These systems combine optical, ionization, and gas sensing techniques, often integrated with advanced algorithms, to detect smoke while minimizing false alarms [78,80]. For example, photoelectric detectors combined with thermal or gas sensing can differentiate between fire-related smoke and non-combustion sources, like cooking aerosols [76]. Studies also suggest that multisensory methods utilizing spectral and polarization data can enhance sensitivity and specificity [81,84]. In order to reduce false alarms, Gottuk et al. [75] introduced multisensory smoke sensing techniques that use different smoke detection configurations, like photoelectric-based gas sensing and ion-sensing gas sensing. In order to determine the rate of fire spread, Jeong et al. [85] constructed a fire field model in addition to studying the smoke path. A smoke particle detector using an amorphous silicon sheet as the radiation source was proposed by Liu et al. [86].

5.2.4. Advanced Smoke Detection Techniques Using Visual Surveillance

As long as there is sufficient ambient light, visual surveillance systems can detect smoke [87]. In order to detect smoke, these systems use CCTV cameras connected to a processing unit that separates image elements such as color, texture, motion, and energy analysis [87]. To improve detection accuracy, sophisticated techniques have been used, such as the statistical assessment of color models and dynamic texture analysis [88]. When adequate datasets are available, techniques such as Adaboost, in conjunction with staircase discovery approaches, allow for effective smoke detection [89]. The system’s capacity for smoke and flame detection is further enhanced by morphological processes and motion cues [90]. Figure 8 shows the basic design of the convolutional neural networks (CNNs) [2]. CNNs and the YOLOv2 algorithm are two recent developments that have greatly enhanced real-time detection capabilities and decreased false positives [84]. Convolutional neural networks (CNNs) have significantly advanced vision-based fire and smoke detection in surveillance systems, leveraging recent embedded processing developments [91]. Sergio et al. [84] introduced a real-time, YOLOv2, CNN-based smoke detection system compatible with standard surveillance cameras. Khan et al. [92] designed an early fire detection framework for CCTV cameras using fine-tuned CNNs, while also proposing a semantic segmentation model for hazy environments. However, because of their reliance on high frame counts, some techniques, such as Bayesian networks and SVM algorithms, continue to have limitations in terms of their response time [93]. These methods have been useful in a variety of contexts, such as historical sites, paper mills, and power plants [87].

Figure 8. A schematic representation of a fire detection system using a convolutional neural network (CNN). The model processes sensor inputs (e.g., gas concentration, temperature, or smoke) through an input layer, which feeds into multiple interconnected hidden layers to extract and analyze complex patterns. The final output layer classifies the result into two categories, “Fire” or “Non-fire”, enabling intelligent fire prediction and decision-making in real time. Reproduced with permission from [2].

Figure 8. A schematic representation of a fire detection system using a convolutional neural network (CNN). The model processes sensor inputs (e.g., gas concentration, temperature, or smoke) through an input layer, which feeds into multiple interconnected hidden layers to extract and analyze complex patterns. The final output layer classifies the result into two categories, “Fire” or “Non-fire”, enabling intelligent fire prediction and decision-making in real time. Reproduced with permission from [2].

Table 4. Comparison of fire detection techniques, based on principles, applications, advantages, and limitations.

Technique	Principle	Applications	Advantages	Limitations	References
Optical Smoke Detection	Measures light scattering or absorption by smoke particles.	Residential and commercial settings.	Sensitive to large particles; reduces false alarms with multi-wavelength illumination.	False alarms due to dust and aerosols; limited to smoldering fires.	[15,76,80,81]
Ionization Smoke Detection	Detects ion flow changes caused by smoke particles using radioactive sources (e.g., Americium–241).	Suitable for detecting flaming fires.	Highly responsive to small particles.	Handling radioactive materials; declining usage.	[74,76]
Multisensing Systems	Combines optical, ionization, and gas sensing techniques with algorithms.	Differentiates fire-related smoke from non-combustion sources, like cooking aerosols.	Reduces false alarms; enhances sensitivity and specificity.	Higher complexity; requires integration of multiple sensors and algorithms.	[75,78,81,84]
Advanced Visual Surveillance	Uses CCTV cameras and image processing (e.g., color, texture, and motion analysis).	Historical sites, paper mills, cement plants, power plants.	Real-time detection with CNNs and YOLOv2; reduced false positives.	Requires sufficient ambient light; limitations in low-light conditions.	[84,87,90,92]
Algorithmic Detection	Employs techniques like CNNs, Adaboost, and Bayesian networks for enhanced pattern recognition.	Both interior and outdoor scenarios with challenging environments, like haze.	High accuracy with adequate datasets; adaptable for diverse environments.	High computational requirements; response time limitations in some models.	[84,88,91,93]

Table 4. Comparison of fire detection techniques, based on principles, applications, advantages, and limitations.

Technique	Principle	Applications	Advantages	Limitations	References
Optical Smoke Detection	Measures light scattering or absorption by smoke particles.	Residential and commercial settings.	Sensitive to large particles; reduces false alarms with multi-wavelength illumination.	False alarms due to dust and aerosols; limited to smoldering fires.	[15,76,80,81]
Ionization Smoke Detection	Detects ion flow changes caused by smoke particles using radioactive sources (e.g., Americium–241).	Suitable for detecting flaming fires.	Highly responsive to small particles.	Handling radioactive materials; declining usage.	[74,76]
Multisensing Systems	Combines optical, ionization, and gas sensing techniques with algorithms.	Differentiates fire-related smoke from non-combustion sources, like cooking aerosols.	Reduces false alarms; enhances sensitivity and specificity.	Higher complexity; requires integration of multiple sensors and algorithms.	[75,78,81,84]
Advanced Visual Surveillance	Uses CCTV cameras and image processing (e.g., color, texture, and motion analysis).	Historical sites, paper mills, cement plants, power plants.	Real-time detection with CNNs and YOLOv2; reduced false positives.	Requires sufficient ambient light; limitations in low-light conditions.	[84,87,90,92]
Algorithmic Detection	Employs techniques like CNNs, Adaboost, and Bayesian networks for enhanced pattern recognition.	Both interior and outdoor scenarios with challenging environments, like haze.	High accuracy with adequate datasets; adaptable for diverse environments.	High computational requirements; response time limitations in some models.	[84,88,91,93]

5.3. Heat Detectors

Heat detection systems play a pivotal role in fire detection by identifying thermal energy changes in the environment, typically caused by combustion. These systems rely on sensors, such as heating elements paired with signal conditioning circuits, to measure ambient heat level. Heat detectors are broadly classified into static detectors, which respond to specific temperature thresholds, and rate-of-rise detectors, which react to rapid temperature increases. Advanced techniques, including thermal imaging, transform invisible radiation patterns into visible images for detailed feature analysis. Similarly, thermal radiation is instrumental in fire spread due to the high temperatures involved, and its spectral characteristics, particularly in the infrared range, can be exploited for intelligent detection [94,95,96]. Infrared-based algorithms compare emission intensities at specific wavelengths, such as 4.4 μm for CO₂ and 3.8 μm for non-flame emitters, to enhance accuracy while minimizing false alarms. These methods unambiguously distinguish flames from other heat sources, ensuring reliable detection, even in challenging environments.

5.3.1. Fixed Temperature Heat Detectors

Fixed-temperature heat detectors, also referred to as static heat detectors, activate when the surrounding temperature exceeds a predefined threshold [97]. These sensors are commonly designed to respond to temperature levels of 58 °C or higher, depending on the application [98]. Among the various types, fusible-element detectors are widely utilized and operate by melting the heating element at the threshold temperature, triggering the system. For instance, in fusible-element detectors, the melted element creates a short circuit within the sensing mechanism, as shown in Figure 9, making them non-restorable after activation. Additionally, these detectors are integral to systems like fire sprinklers due to their reliability in high-temperature scenarios [98]. A key concept in static heat detection is thermal lag, defined as the difference between the device’s activation temperature and the ambient temperature, which correlates directly with the rate of temperature increase [97]. Beyond fusible elements, other variants include distributed and bimetal sensors, expanding their adaptability to diverse setting [98].

5.3.2. Distributed Optical Fiber Heat Sensing Techniques

Distributed optical fiber heat sensing is a promising technology for fire detection and prevention, leveraging the fiber itself as a sensing module to measure heat distribution along its length [99,100]. Using the Raman effect, this technique processes the backscattered light intensity from optical pulses traveling through the fiber to determine the temperature at different points along the cable, as illustrated in Figure 10. Additionally, Brillouin backscatter has been proposed as an alternative for temperature detection, offering an effective replacement for Raman scattering in certain applications [101,102].

Hoff et al. [103] introduced a distributed temperature sensing (DTS) system utilizing fiber optics for detecting fires in industrial conveyor belts. This approach analyzes the ratio of anti-Stokes to Stokes Raman lines from backscattered spectra, a method illustrated in Figure 11, to accurately gauge the surrounding temperatures. Furthermore, acoustic-based fire measurement technology was also suggested by Hoff et al., expanding the potential applications of fiber optic systems in fire safety.

Research by other authors explored additional uses of distributed heat sensing, such as monitoring groundwater flow and heat transport in fractured media [104] and verifying surface flux and heat balance measurements at lake surfaces using temperature recordings from 0.5 m above to 1.5 m below the surface [105]. This diverse range of applications underscores the versatility and efficiency of optical-fiber-based temperature sensing for fire detection and beyond.

