Cyber-Physical Systems for Smart Farming: A Systematic Review

Abstract

In recent decades, climate change, increasing demand, and resource scarcity have transformed the agricultural sector into a critical field of research. Farmers have been compelled to adopt innovations and new technologies to enhance production efficiency and crop resilience. This study presents a systematic literature review, supplemented by a bibliometric analysis of relevant documents, focusing on the key applications and combined techniques of artificial intelligence (AI), machine learning (ML), and digital twins (DT) in the development and implementation of cyber-physical systems (CPS) in smart agriculture and establishes whether CPS in agriculture is an attractive research topic. A total of 108 bibliographic records from the Scopus and Google Scholar databases were analyzed to construct the bibliometric study database. The findings reveal that CPS has evolved and emerged as a promising research area, largely due to its versatility and integration potential. The analysis offers researchers and practitioners a comprehensive overview of the existing literature and research trends on the dynamic relationship between CPS and its primary applications in the agricultural industry while encouraging further exploration in this field. Additionally, the main challenges associated with implementing CPS in the context of smart agriculture are discussed, contributing to a deeper understanding of this topic.

1. Introduction

The trend in the world population is that the urban population is increasing, while the rural population is decreasing. Therefore, with this trend, the current rural population will age. Since the rural sector is the main food provider for cities [1], this could trigger a point where the demand for food from the population will exceed the supply, causing various consequences such as deprivation, malnutrition of children, and turning good nutrition into a privilege that very few will be able to access. Furthermore, during 2020, the world suffered the scourge of a global pandemic, and the global economic and food impact was enormous.

Food crises continue to be a significant global challenge, with hunger and food insecurity impacting millions of people across all continents. From 2019 to 2020, there was a significant increase in the number of people suffering from hunger, intensifying an already severe situation. About $9\%$ of the world’s population is affected by hunger, while roughly $41\%$ experience food insecurity, affecting billions worldwide [2,3,4]. The concept of food security encompasses four essential aspects: availability, access, utilization, and stability [5]. Despite advances in improving availability, access, and utilization in several areas, maintaining stable food systems continues to be a critical issue, mainly due to climate change. Challenges such as declining soil fertility, the spread of pests, trade interruptions, and transportation obstacles exacerbate the situation, highlighting the need for innovative and sustainable methods of food production [2,3].

Agriculture has been a cornerstone of human civilization and has shaped lifestyles since the very inception of humanity. Throughout history, six key periods can be identified that have revolutionized agricultural practices, bringing about significant changes. The investigation by Idoje [6] provides further details on these transformative periods.

Today, new technologies and strategies, such as sensors, the Internet of things (IoT), cyber-physical systems (CPS), machine learning, artificial intelligence (AI), digital twin (DT), and networks for agricultural systems [6], are implemented in what is known as smart farming or smart agriculture. Smart farming use cases include precision agriculture, irrigation control, automated drones, and, above all, smart greenhouses [7].

CPSs are essential in smart farming, as they combine computational algorithms with agricultural processes to allow real-time monitoring, control, and automation. These systems connect the physical and digital realms, supporting the continuous flow of data and feedback to improve various farming operations. Unlike technologies such as IoT or AI, which mainly deal with data collection, analysis, and prediction, CPSs offer an integrated framework that not only collects information but also empowers intelligent decision making and precise action within a unified system [8].

Through the integration of state-of-the-art sensing technologies, edge computing, and cloud analytics, CPSs provide farmers with critical insights that improve resource management, reduce operational expenses, and reduce environmental impact. The interconnected structure of the CPS enables adaptive reactions to changing environmental conditions, such as varying weather patterns, soil health fluctuations, and crop growth cycles, thus guaranteeing that agricultural procedures are continuously refined in real time [9].

Moreover, the adaptability and extensibility of cyber-physical systems make them suitable for diverse applications, ranging from compact farms to extensive industrial agricultural enterprises. They facilitate the unification of multiple subsystems, such as irrigation, pest control, and fertilization, into a unified self-management network that functions with minimal human input. This comprehensive method increases both productivity and efficiency and also facilitates sustainable agriculture by advancing data-informed decision making alongside precision agriculture methodologies [10,11].

Using the potential of cyber-physical systems, smart agricultural systems can achieve unprecedented levels of precision, reliability, and automation. These advances contribute to enhanced food security, reduced waste, and improved sustainability in modern agriculture.

A key application of such technologies is greenhouse farming, which has seen significant global adoption due to its ability to provide controlled environments for optimal crop production. Recent estimates indicate that greenhouse cultivation covers approximately 3.64 million hectares worldwide [12]. The popularity of greenhouses stems from their ability to regulate climatic, nutritional, and biotic factors throughout the various stages of growth, thus maximizing yields and ensuring consistent quality. In addition, greenhouse agriculture facilitates the production of off-season horticultural crops, improving both food availability and economic profitability [13]. The integration of CPS into greenhouse operations further enhances these benefits by enabling real-time monitoring and adaptive control, ensuring efficient resource utilization and sustainable agricultural practices.

Despite their advantages, greenhouses present significant challenges in terms of modeling and control. Numerous research efforts have focused on developing mathematical models for greenhouses based on energy balance principles [14,15,16,17,18]. However, the absence of a standardized model stems from the complexity of greenhouse ecosystems, where a multitude of interdependent variables interact in a highly dynamic and nonlinear manner. This complexity makes it difficult to achieve precise control using conventional model-based approaches.

Greenhouses exhibit strong nonlinear behavior, which poses significant challenges for the implementation of classical control strategies [19]. The complexity arises due to the intricate relationships between key variables such as temperature and humidity, the inherent delays in the control cycle, and the nonlinear interactions between physical and biological subsystems. As a result, achieving optimal climate conditions—often defined through extensive experimental knowledge by growers and researchers [14]—remains a challenging task. These challenges have historically limited the automation of greenhouse operations compared to other industrial sectors. However, with the advent of Industry 4.0 technologies, including CPS, IoT, and AI, the agricultural sector is undergoing a significant transformation, paving the way for smarter, more adaptive, and efficient greenhouse management [20,21].

In response to these challenges, researchers have investigated a range of sophisticated control methods that extend beyond conventional techniques to effectively handle the dynamic characteristics of greenhouse environments. Notably, adaptive control methods, AI-based approaches, and hybrid control strategies have been identified as promising solutions. Adaptive control systems are designed to dynamically modify their parameters in reaction to evolving conditions, thereby enhancing robustness and efficiency. Concurrently, AI-based methods, including machine learning and deep learning, utilize extensive datasets to fine-tune decision making and precisely forecast environmental changes. Hybrid control methods, which integrate model-driven and data-oriented approaches, provide a comprehensive solution that utilizes both theoretical insights and empirical data.

Incorporating these advanced control strategies into CPS frameworks could transform greenhouse automation, allowing real-time modifications and improved predictive control. By embedding AI algorithms and adaptive mechanisms, the CPS can offer intelligent decision support systems that maximize resource efficiency, reduce energy use, and improve crop productivity. This progression signifies a major transition from traditional manual or semi-automated systems to fully autonomous, self-regulating greenhouse setups that can proactively adapt to environmental variations.

Through a systematic literature review and bibliometric analysis, it is possible to identify the current state of cyber-physical systems in agriculture, highlighting research trends, key areas of development, and existing gaps. By classifying the studies reviewed according to their research focus, this analysis provides a structured understanding of CPS applications, facilitating the identification of technological advancements, limitations, and future research directions necessary for their effective implementation in the agricultural sector.

In summary, the key contributions of this work are as follows:

• Comprehensive Literature Review: A systematic analysis of existing research, including methodologies, bibliometric data, application areas, and current debates, is conducted.
• CPS Technique Taxonomy: A classification system for common CPS techniques is developed, supported by illustrative research examples.
• Practical Research Framework: A standardized framework is proposed to guide researchers in conducting applicable studies and sharing their CPS-based solutions.

The remainder of this paper is organized as follows: Section 2 introduces the origins and fundamental concepts of the CPS. Section 3 presents the systematic literature review methodology. Section 4 shows the findings derived from the bibliometric analysis. Section 5 presents the most interesting research on CPSs in smart agriculture, and Section 6 explains some remarks. Finally, Section 7 provides the conclusions.