5.3.3. Thermal-Radiation-Based Techniques for Flame Detection

Flame detection using thermal radiation is one of the quickest methods and is widely employed in vital industrial settings, like aircraft hangars and petroleum production or storage facilities. A key challenge lies in differentiating flame radiation from other sources, such as heated surfaces or solar radiation. Although the fundamental principles of radiation-based flame detection have been understood for many years, detailed feasibility studies and openly available data remain limited. Hadi Bordbar et al. [8] utilized advancements in experimental and numerical methods for characterizing infrared spectra. By combining high-resolution spectral data from flames and blackbody emitters with virtual low-pass filters, they simulated the response of a hypothetical sensor. To enhance the differentiation between flame and blackbody responses, a pattern search algorithm was applied to determine the optimal filtering wavelengths for two detection strategies involving three or four optical low-pass filters. The optimal wavelengths were presented alongside an analysis of the detection signal’s sensitivity to non-ideal filter performance.

Table 5. Overview of thermal-based fire detection techniques, their mechanisms, applications, advantages, and challenges.

Technique	Mechanism	Applications	Advantages	Challenges
Fixed-Temperature Detectors	Activates at ≥58 °C; fusible elements melt to trigger.	Fire sprinklers, industrial use	Reliable; simple; cost-effective	Non-restorable; thermal lag
Distributed Fiber Sensing	Measures temperature via Raman/Brillouin backscatter.	Conveyor belts, groundwater flow	High sensitivity; wide range	Complex setup; environmental noise
Thermal Radiation Detection	Infrared spectra analyzed via low-pass filters.	Hangars, petroleum facilities	Rapid; highly selective	False positives from hot surfaces or sunlight

shows the overview of thermal-based fire detection techniques, their mechanisms, applications, advantages, and challenges. These findings provided guidelines for designing efficient and highly selective flame-radiation-based fire detectors.

5.4. Flame Detectors

Flame detectors are essential devices used to identify flames in various industries, including chemical plants, hydrogen stations, drilling and construction sites, industrial gas turbines, heating and drying systems, and the printing and paper manufacturing sectors. Flames, the visible and gaseous component of a fire, arise from an exothermic reaction between fuel and an oxidant, such as oxygen. Fire serves as a source of radiation, detectable by analyzing the emitted radiation within the combustion zone. The characteristics of a flame, including its chromatic properties and radiation, vary depending on the material being burned and the resulting flame temperature.

5.4.1. Deep Learning-Based Flame Detection

Flame detection has gained increasing significance in intelligent surveillance systems. Dongqing Shen et al. [106] extracted visual features from video frames for both training and testing purposes in the context of fire detection. Several shallow learning models, such as color-based, fuzzy-based, and motion- and shape-based models, were developed to detect flames. However, deep learning emerged as a more efficient and accurate approach for this task. In their study, the YOLO model was employed to perform flame detection, and its performance was compared with shallow learning methods to identify the most effective technique. The primary contribution of their research was the application of an optimized YOLO model for detecting flames in video frames. The dataset was collected and trained using Google TensorFlow, achieving an accuracy of up to 76% for the proposed flame detection approach. The YOLO model utilized a distinctive method to extract visual features for flame detection. The detection results demonstrated high accuracy in most frames. However, in scenes with numerous bright objects, the bounding box occasionally shifted slightly from the precise flame location.

The flame’s emitted radiation is influenced by its temperature and the type of fuel being combusted. Flame sensing utilizes ultraviolet, visible, and infrared sensors, which are classified based on their spectral range.

5.4.2. Multicell Flame Monitor for Enhanced Stability and Self-Checking in Combustion Systems

Lijun Xu et al. [107] introduced an innovative flame monitor utilizing three photovoltaic cells, each targeting a specific spectral band: ultraviolet (UV), visible, and infrared (IR). A gain-adjustable amplifier was integrated into the design, enabling its application for flames fueled by coal, oil, or gas. The self-checking functionality was achieved through the cross-correlation of signals from the three cells, eliminating the need for additional self-checking hardware. The monitor assessed flame stability and presence, as well as sighting-tube blockages, by analyzing both the oscillation frequency and brightness of the flame. Unlike conventional single-cell detectors, this multicell device continued functioning even after a cell failure, triggering an alarm but remaining operational until repairs were made. Experiments conducted at an industrial-scale combustion test facility demonstrated the practicality and effectiveness of the proposed flame monitor.

The combustion flames emitted energy across a continuous spectrum, with varying intensities ranging from the ultraviolet (UV) region to the far-infrared (IR) region, as outlined in [108]. Oil and pulverized fuel flames appeared more reddish in comparison to the bluish hue of gas flames. Most industrial boiler burners were equipped with pre-installed flame-sighting tubes aimed at the zone of initial combustion. Figure 12 depicts a typical installation arrangement for these sighting tubes.

A newly designed flame monitor, adhering to identical installation specifications, was developed. Its schematic diagram is presented in Figure 13. This monitor employed three distinct photovoltaic cells, collectively covering the UV, visible, and IR spectral regions, to detect flame radiation signals. While the spectral responses of the visible and IR cells were confined to their respective bands, the UV-enhanced cell spanned all three regions—UV, visible, and IR [107].

5.4.3. Ultraviolet Flame Detectors

Studies on chemical reactions in gaseous or solid reagent mixtures initiated by pulsed electric discharges [109,110] have shown that flame optical emission dynamics vary significantly across spectral intervals. During the initial inflammation phase, ultraviolet (UV) emission dominates due to nonequilibrium processes. However, the UV emissions associated with combustion and explosions, particularly in the vacuum ultraviolet range, remain poorly studied.

The stationary flame spectrum also contains nonequilibrium UV radiation, where thermal (Planck) radiation is negligible [111]. This is attributed to excited-state chemical reaction products, such as NO, CO, OH, C₂, and CN, whose electrons’ vibrational–rotational bands appear in the UV range [112,113]. This property enables flame detection, especially for large fires, using solar-blind UV photodetectors [114,115], as solar UV radiation is absorbed by atmospheric oxygen (λ < 185 nm) and ozone (185–300 nm).

E.V. Gorokhov et al. developed a natural-diamond-based UV flame detector using the photo resistance of a metal–diamond–metal structure. This photodetector, sensitive to wavelengths below 300 nm, was tested for detecting gas burner flames at 1900 K. With its peak sensitivity at λ = 210 nm and a current responsivity of 2 A/W, its effectiveness lies in the nonequilibrium flame radiation intensity being six orders of magnitude higher than the thermal emission in the detector’s sensitivity range. A natural-diamond-based photo detector has been utilized to measure the total ultraviolet (UV) emission from a natural gas flame at wavelengths near 200 nm, even in the presence of background illumination from gas-discharge lamps. The detected photo response is attributed to the nonequilibrium UV emission produced by the flame [110].

Nitrogen-free natural diamond is considered a highly promising material for developing solar-blind photodetectors capable of operating within the UV range. A critical objective in this area is to extend the range of high sensitivity toward shorter wavelengths, specifically into the vacuum UV region.

5.4.4. Infrared (IR) Sensors for Fire Detection

Infrared (IR) flame detectors are developed to identify the occurrence of open flames by analyzing the infrared spectral characteristics of fire. Due to their exceptional selectivity, greater measurement precision, and quick response time, infrared gas sensors have been extensively adopted in various applications [116,117,118].

Yafei Li et al. [119] created a mid-infrared carbon monoxide (CO) and carbon dioxide (CO₂) dual-gas sensor system. As illustrated in Figure 14, the system was composed of a gas pretreatment unit, CO₂ and CO sensor sections, and a laptop-based monitoring platform. In the gas pretreatment unit, a vacuum pump directed the gas via a ball valve, then through coarse and fine filters, followed by a drying tube, before entering the multipass gas cell (MPGC). The filters effectively removed impurities, while the drying tube, loaded with anhydrous calcium chloride, eliminated moisture. To keep contaminants out of the MPGC, more filters were added after the vacuum pump. The CO₂ and CO sensors used LMM–242–TH and LMM–242–IH detectors (InfraTec., Dresden, Germany), respectively, with the MPGC outlets connected for simultaneous detection. A digital signal processor (DSP) controlled the broadband infrared light source and modulated the signal, which was then split and reflected through gold-plated spherical reflectors, forming an effective 1.8 m optical path. The light output was detected by a dual-channel pyroelectric detector. The signals were amplified and converted into voltage signals, which were processed by the DSP to obtain concentration data. A wireless network transmitted the data to a LabVIEW-based platform, which monitored the CO₂ and CO levels. If a fire was detected, the system triggered an alarm and displayed the fire’s location on a map.

A mid-infrared sensor system for early fire detection and location was developed. The absorption lines at central wavelengths of 4.26 µm and 4.66 µm were selected as the target absorption lines for CO₂ and CO detection, respectively. The CO₂ sensor module had a concentration range of 200–1600 ppmv, while the CO sensor covered a range of 0–200 ppmv. The dynamic response times for both sensors were measured to be 30 s at a flow rate of 1 L/min. The limits of detection (LoDs) were determined to be 5.66 ppmv for CO₂ and 0.94 ppmv for CO, with an average time of 0.25 s. By examining the smoldering behavior of various materials, an early approach for fire detection was explored, and a threshold value of 0.1 for C(CO)/C(CO₂) was established as the fire alarm indicator. Furthermore, a mobile early-fire location method was proposed, integrating the GTD model with the PSO algorithm. Experimental findings confirmed the reliability, as well as the effectiveness, of the proposed location method.

5.5. Gas Sensors

Gas sensors are critical components in fire detection systems, enabling the identification of the hazardous gases produced during combustion or gas leaks. During fire incidents, gases like carbon monoxide (CO), carbon dioxide (CO₂), and ethane (C₂H₆) are typically of primary concern. CO is released due to incomplete combustion, CO₂ is generated from complete combustion, and C₂H₆ is associated with fires originating from natural gas or liquefied petroleum gas (LPG) leaks. Additionally, nitrogen dioxide (NO₂) concentrations can rise due to crowd breathing, making NO₂ an important indicator in confined space disasters, such as crush injuries. Therefore, effective gas detection is vital for early disaster response and prevention. Conventional gas sensing technologies have been extensively used in this domain, leveraging various mechanisms to detect specific target gases. The most widely employed sensor types include electrochemical sensors, metal-oxide semiconductor (MOS) sensors, and nondispersive infrared (NDIR) sensors. However, conventional gas sensing technologies, including NDIR, electrochemical, and MOS sensors, have distinct limitations that hinder their effectiveness in such complex environments.

Liu et al. [47] comprehensively evaluated gas sensing technologies, focusing on key performance metrics, such as selectivity, sensitivity, energy usage, stability, adsorption capacity, reversibility, response time, and manufacturing expenses. Gas sensors operate by detecting changes in output signals, which indicate the presence of specific gases at a given location. In fire-related scenarios, the release of significant quantities of carbon dioxide (CO₂) is a common characteristic, along with the presence of toxic gases, such as carbon monoxide (CO) and hydrogen cyanide (HCN) [77,120]. The depletion of oxygen is also observed as CO concentrations rise. Low oxygen levels are indicative of smoldering fires, while rapid changes in oxygen concentrations often signal the combustion of liquid fuels. Among the various technologies, catalytic bead sensors, also referred to as pellistors, are widely utilized for detecting combustible gases in air. These sensors are integrated into both portable and stationary devices in industries such as mining to detect explosive conditions (Figure 15). However, catalytic bead sensors are significantly influenced by fluctuations in ambient temperature, leading to variability in their output signals. These fluctuations can result in false alarms or failures to respond during explosive conditions, posing a challenge to their reliability [121,122].

While numerous gas sensing technologies are available, semiconductor metal oxide gas sensors have garnered considerable interest due to their cost-effectiveness, operational simplicity, great stability, and broad chemical detection range. Their adaptability and reliability have positioned them as a preferred choice for various applications, particularly in environments requiring cost-effective and stable sensing solutions.

5.5.1. Metal Oxide Semiconductor Gas Sensors

Metal oxide semiconductor (MOS) gas sensors are widely recognized for their high sensitivity and cost-effectiveness. The core principle behind these sensors’ mechanism relies on changes in electrical resistance, which occur because of chemical reactions between the molecules of the target gas and oxygen ions that have been absorbed on the MOS material’s surface. Upon exposure to the target gases, these interactions alter the resistance of the material, enabling gas detection. Despite their advantages, MOS sensors face challenges related to stability, which can result in false alarms. Enhancing the discrimination capability of these sensors has been achieved by depositing additional layers, like zeolites, onto the metal oxides, improving their performance and selectivity [123]. Materials like nickel oxide, tin oxide, iron oxide, and zinc oxide are commonly employed in the fabrication of gas sensors, particularly those designed for detecting oxygen-containing gases [124,125,126,127]. These materials offer several advantages, including enhanced sensitivity and cost-effectiveness. However, they often face challenges related to reliability. In contrast, carbon-nanomaterial-based gas sensors, such as graphene and carbon nanotubes, demonstrate superior sensitivity and detection capabilities compared to traditional metal oxide semiconductor (MOS) sensors. Among these, graphene derivatives, including graphene oxide (GO) and reduced graphene oxide (rGO), stand out for their exceptional sensitivity to a wide range of gases [128,129,130,131]. The advancements in graphene-based materials have significantly expanded the potential of these sensors, making them highly effective for various gas detection applications. Composite materials have also been employed to enhance the performance of such optical gas sensors, further increasing their utility [132]. To address the challenge of detecting multiple gases simultaneously, sensor arrays have been employed in multi-parameter gas sensing applications [133]. Polymers have proven to be effective in these applications due to their enhanced sensitivity, which arises from their ability to undergo doping-level modifications through chemical interactions with various analytes. This property allows for straightforward gas detection at ambient temperatures [134]. For electronic fire detection, Riches et al. [135] investigated the sensitivity of surface acoustic wave (SAW) and MOS sensors. The SAW (surface acoustic wave) sensor detects fire by monitoring shifts in frequency resulting from the adsorption of gases or vapors onto the surface of a piezoelectric crystal, while the MOS sensor measures conductivity changes in a metal oxide sheet when exposed to organic vapors. Several innovations have been reported in the advancements of MOS-based sensors for specific gas detection. Mandayo et al. [136] developed a micro-machined tin oxide system for detecting CO. Juang et al. [137] designed a Au/MO/n low-temperature polysilicon MOS Schottky diode on a glass substrate, which demonstrated superior performance due to the large band gap and high surface-to-volume ratio of SnO₂. Among various metal oxides, such as SnO₂, TiO₂, and ZnO, SnO₂ exhibits the greatest sensitivity ratio. Abid et al. [138] fabricated a gas sensor using SnO₂ nanowires to identify gases responsible for fire, such as isopropanol, CO, and benzene. This sensor operates by measuring resistance variations caused by these gases and is particularly effective when heated to 200 °C, enabling it to detect combustion odors from materials such as cotton, beech wood, and printed circuit boards.

Recent advancements in metal oxide sensors have introduced metallic heaters, including platinum (Pt), palladium (Pd), silver (Ag), nickel (Ni), and chromium (Cr), to elevate the sensor temperature to optimal levels [139]. These heaters improve sensor performance by enabling precise operation at desired temperatures, facilitating the detection of fire-related gases. Consequently, a new class of simple, cost-effective, and exceptionally sensitive gas sensors has been developed, utilizing resistance measurements as a core detection principle (Figure 15b). These innovations represent a significant step forward in the creation of reliable and affordable gas sensing solutions for fire detection.

5.5.2. Electrochemical Sensors

Electrochemical sensors play a crucial role in fire detection systems, owing to their excellent sensitivity and selectivity toward specific gases emitted during combustion. These sensors operate by measuring the electrical current produced through the electrochemical oxidation or reduction of target gas molecules at the sensor’s electrodes. This electrical current varies proportionally with the gas concentration, enabling accurate detection. For instance, electrochemical gas sensors on General Monitors are extremely sensitive to toxic gases, including carbon monoxide, hydrogen sulfide, hydrogen, ammonia, and oxygen deficiency. One of the primary advantages of electrochemical sensors is their ability to detect gases at low concentrations, which is crucial for early fire detection. They are particularly effective in identifying toxic gases such as carbon monoxide (CO), a common byproduct of fires. The ECO–5011 sensor (Cubic Sensor and Instrument Co., Ltd., Wuhan, China), for example, is widely used in residential and commercial CO sensors, fire alarms, industrial CO monitoring, home gas alarms, parking garages, and automotive safety alarms. Moreover, electrochemical sensors are known for their low power usage, making them ideal for battery-operated devices and remote monitoring systems. Their compact size allows for integration into various fire detection equipment without significant design alterations. Additionally, these sensors exhibit minimal interference from other gases, enhancing their reliability in diverse environments. The leading sensing technology company SemeaTech provides a comprehensive range of electrochemical sensors for identifying dangerous substances and gases for environmental and occupational safety. However, certain limitations exist. The lifespan of electrochemical sensors can be affected by environmental factors, such as temperature and humidity fluctuations. Regular calibration is necessary to maintain their accuracy over time. Regardless of these challenges, advancements in sensor technology continue to address these issues, improving durability and performance. For example, chemical sensor systems and their associated algorithms have been developed to enhance fire detection capabilities, providing faster alarm signals when gases are released before smoke particles [3]. Figure 16 illustrates the working principle of an electrochemical gas sensor used for detecting gases, like hydrogen (H₂), or oxygen (O₂) [140]. The sensor consists of a porous measuring electrode, a diffusion membrane, an electrolyte, and a reference electrode. At the measuring electrode, the target gas, H₂ undergoes oxidation, producing hydrogen ions (H⁺), and electrons (e⁻). The H⁺ ions migrate through the electrolyte to the reference electrode, where oxygen, O_2, is reduced, forming water (H₂O). The flow of electrons generates a current proportional to the gas concentration, which is measured by an amperimeter. This mechanism ensures high sensitivity and is widely applied in fire detection systems. In summary, electrochemical sensors play a crucial role in modern fire detection systems, offering the precise and early identification of hazardous gases. Their integration into safety protocols significantly enhances the ability to respond promptly to fire incidents, thereby protecting lives and property.

5.5.3. Nondispersive Infrared Sensors

Carbon dioxide (CO₂) is a significant component of fire emissions. Detecting CO₂ at concentrations relevant to fire scenarios is effectively achieved using miniature nondispersive infrared (NDIR) sensors. These sensors operate on the principle that CO₂ molecules absorb specific wavelengths of infrared (IR) radiation, particularly at 2.7, 4.3, and 15 μm. An NDIR sensor typically comprises an IR light source, an absorption chamber, wavelength filters, and IR detectors. Modern advancements have led to the integration of these components into compact systems, enhancing their applicability in various settings. For instance, a fully integrated, on-chip NDIR CO₂ sensor has been developed, featuring an integrating cylinder with access waveguides, a mid-IR LED as the optical source, and mid-IR photodiodes as detectors, as shown in Figure 17. This miniaturized sensor demonstrated a detection limit of approximately 750 ppm, indicating its potential for applications requiring compact and sensitive CO₂ monitoring [141]. Additionally, NDIR sensors are valued for their stability, accuracy, and low power consumption, making them suitable for deployment in dense sensor networks aimed at monitoring CO₂ concentrations with a high spatial resolution [142]. These attributes make NDIR sensors a reliable choice for CO₂ detection in fire safety applications.

In evaluating different sensor types, it becomes clear that their performance must be assessed not only by sensitivity or fabrication techniques but also by their real-world applicability. MOS sensors, for example, provide high sensitivity and fast responses but typically require elevated operating temperatures, increasing their power consumption and limiting their portability [143]. Carbon-based sensors, including CNTs and graphene derivatives, like GO, exhibit excellent sensitivity at room temperature, yet they often face challenges in stability and reproducibility under varying humidity [144]. Polymer-based sensors are flexible and cost-effective but generally show longer response/recovery times and lower stability in harsh environments [145].

Table 6. Comparison of response times for commercially available, fire-related gas sensors, including MOS, electrochemical, optical, and NDIR sensors.

Sensor Type	Target Gas	Response Time	Operating Temp	Model/Source
Electrochemical Sensor	CO	~15–60 s	Ambient	Figaro TGS5042
MOS Sensor	CO, VOCs	~10–30 s	150–300 °C	Figaro TGS2611
NDIR Sensor	CO2	~20–30 s	Ambient	Senseair S8
IR Flame detector	Open flame	<5 s	−40–75 °C	Honeywell FS24X
Electrochemical Sensor	CO	~30–60 s	0–50 °C	FECS40-1000
NDIR Sensor	CO2	<120 s	0–50 °C	CDM7162-C00
Optical Sensor	Flame	<0.01 s	20–50 °C	UVTRON R2868
Optical Sensor	Flame	<0.039 s	−40–85 °C	Honeywell SS2
Optical Sensor	Flame	<120 s	0–40 °C	Honeywell i3 Series 4WITAR-B

summarizes the response times of commercially available, fire-related gas sensors, including metal-oxide-, electrochemical-, optical-, and NDIR-based sensors. This comparison provides useful benchmarks for evaluating the performance of advanced sensor technologies.

6. Advances in Gas Sensing for Fire Detection

6.1. Gasistors

In complex disaster situations, the rapid and accurate detection of various gases is crucial for ensuring timely responses and preventing accidents. During fire incidents, gases, such as carbon monoxide (CO), carbon dioxide (CO₂), and ethane (C₂H₆), are typically of primary concern. CO is released due to incomplete combustion, CO₂ is generated from complete combustion, and C₂H₆ is associated with fires originating from natural gas or liquefied petroleum gas (LPG) leaks. Additionally, nitrogen dioxide (NO₂) concentrations can rise due to crowd breathing, making NO₂ an important indicator in confined space disasters, such as crush injuries. Therefore, effective gas detection is vital for early disaster responses and prevention. However, conventional gas sensing technologies, including nondispersive infrared (NDIR), electrochemical, and metal oxide semiconductor (MOS) sensors, have distinct limitations that hinder their effectiveness in such complex environments.

NDIR gas sensors rely on the absorption of infrared light by specific gases, which necessitates the inclusion of an infrared light source, gas sample chamber, filter, and infrared detector. While effective in certain applications, the need for these components results in limitations in size and power consumption, making it difficult to monitor atmospheric gases continuously. Electrochemical sensors, which detect the change in current caused by a gas reaction with an electrode, also face challenges. Their reliance on an electrolyte in the reaction mechanism restricts their miniaturization and power efficiency. MOS sensors, which detect gas concentrations by observing changes in resistance as a result of gas adsorption on the metal oxide surface, are the smallest of these sensors. However, they are susceptible to external environmental factors, such as humidity and temperature, requiring external heaters to maintain their reliable operation.

To overcome these limitations, there is a need for the development of advanced gas sensing technologies that are miniaturized, consume less power, and offer high reliability in extreme disaster conditions, such as varying temperature and humidity. The gasistor, a novel type of semiconductor gas sensor utilizing a memristor, presents a promising solution. A gasistor is a type of memristive device specifically engineered to detect and respond to gas molecules. It combines the functionalities of a gas sensor and a memristor, meaning it not only senses gases through changes in electrical properties but also retains memory of these changes, enabling enhanced signal processing and potentially neuromorphic applications. The device’s resistance state is influenced by gas interactions at the sensing material’s surface, making it sensitive to the gas concentration and type. This dual capability allows gasistors to operate as intelligent sensing units with memory features, offering advantages like low power consumption, miniaturization, and real-time monitoring. Figure 18 illustrates a gasistor-based sensing system integrated with an ESP32 module for real-time monitoring via a smartphone. The right panel shows the working mechanism, where gas adsorption modulates the CF in the active layer, switching the device between a low-resistance state (LRS) and a high-resistance state (HRS) depending on the interactions between gas molecules and oxygen vacancies [146]. Unlike traditional gas sensors that operate based on surface reactions, the gasistor detects gases through a single filament local reaction, as shown in Figure 19. This mechanism is based on the principle that the resistance of the filament increases when a reducing gas (negative charge) reacts with an oxygen–nitrogen vacancy-chain-structured conductive filament (positive charge) and decreases when the filament reacts with an oxidizing gas (positive charge) [7]. The gasistor has garnered significant attention due to its superior gas sensing characteristics, such as miniaturization, low power consumption, and high recycling rate, especially when compared to MOS-type sensors. This new approach to gas sensing holds the potential to significantly improve detection capabilities in complex disaster situations.

6.2. Experimental Setup

In the field of gas sensing, experimental setups are carefully designed to simulate real-world conditions in which gas sensors (or gasistors) can detect and respond to specific gases. These setups typically consist of several key components that ensure the accurate measurement of the sensor’s performance. A typical gas sensor testing setup includes a sealed gas chamber, where the sensor is exposed to various target gases. This chamber is often equipped with gas flow controllers to regulate the introduction of specific gas mixtures into the system. Additionally, temperature and humidity control systems are integrated into the chamber to maintain consistent environmental conditions, as these factors can significantly affect the sensor’s response. The sensors used in these setups can be-based on a variety of materials, such as metal oxide semiconductors, graphene, or carbon nanotubes (CNTs), each of which interact differently with the gas molecules, resulting in measurable changes in electrical properties, such as resistance or capacitance [9].

To monitor the sensor’s response, data acquisition systems are employed to record changes in electrical signals over time. These setups allow researchers to analyze how the sensor performs under different conditions, such as varying gas concentrations, temperature, and humidity levels. Multisensor arrays are sometimes used to detect multiple gases simultaneously or to enhance the selectivity and sensitivity of the measurements. The experimental design is crucial for ensuring reliable and reproducible results, and it often includes sensitive measurement instruments like digital multimeters, oscilloscopes, or even optical detectors, depending on the sensing mechanism.

As shown in Figure 20a, the gas sensing system is designed to create a controlled environment where the sensor is exposed to precise concentrations of gases. This system features a sealed gas chamber where the gases are introduced using a gas flow controller. This setup allows for the regulation of the gas concentrations with high accuracy, as well as the control of both temperature and humidity to ensure stable testing conditions. The sensor, embedded with MH-embedded CNTs, is placed in the chamber, where it can interact with the target gases. The change in the electrical resistance of the sensor, caused by the interaction of gas molecules with the sensor material, is measured and recorded. The setup also includes valves that can control the gas flow, allowing for dynamic adjustments during the experiment [147].

In Figure 20b, the schematic of the enlarged gas chamber with the MH-embedded CNTs-based gas sensor is shown. This figure provides a closer look at the sensor’s placement inside the chamber, which ensures that it is fully exposed to the gas flow. The chamber is designed to allow for efficient interaction between the gas molecules and the sensor material. As the gas flows through the chamber, the sensor detects the target gas, and the electrical response is measured. The MH-embedded CNTs play a critical role in this process, as their unique structure allows for high sensitivity to gases, with the resistance changing in response to gas adsorption on the CNT surface. This schematic shows the precise arrangement of the components in the chamber, ensuring that the sensor’s response is both accurate and reproducible [148].

Other research studies in the field of gas sensing also employ similar experimental setups. These setups typically involve gas flow controllers, mass flow meters, and precise gas mixtures to expose the sensor to control concentrations of gases. In some studies, multisensor arrays are used to detect a range of gases simultaneously, which improves the overall sensitivity and selectivity of the system. Additionally, temperature and humidity control are a crucial aspect of these setups, as both factors can influence the sensor’s response. Researchers often use optical or electrical measurements to record the sensor’s behavior under varying gas concentrations, and data acquisition systems are employed to monitor real-time changes in the sensor’s electrical properties.

Overall, the experimental setup for gas sensors is designed to be flexible, allowing for precise control over environmental conditions and gas concentrations. These setups are fundamental for studying the response of different sensor materials, such as MH-embedded CNTs, semiconducting metal oxides, or graphene-based sensors, under well-defined conditions, ensuring that the resulting data is both reliable and reproducible.

Gasistors typically adopt a metal–oxide–metal architecture, where the active sensing occurs via the formation and rupture of conductive filaments within the oxide layer. These filaments consist of oxygen or nitrogen vacancies and serve as nanoscale, electrically re-configurable pathways whose resistance can be modulated by gas interaction. For in-stance, in IGZO-based gasistors, exposure to reducing gases, such as isopropanol, leads to the adsorption of negatively charged molecules that disrupt CFs, increasing the device’s resistance [146]. This process contrasts starkly with MOS sensors, which rely on surface charge modulation via gas adsorption at elevated temperatures. Notably, gasistors exhibit room-temperature operation, low power consumption, and rapid recovery times (as low as 50 µs via pulse biasing), while offering high sensitivity and excellent integration potential with IoT systems. Moreover, Vidiš et al. [149] demonstrated a gas-triggered switching behavior, enabling thyristor-like operation and long-term memory retention, which allows environmental history tracking—an attribute absent from MOS devices. This novel sensing mechanism enables dynamic threshold-based switching and memory storage, making gasistors highly attractive for low power, real-time environmental monitoring applications.

Memristor-based gas sensors, known as gasistors, represent a significant advancement in gas sensing technology by integrating the functionalities of gas sensors and memristors to address the limitations of conventional methods. These devices are particularly well-suited for fire detection due to their high sensitivity, selectivity, and efficient operation in challenging environmental conditions, such as varying temperatures and humidity. A variety of materials have been employed to enhance their performance in fire-related scenarios. Marek Vidis et al. introduced the term gasistor for memristor-based gas sensors, highlighting their ability to detect gases while retaining memory. This study demonstrated an ultrasensitive gasistor for H₂ detection, featuring a gas-triggered switching mechanism. The proposed structure (Figure 21a) includes a 100 nm activation area, where resistive switching was analyzed. The I–V characteristics in synthetic air (Figure 21b) and the response to 10,000–ppm H₂ at room temperature (Figure 21c) showed a significant resistance drop upon gas exposure. Notably, the device exhibited a self-resetting behavior, where exposure to H₂ naturally transitioned it to a low-resistance state, resembling conventional semiconductor gas sensor mechanisms (Figure 21d). Such gasistors hold promise for fire detection by sensing flammable gases, like H₂ [149].

Peilun Qiu et al. developed a memristor-based gas sensor utilizing a TiO₂ nanostructure to detect low concentrations of NH₃ at room temperature (RT) [11]. The device structure is depicted in Figure 22a, while Figure 22b presents its typical bipolar I–V characteristics, with set and reset voltages of 1.25 V and –1.04 V, respectively. The gas sensing performance was analyzed in different resistive states, with the HRS demonstrating a significantly higher response (164.2) compared to the LRS (10.1) for 1 ppm NH₃, as shown in Figure 22e,f. The sensor exhibited remarkable sensitivity, even at 250 ppb NH₃, with a response of 35.7 under HRS, and fast response and recovery times within 1 s. Selectivity tests against other reducing gases, including methanol, ethanol, acetone, isopropanol, toluene, and triethylamine, confirmed a strong preference for NH₃ detection. Although increasing humidity (20–85% RH) reduced the response, it remained high at 76.6 for 1 ppm NH₃, with slightly prolonged response and recovery times (6 s and 10 s). Similar to Marek Vidis et al., they applied a gas-triggered set process and found that the triggering voltage increased with NH₃ concentration, surpassing the initial –1.04 V due to gas interactions. The study emphasized that oxygen adsorption on the TiO₂ surface plays a crucial role in NH₃ sensing, as NH₃ reacts with adsorbed oxygen ions, altering the surface depletion layer. These TiOs₂-based gasistors, with their high sensitivity and selectivity, have potential applications in fire detection by sensing ammonia and other combustion-related gases, contributing to early fire warning systems [11].

D. Lee et al. investigated memristor-based gas sensors using SnO₂, HfO₂, and Ta₂O₅ to compare their response characteristics for NO, C₂H₆, and O₂ detection [150]. The device structures are shown in Figure 23a–c, while their resistive switching behaviors—unipolar for SnO₂ and Ta₂O₅ and bipolar for HfO₂—are illustrated in Figure 23d–f. Gas sensing was performed in the LRS with oxygen vacancy-based CFs, where the SnO₂ memristor exhibited the highest response (118.9) for 50 ppm C₂H₆ (Figure 23g). The HfO₂-based sensor showed responses of 5.2 for NO and 4.7 for O₂ (Figure 23h), while the Ta₂O₅ sensor displayed responses of 18.7 for C₂H₆ and 11.7 for NO (Figure 23i). These sensors demonstrated rapid response times of under one second for all the tested gases, with fast recovery for low concentrations but a prolonged recovery time of over 340 s at 50 ppm. To accelerate recovery, the study introduced DC and pulse bias methods, achieving restoration in 28 s with the DC bias and within 90 ns using the pulse bias. These advancements in memristor-based gas sensing highlight the potential of gasistors for the real-time detection of hazardous gases, making them promising candidates for fire detection applications by enabling the rapid identification of combustion-related gases.

For instance, SnO₂-based gasistors demonstrate high responsiveness to reducing gases, like ethane (C₂H₆), with rapid recovery and excellent sensitivity, while TiO₂-based gasistors offer superior selectivity and strong response characteristics for detecting carbon monoxide and ammonia, key byproducts of combustion [7]. Ta₂O₅-based gasistors, on the other hand, are valued for their robust operation and stability in detecting gases within complex environmental backgrounds. These material-specific advancements make gasistors highly adaptable to diverse gas profiles encountered during fire incidents, thereby increasing their reliability in real-world applications. Furthermore, a defining feature of gasistors is their advanced recovery systems, which utilize pulse- or trigger-based mechanisms to return to baseline resistance states within milliseconds, a crucial capability in dynamic fire scenarios where gas concentrations fluctuate rapidly. In addition to their material and functional advantages, gasistors exhibit exceptional stability under various environmental conditions. Their design minimizes the impact of humidity on resistance changes and enables consistent operation across wide temperature ranges, unlike traditional sensors that rely on external heaters. With their compact form factor, low power requirements, and real-time detection capabilities, gasistors hold significant potential for integration into fire safety systems [7]. They are particularly suited for deployment in IoT-enabled fire detection networks, providing continuous monitoring and early warning capabilities for residential, industrial, and public safety applications.

Table 7. Summary of key sensing characteristics of memristor-based gas sensors, including sensing materials, target gases, response times, and recovery times.

Target Gas	Sensing Material	Driven Source	Concentration (ppm)	Response (Rair/Rgas)	Recovery Time (S)	Ref.
NO2	CNT-based gasistor	MHH MHN	10	1.46 1.77	0.001	[148]
NO2	CNT-based gasistor	MH (HRS)	50	52	0.001	[147]
CH3OH	ISnS/TiO2-based gasistor	RT	1	85.2	0.65	[151]
NO	Zr3N4-based gasistor	RT	0.05	25.8	0.0000016	[152]
NO C2H6 O2	HfO2-based gasistor	RT	50	5.2 1.6 4.7	0.00000009	[150]
O2	IGZO-based gasistor	RT	0.5	14	0.00009	[153]
H2	TiO2-based gasistor	Dry air Humid air	10000	~5 × 104 ~6 × 105	180 11	[154]
Ethanol	TiO2-based gasistor	350–750 °C	1000	~15	_	[155]
NO	Bn- based gas sensor	RT	5	18	81	[156]
IPA	a-IGZO	RT	10	2.51	0.000005	[146]

shows the response characteristics of the reported memristor-based gas sensors.

6.3. Comparative Analysis of Gasistors and Traditional MOS Sensors

The performance differences between gasistors and traditional MOS sensors stem from fundamental distinctions in their sensing mechanisms and material properties, which directly impact their sensitivity, selectivity, power consumption, and response/recovery characteristics. Traditional MOS sensors operate based on surface reactions, where target gases interact with adsorbed oxygen species on the entire metal oxide surface, leading to an electron transfer that changes the material’s conductivity [157]. In contrast, gasistors utilize a localized CF mechanism, where the sensing occurs primarily at the filament rather than across the entire material surface [157,158]. When reducing gases interact with the oxygen vacancy-based CF (positive charge), the resistance increases due to the electron donation to the filament; conversely, oxidizing gases extract electrons, decreasing resistance [159]. This fundamental mechanistic difference explains why gasistors can operate effectively at room temperature, while MOS sensors require elevated temperatures.

The material composition of traditional MOS sensors (typically SnO₂, ZnO, TiO₂, etc.) necessitates a porous structure to maximize the surface area for gas adsorption [160]. While this enhances sensitivity, it simultaneously increases susceptibility to interference from humidity and other environmental factors [161]. In contrast, gasistors employ a compact active layer with controlled defect structures (primarily oxygen vacancies) that form conductive filaments, resulting in a more stable sensing platform that is less affected by environmental variables [11]. The use of transition metal oxides with specific defect engineering in gasistors allows for greater control over sensing properties through manipulation of the oxide’s stoichiometry and electronic structure [157].

The superior sensitivity of gasistors (response values of 118.9 for SnO₂-based gasistors to 50 ppm C₂H₆ [159], compared to typical responses of 5–10 for traditional SnO₂ MOS sensors at similar concentrations [161]) can be attributed to the concentrated nature of the sensing mechanism. In gasistors, the response occurs along a nanoscale filament, where even small numbers of gas molecules can significantly alter the electrical properties [11]. By comparison, MOS sensors require gas molecules to interact with a substantial portion of their surface to generate measurable signals. Selectivity improvements in gasistors derive from their bipolar resistive switching properties, which enable different gas detection modes based on the resistive state [159]. When operated in an HRS, gasistors demonstrate enhanced selectivity for specific gases compared to their LRS, a tunability not available in traditional MOS sensors. This adaptability allows gasistors to be optimized for detecting specific fire-related gases, like NO, CO, or C₂H₆, through controlled resistance state modulation [11].

Finally, the room temperature operation of gasistors represents a substantial advantage over MOS sensors’ high temperature requirements, significantly reducing power consumption. This improvement stems from the fundamentally different sensing mechanism that does not rely on thermally activated surface chemistry, making gasistors particularly suitable for battery-powered and IoT-integrated fire detection systems [159].

Gasistor-based sensors offer a compelling combination of high sensitivity, room-temperature operation, rapid response and recovery, and compact architecture. Their operation is governed by a filament-based resistive switching mechanism, which enables the selective detection of specific gas molecules while retaining non-volatile, memory-like behavior. These attributes make gasistors particularly advantageous in scenarios requiring fast and reliable gas identification, such as in early fire detection or environmental monitoring. This performance distinction highlights the importance of evaluating sensor technologies not only by their material properties but also by their potential for integration into real-world applications, including wearable electronics, embedded systems, and IoT-based platforms.

7. Deep Learning and Artificial Intelligence in Fire Detection

Despite continuous advances in sensor technology and significant research investment, fire incidents still occur frequently. This is mainly due to the limitations of traditional sensors, including their sensitivity to environmental changes (temperature extremes, humidity, smoke density, sensor drift, and mechanical degradation over time). In addition, many existing systems rely on human monitoring, which can lead to delays in detection and response.

To address these problems, researchers are exploring the integration of artificial intelligence (AI) with sensor-based fire detection systems. AI technologies provide advanced abilities, such as pattern recognition, adaptive learning, and predictive analytics, to enable the detection of earlier fires in a variety of environments. Therefore, this section analyzes AI-based fire detection strategies from three perspectives: data collection and preprocessing, model development, and practical application. In this section, we present some representative approaches and studies that focus on collecting training data for early fire detection.

7.1. Data Collection and Preprocessing Methods for Early Fire Detection

One of the most critical issues in applying AI to fire detection is the quality and structure of the data. Unlike traditional fire sensors that require a single point signal, such as a temperature threshold or gas concentration limit, AI models require various datasets to learn patterns that distinguish early fire events from non-hazardous conditions. So, data collection and preprocessing are critical to the effectiveness of the final AI model.

Wang et al. built a large public dataset called the Flame and Smoke Detection Dataset (FASDD) to improve deep learning detection performance in various fire scenes [162]. The dataset consists of more than 120,000 images and 180,000 bounding box annotations, including three subsets from different platforms (ground cameras, drones, and remote sensing satellites). The dataset contains a mixture of smoke and flames, varying illumination and resolution, and complex weather conditions. The samples have been classified into four categories: “fire”, “smoke”, “fire and smoke”, and “neither fire nor smoke”. The annotation format supports a variety of formats, including YOLO, VOC, and COCO, for greater flexibility in model training, and can be used as a key resource in a variety of deep learning-based fire detection scenarios, including urban fires, wildfires, and remote surveillance.

To overcome the lack of real-world fire data and improve the performance of deep learning wildfire detection models, Fernando et al. propose SWIFT (Simulated Wildfire Images for Fast Training), a synthetic fire dataset generated through 3D simulations based on Unreal Engine [163]. SWIFT consists of a total of about 70,000 high-resolution images (1920 × 1080), as shown in Figure 24, and 15 simulated images, including RGB images and corresponding mask images annotated into four classes: smoke–fire–both–no fire. The data was collected from a variety of viewpoints (ground, aerial, and interior views) and conditions (temperature, humidity, wind speed, wind direction, etc.) and was categorized into fire-only, smoke-only, and fire + smoke scenarios for different detection purposes. In particular, the Niagara particle system and UE graphics technology were used to increase the visual precision, and automated mask generation techniques were used to reduce the manual labeling effort. The dataset has visual characteristics similar to real-world data, enabling sim-to-real learning, and has been used to train models, such as BoucaNet, CT-Fire, and DC-Fire, which have shown high accuracy in real-world fire images.

El-Madafri et al. proposed a hierarchical domain adaptive learning framework using a dual-dataset-learning approach to improve the accuracy of wildfire detection in different environments [164]. The approach consists of a separate auxiliary dataset consisting of non-wildfire images and a primary dataset centered on wildfire images collected in different terrains and climatic conditions. The secondary dataset contains open-source building fire and smoke images from Roboflow and GitHub, while the primary dataset consists of 2700 high-resolution wildfire images collected from government agencies, Flickr, and Unsplash. In particular, different resolutions, viewing angles, weather conditions, and lighting variations were considered to increase realism, and the initial class balance was corrected using data augmentation techniques. The two datasets were not mixed during training, but were processed on separate paths, with common features being learnt through a common “shared layer” and individual features being considered through a “task-specific layer” specific to each environment.

Altaf et al. utilize the Laplacian Pyramid Super-Resolution Network (LapSRN) in the preprocessing stage to address the quality of low-resolution image data [165]. This method gradually restores the resolution of the input low-resolution images, allowing for the more-accurate recognition of visual elements, such as the fine boundaries of fire features, smoke textures, and flame patterns. The preprocessed high-resolution images are then fed into an Xception-based attention deep learning network, which is optimized to focus on visually significant areas through its Adaptive Spatial Attention (ASA) module. In this study, a custom dataset was built to reflect realistic complexity, including fire and non-fire scenes captured indoors and outdoors, day and night, and under different brightness conditions. The experimental results show that the proposed network outperforms several existing SOTA models on key evaluation metrics, such as precision, recall, F1-score, and accuracy, and demonstrates its applicability to edge devices as a real-time, lightweight model.

Meanwhile, Diaconu et al. published a review paper on the impact of the structural quality factors of datasets on the model performance in AI-based fire detection systems, in which he conducted a comparative analysis of 14 public datasets [166]. The study found that class imbalance, a low sample size, and a lack of scene diversity in the dataset are important limiting factors in high-performance learning models. He notes that while many studies focus on improving quantitative metrics, he suggests that in real-world fire alarm systems, the socioeconomic costs of false positives and false negatives may be more important. Therefore, he proposes improving dataset quality as a key strategy for improving performance and presents specific approaches, including applying contextual loss functions and class redistribution strategies.

7.2. Comparative Approaches in Fire Detection Model Design

A major decision in developing models based on these datasets is whether to choose a lightweight model for low-power, real-time environments or a high-performance model for precision detection over large areas. In this section, we compare these two model families, focusing on performance, application, data requirements, and real-world use cases. We first introduce several representative lightweight models, highlighting their suitability for embedded environments. Afterwards, we explore high-performance architectures that focus on detection accuracy and scalability in more complex or expansive settings.

Muhammad et al. present a lightweight network for devices in resource-constrained environments within smart cities [167]. The Flame Net detects fires, triggers alarms, and communicates to fire, medical, and rescue departments. It also integrates MSA modules to efficiently prioritize and enrich key fire-related functions for effective fire detection. Compared to the traditional SOTA method, the model achieved a high test accuracy of 99.40% with fewer parameters, with a precision of 0.99 for the fire class and 1.00 for the non-fire class, recalls of 1.00 and 0.98, and an F1-score of 0.99. A new dataset, new Ignited-Flames dataset, was created and applied to this model to obtain accuracy on two local machines (RPi, CPU).

On the other hand, an attempt to integrate various systems to detect small-scale, real-time fires was made by Titu et al. [168]. They proposed a system that integrates drones, edge devices, and lightweight deep learning models for real-time fire detection. The study collected and augmented 7187 images, including various fire situations (forest, vehicle, industrial, residential, drone view, etc.) to form a dataset for training and created a lightweight model using YOLOv8m as a teacher model and YOLOv8n and DETR as student models. They automated labeling using the Autodistill-based knowledge distillation technique and real-time deployed YOLOv8n on a DJI F450 drone equipped with a Raspberry Pi 5 and Pi Camera Module 3 to evaluate its fire detection performance in real-world environments. The distilled YOLOv8n model achieved a 95.21% accuracy, 98.9% precision, 98% recall, an F1-score of 0.985, and a mAP of 93.31%. In the drone experiment, as shown in Figure 25, we achieved 89.23% real-time detection accuracy while operating at an average of 8 FPS. This study demonstrates the feasibility of high-performance, real-time fire detection in real-world, low-power edge environments.

Specifically, Safak et al. sought to develop a model that could perform real-time fire and smoke detection on mobile devices [169]. The proposed model was able to work effectively, even in low-power environments. To achieve this, training was performed on a large dataset consisting of a total of 43,355 images, and the model was lightweighted to allow image analysis directly on the mobile device, instead of the traditional complex server-based processing. By comparing the performance of various lightweight CNN models, the proposed model showed excellent results, with a high accuracy of 98.37%. In particular, this study experimentally confirmed that it maintained a high generalization performance even under various weather, illumination, and image conditions. Furthermore, it showed the potential to contribute to the development of next-generation, IoT-based fire detection systems based on its fast response and low computational cost compared to traditional fire detection methods.

Kwon et al. compared the performance of segmentation-based deep learning models for the real-time detection of electric vehicle (EV) fires [170]. For this purpose, they constructed 3000 frames of data based on 60 real EV fire videos and conducted training, validation, and testing in various environments and vehicle types. YOLOv5/8/11-Seg, Mask R-CNN, and Cascade Mask R-CNN models were evaluated for precision, recall, F1-score, mAP50, and FPS metrics, and YOLOv11-Seg showed the best performance in both accuracy and speed (F1-score 0.7592, mAP50 0.7635, FPS 136.99). In particular, it demonstrated a high detection performance and a high generalization ability, maintaining the most stable detection performance, even in external, real-world fire training videos. On the other hand, the Mask-based model has relatively low real-time performance and precision, which limits its practical use.

Jacek et al. present a modern fire detection method that uses intelligent sensors, such as artificial vision systems and video cameras [171]. They utilize artificial intelligence to detect flames in places where the use of conventional sensors is impossible or severely restricted, enabling them to operate as a standalone platform in conjunction with environmentally friendly acoustic flame extinguishing technology. The authors present the possibility of using the acoustic method for fire detection and preventing and extinguishing fires regardless of the type of fire, whether liquid, solid, or gaseous. In particular, a mathematical model for temperature control in a local room and the detection of fires and outbreaks through the means of pulse acoustic probing was presented. The model consisted of three dependencies: a first dependency that calculates the technical parameters of the acoustic device designed for the room, a dependency that determines the average air temperature in the room, and a dependency that detects the occurrence of a fire in the room by detecting the exceedance of a threshold value of temperature rise per second. Based on this model and an artificial intelligence computer vision system, it is expected to be implemented in various fire protection systems, contributing to the construction of production lines, smart city infrastructures, Internet of Things networks, smart buildings, and evacuation plans.

There have also been attempts to improve detection accuracy by analyzing gases with AI, aside from smoke and flame. For example, efforts to improve the response accuracy of existing gas sensors have been studied by Efenwengbe et al. [172]. They note that conventional gas sensors are unable to accurately analyze the composition of gas streams in real time, citing their weaknesses, including mechanical wear or damage, and environmental factors, such as exposure to unusual temperature and pressure conditions. While machine learning models have been found to prevent such failures, they say the models still require users to spend unnecessary and additional resources to identify failures between sensor readings and machine learning predictions. Therefore, they propose a hybrid artificial intelligence system to monitor mixed gas concentrations in gas streams in real time, as shown in Figure 26, and to identify when sensors need to be recalibrated. They presented the XGBoost model to analyze the gas concentration. Utilizing Cr470 and Au475 sensors, the mean or median absolute error was used as the error limit.

While models utilizing lightweight networks exist, models for fire monitoring at very large scales are also being actively researched. Abdallah et al. proposed a fire detection model utilizing Earth observation (EO) and demonstrated the feasibility of fire detection based on the precise spatial, temporal, and spectral resolution of satellite-based EO [173]. In this study, they developed algorithms for detecting, classifying, and segmenting fires based on EO data and compared the performance of Autoencoder, U-Net, and convolutional neural network (CNN) models. Among them, the CNN model performed the best in terms of the balance between various performance metrics and the efficiency of the number of trainable weights. However, EO-based approaches are limited by the nature of satellite data, which can only detect fires after they have progressed to a certain point. Nevertheless, EO data is a viable alternative for AI-based fire risk classification, as it can compensate for the limitations of traditional methods in real-time detection in isolated areas or rugged topography where ground access is difficult.

Basturk et al. proposed a deep ensemble network (DEM) that combines various deep learning models for forest fire detection based on drone imagery [174]. Four CNN structures, Faster R-CNN, RetinaNet, YOLOv2, and YOLOv3, were combined to improve the fire detection performance in three ensemble methods, and the detection results from each model were merged based on logical operations to improve the accuracy and stability of the predictions. They utilized images and segmentation information from the Corsican Fire Dataset for training and generated a total of more than 200,000 fire images through data augmentation to train the model. Their testing on real-world drone imagery showed that the ensemble approach significantly improved the performance of the single model, with the simple combination approach being the most effective, especially when multiple small fires were present. This study demonstrates that fire detection is possible, even in complex drone imagery, and provides a direction for improving the reliability of real-time wildfire detection through complementary combinations of deep learning models.

The integration of gasistor-based sensors within Internet of Things (IoT) architectures marks a significant advancement in contemporary fire detection technologies. Owing to their miniaturized form factor and low energy demands, gasistors are particularly advantageous for deployment in decentralized IoT systems [146]. When embedded within wireless sensor nodes, these devices can continuously monitor and transmit data on fire indicative gases, such as carbon monoxide (CO), nitrogen oxides (NO_x), and ethane (C₂H₆), using communication protocols, like Wi-Fi, Zigbee, or LoRa. This connectivity supports real-time fire detection, remote surveillance, and automated emergency responses. Furthermore, integrating gasistors with edge-based artificial intelligence enables on-site data analysis and rapid decision-making, thereby minimizing latency and enhancing detection reliability. Such a synergistic approach holds considerable promise for applications in smart homes, industrial safety systems, and largescale environmental monitoring, including early forest fire detection as shown in

Table 8. Lightweight and high-performance models for early fire detection.

Category	Model	Data Size	Predction	Performance	Platform	Reference
Lightweight	Flame Net	Fire images (RGB)	Fire classification (binary)	Accuracy 99.4%, F1-score 0.99	Smart city, Raspberry Pi	[167]
Lightweight	YOLOv8n	Fire scene images	Fire classification (real-time)	Accuracy 95.21%, F1-score 0.985	DJI drone + Raspberry Pi 5	[168]
Lightweight	Mobile CNN	Smoke and fire images	Real-time smoke/fire detection	Accuracy 98.37%	Mobile device	[169]
Lightweight	DT/NB	CO/O2 gas ratio	Fire hazard (binary)	DT: 100%, NB: 93.8%	Coal mine, embedded	[175]
High-performance	XGBoost	Gas sensor data (Cr470, Au475)	Sensor error & gas concentration	R2 > 0.99	Industrial sensing	[172]
High-performance	EO CNN	EO satellite data	Fire detection/classification	Fire ratio > 0.8	Satellite-based monitoring	[173]
High-performance	VGG16-Flame	Flame images + HRR	HRR estimation	R2 > 0.8	Indoor/outdoor fire scenes	[176]
High-performance	Random Forest	Fire size, location, resources	Fire duration prediction	Classification error: 8.13%	National wildfire database	[177]

.

7.3. Real-World Applications of Deep Learning-Based Fire Detection Models

While many fire detection models are developed in controlled environments, only a few are successfully deployed in real-world scenarios. This section introduces deep learning-based fire detection models that have been tested or implemented in practical environments, such as smart cities, drones, electric vehicles, and satellite systems.

Among them, Fouda et al. proposed a hierarchical, lightweight AI framework tailored for edge computing using UAVs in forest fire detection [178]. The authors recognized the severe limitations of the onboard computing, energy, and communication resources of commercial drones and designed a dual-layer inference system. The architecture consists of a simple model (M1: decision tree or logistic regression) for low-complexity inputs and an advanced deep learning model (M2: custom CNN with global average pooling) for ambiguous cases. The key innovation of the framework lies in the adaptive confidence score thresholds, which are optimized using TOPSIS-based Pareto analysis to balance the detection accuracy and the computational cost. Experiments were performed using a publicly available dataset containing 1339 fire and 2668 non-fire forest images. The system was distributed on a simulated DJI Mavic Mini 2 configuration and demonstrated remarkable adaptivity. By setting the threshold to 0.2 in the DT-M1 + DL-M2 combination, they reduced the load of the deep model by 36.48%, while maintaining an overall accuracy of 95.53% and an F1-score of 0.9543. These results demonstrate that lightweight hierarchical inference can provide reliable real-time forest fire detection, even on limited hardware and is highly suitable for drone swarms and isolated environmental monitoring tasks.

Meanwhile, Kadir et al. developed a fast and lightweight deep learning model based on the YOLOv5-s architecture, focusing on the problem of post-fire detection [179]. Their work emphasized not only detection accuracy, but also the real-time usability potential for the de-deployment of drones. To achieve this, they augmented the YOLOv5-s backbone by incorporating an improved bottleneck cross-stage partial (CSP) module and a pyramid attention network (PAN) layer. These additions significantly improved the feature extraction capabilities while minimizing computational load. They also replaced the traditional ReLU with a hard-switch activation function to increase the convergence stability and maintain the gradient flow in noisy fire environments. The model was trained and validated on a custom aerial dataset consisting of 1900 labelled images (950 fire/950 non-fire), annotated using Roboflow. In terms of performance, the optimized model achieved an accuracy of 97.4%, a false positive rate of 1.258%, and a non-maximum suppression (NMS) latency of 3 ms, outperforming both the traditional CNN model and the Faster R-CNN in the comparison benchmarks. Although the model was tested on a CPU-based platform, its low parameter count (7.1 million) and fast inference suggest a strong potential for application in UAV systems for post-fire assessment in smart city and rural forest environments.

Several recent studies on AI-based fire detection have focused on integrating deep learning models into critical smart city infrastructure [180]. Smart cities are environments where small size, real-time responsiveness, low false alarms, and edge compatibility are critical. Dilshad et al. proposed the Optimized Fire Attention Network (OFAN), an attention-based deep learning architecture designed for fire detection within IoT-enabled smart city environments. Recognizing the need for fast and reliable processing at the edge of the network, such as street surveillance cameras or embedded processors, the authors combined MobileNetV3Small with lightweight attention modules (OCA and OSA) to create a small but accurate model; this model is designed to detect fires in the dark, fog, and clutter at night. They also introduced Diverse Fire, a large dataset that simulates complex smart city conditions. Through extensive training on this dataset, OFAN achieved over 96% accuracy on multiple benchmarks. They also kept the model size under 12 MB, making it highly deployable across a city-wide sensor grid and camera system. The system demonstrates how optimized deep models can support the real-time risk response in dynamic and visually complex urban environments.

Similarly, Talat et al. have developed a complete smart fire detection system (SFDS) based on the YOLOv8 architecture integrated into a multi-tiered smart city computing framework [181]. Their approach splits the system into IoT, fog, cloud, and application layers to enable efficient data collection, processing, and storage across city infrastructure. They specifically utilized YOLOv8’s real-time detection capabilities and anchor-free architecture to speed up response times in environments such as smart streets, hospitals, and transport systems. The system achieved 97.1% precision and demonstrated strong performance in a variety of lighting and environmental conditions. It is also modular in design and can be extended to other smart city monitoring tasks. This study demonstrates how modern object detection frameworks can be applied to city-scale safety management through a structured architecture and scalable deployment.

Recent advances in hyperspectral image and onboard AI computing have significantly expanded the potential of satellite fire detection systems. By directly processing data in orbit, this approach enables the rapid and large scale monitoring of wildfires. Unlike traditional observation methods that rely on post-downlink analysis, AI integrated into satellites can reduce latency, optimize data transmission, and enable real-time fire and smoke detection with energy-efficient operations. The following research is an example of the recent trend of integrating deep learning models into satellite systems for wildfire monitoring and early warning.

Lu et al. developed a lightweight, onboard AI system designed for fire smoke detection using hyperspectral imagery on the upcoming Carnini cubesat mission [177]. To address the computational and bandwidth limitations of the cubesat, they proposed the VIB_SD model, a small CNN with only 1.6 million parameters optimized for embedded inference on low-power hardware (Intel NCS2). Since real-world sensor data was not available before launch, the researchers synthesized a large training dataset using VIIRS imagery from historic Australian bushfire events covering four classes: smoke, cloud, mixture, and clear. Using Raspberry Pi and AI accelerators to build an emulation system that mimics the satellite environment, they were able to test the model’s accuracy, execution time, and energy usage before launch. The results showed that the onboard system could reduce the data downlink capacity by 84% and the energy usage by 84.3%, validating the feasibility of AI satellite autonomy. This research is a key step toward deploying intelligent and energy-efficient bushfire detection models on real-time Earth observation missions.

8. Challenges and Future Directions

8.1. Challenges

Different types of sensors for fire detection are critical for ensuring safety by providing early warnings of potential fire hazards. However, their performance and reliability are often hindered by challenges, such as environmental interference, false alarms, and limitations in detecting various fire types under diverse conditions. Smoke sensors encounter several critical challenges that affect their performance and reliability. One prominent issue is the occurrence of nuisance alarms, which are frequently triggered by cooking fumes, or elevated humidity levels. These false positives can result in users disabling or disregarding the alarms, thereby undermining the effectiveness of the safety system. Despite their proven effectiveness, flame detector-based fire detection systems exhibit specific challenges and constraints that must be carefully addressed during the design and implementation stages. False alarms can occur due to interference from light sources, like sunlight or artificial lighting. Environmental factors, such as extreme temperatures, humidity, and contaminants, can also impact the sensor’s accuracy. Proper sensor placement is crucial to ensure optimal coverage, as inadequate placement may lead to delayed or missed detections. Additionally, flame detectors must detect various flame types, but variations in size, intensity, and spectral characteristics can complicate detection. Gas sensors also face several challenges that can impact their effectiveness and reliability in accurately detecting fire-related gases across various environmental conditions and fire scenarios. The selectivity and the stability of sensors remain significant challenges in practical applications. Environmental humidity can notably influence the sensitivity of porous metal oxide semiconductors (MOS) due to the complex processes of surface adsorption–desorption and catalysis. As a result, the sensor’s performance may degrade under high humidity conditions, such as during exhaled gas detection, which hinders the reliable detection of target gases across varying humidity levels. Achieving a high response sensitivity while minimizing interference is essential for accurate detection, and enhancing selectivity helps prevent false alarms in practical settings [182]. UV flame detectors have certain limitations that need to be addressed to ensure effective fire safety. These detectors require an unobstructed line of sight to the flame, as UV radiation is easily attenuated by smoke, dust, or physical barriers. Moreover, they are less sensitive to smoldering fires, which emit minimal UV radiation until they transition into open flames. Under conditions of heavy smoke, their detection range, typically between 15 to 50 m, can be significantly reduced. Additionally, physical obstructions or the presence of UV-absorbing vapors, such as those from welding, lightning, or x-rays, can hinder or even prevent the detector from identifying the fire. Infrared fire detectors face challenges such as false alarms from non-fire sources and reduced performance in adverse conditions. They require a clear line of sight and can be affected by high ambient temperatures. IR flame detectors can exhibit sensitivity to modulated heat sources and light emissions. Additionally, the presence of water, snow, or ice on the detector’s optical components may impair its ability to detect a fire, either delaying or preventing detection. In wildfire detection, sensor networks, video monitoring, and machine learning have improved detection speed and accuracy, enabling quicker responses and better resource allocation. While each sensor technology offers unique advantages, their real-world applicability varies significantly.

Table 9. Comparative analysis of various sensor types used in fire detection, based on their effectiveness, limitations, and readiness for real-world deployment.

Sensor Type	Effectiveness (Sensitivity/Selectivity)	Limitations	Real-World Deployment Readiness	Ref
Gasistors	High sensitivity, fast response	Humidity/temperature sensitivity	High (compact, low power, IOT compatible)	[149]
MOS Sensors	Moderate sensitivity	High power, slow recovery	Moderate (requires heater, bulkier)	[183]
Electrochemical	High selectivity	Short lifespan, calibration needed	Moderate (limited by environmental factors)	[184]
NDIR	Accurate for CO2	Large size, expensive	Low (complex optics, high cost)	[185]
Optical Smoke Sensor	Early smoke detection	False alarms (dust/steam)	High (widely deployed but prone to nuisance alarms)	[186]

summarizes the effectiveness, limitations, and deployment readiness of the key sensor types discussed in this review, including gasistors.

8.2. Future Directions

In the realm of advanced fire detection technologies, memristor-based gas sensors, also known as gasistors, represent a promising frontier for future development. These hybrid devices, combining the functionalities of gas sensors and memristors, offer significant advantages in terms of enhanced gas detection performance and energy efficiency. The integration of memristor technology with traditional gas sensing materials has demonstrated remarkable improvements in sensitivity, selectivity, and stability. The unique properties of gasistors, including their ability to operate at room temperature and their rapid recovery using pulse biasing, make them particularly suitable for next-generation fire detection systems [12]. Furthermore, the precise control of CFs within the memristor structure enables more reliable and accurate gas detection, potentially allowing for the development of highly sensitive and selective fire alarm systems [187]. As research in this field progresses, we can anticipate the emergence of transparent, low-power gasistors that could revolutionize early fire detection in both industrial and residential settings. Moreover, these advancements in gasistor technology show great promise for addressing critical challenges in wildfire detection. By leveraging their high sensitivity, low power consumption, and ability to operate in remote areas, gasistors could significantly enhance early warning systems for wildfires. Their potential to reduce false alarms and enable real-time data processing directly at the sensor level addresses key limitations of current wildfire detection methods.

9. Conclusions

The field of fire detection has evolved significantly with the development of intelligent sensor technologies. While traditional systems are still widely used, they often fail to detect early fire indicators promptly and are highly susceptible to false alarms. Gasistor-based sensors, in particular, demonstrate considerable promise due to their high sensitivity to toxic combustion gases, low power consumption, miniaturized design, and rapid response capabilities. This review highlights how the integration of gasistors with AI and IoT platforms enhances system intelligence, enabling real-time monitoring, data-driven decision-making, and improved situational awareness. In addition to the points discussed, it is important to highlight that recent research has shown gasistors to be capable of achieving response times as low as 1–2 s and having sub-ppm-level detection limits for critical fire-related gases, such as CO, NO₂, and NH₃. Their ability to operate under room-temperature conditions or with minimal thermal assistance significantly reduces their energy consumption compared to traditional metal oxide sensors. Furthermore, advancements in fabrication techniques, including the integration of 2D materials and pulse-bias-enhanced recovery, are pushing gasistors closer to commercialization. When deployed in wireless sensor networks or edge-AI platforms, these sensors can deliver real-time, on-site fire analysis, minimizing latency and improving the emergency response in both residential and industrial environments. Such capabilities underscore their transformative potential in next-generation fire safety systems. Moreover, hybrid approaches combining multiple sensing modalities—such as gas, thermal, and visual data—further contribute to their increased reliability and detection accuracy. Despite these advancements, challenges persist, including the need for improved long-term stability, robustness in harsh environments, and standardized testing protocols. Addressing these issues through material innovation, algorithmic refinement, and system-level optimization will be key to the successful deployment of next-generation fire detection technologies across diverse applications.