2. Origin and Fundamentals of CPSs

The term “Industry 4.0” emerged within the initiative of the German government “High-Tech Strategy 2020 Action Plan” in 2011 [20]. Other countries have presented similar proposals and strategies, such as the USA with its “Industrial Internet ” program and China with its “Industrial + ” program [21].

With the arrival of Industry 4.0, new concepts and paradigms were established. The term “cyber-physical system” was coined by the National Science Foundation (NSF) of the United States around 2006 [22]. Such systems aim to connect genuine physical things with ubiquitous computing and networking technologies. That is, taking the physical system (sensors, actuators, analog-to-digital converters, digital-to-analog converters, and so on) and the virtual system (data, signals, communication networks, and so forth) and coupling them within a new all-encompassing system called a cyber-physical system.

A CPS has the potential to perceive and even understand physical changes in its environment, analyze these changes, consider their impact, and ultimately make decisions. The purpose of these decisions is to control the physical elements of the system, thereby affecting its environment autonomously [22,23].

Cyber-physical systems are the next-generation paradigm for designing systems that are increasingly being adopted in industry, with the ability to describe, manage, and integrate heterogeneous systems [24]. Gunes defines CPSs as “complex, multidisciplinary, physically aware next-generation engineering systems that embed embedded computing technology (part cybernetics) in physical phenomena through the use of transformative research approaches” [22]. Rajkumar describes CPSs as “physical and engineering systems whose operations are monitored, coordinated, controlled, and integrated by a computing and communication core” [25]. Zhou describes CPSs as “intelligent systems, where sensors, actuators, and controllers are integrated to support the interaction between the physical world and the cyber world” [26]. Burg describes CPSs as “complex and heterogeneous distributed systems typically consisting of many sensors and actuators, which are connected to a group of computing nodes” [23]. Khaitan describes a CPS as a “system of systems” in which complex, heterogeneous systems interact continuously. Proper regulation of such systems requires careful codesign of the overall CPS architecture [27].

It is worth mentioning that there is no standardized model of a CPS; however, there are specific characteristics that a system must have to be considered a CPS. Authors such as Burg [23], Tan [28], and Rajhans [29] propose that the architecture of a CPS is composed of three layers. Figure 1 illustrates the schematic of the standard CPS architecture, with each layer or sub-layer depicted in a different color.

• Physical layer. In this layer, we find sensors (temperature, humidity, pressure, light, movement) and actuators (motors, luminaries, valves) according to their application (purple layer).
• Communication layer. This layer serves as a means of information exchange for the different devices that make up the system. Additionally, it allows elements of the perception layer to access the data generated by the sensors in the physical layer (blue layer).
• Perception Layer.

  −

  Management Sub-layer. This part of the architecture is responsible for supervision (assisted, artificial, hybrid). Within this sub-layer are the HMIs, data managers (big data), and virtual observers (AI, ML) (green sub-layer).

  −

  Control Sub-layer. This layer can also be called the control layer. Generally, in this layer, decisions are made based on the results obtained from the physical layer (yellow sub-layer).

In the basic architecture of a CPS, the sensors and actuators form a network. The sensor–actuator network (SAN) consists of sensors and actuators to enable close interactions between humans and the environment [30]. SANs are fundamental in the development of a CPS; the coordination in the management of information to know how to make the appropriate decision depending on the scenario in which the system is managed provides the CPS with reliability. In some cases, the sensors are wireless, so the network is called a WSAN. Several parameters must be considered when assigning the control task to an actuator, such as actuator capabilities, guarantee, task completion time, power consumption of each actuator, and physical system requirements [22].

3. Methodology of the Systematic Literature Review (SLR)

This review used an SLR methodology based on four stages: definition of research questions, search process, studio selection, and quality assessment [31,32,33]. The first step to establishing the research questions was to define the objectives and scope of data selection and synthesis. After this, the next step was to search for articles in prestigious databases. Scopus was selected because of its recognized prestige and extensive indexing of high-quality scientific publications. In addition, it includes articles from other reputable databases and publishers, such as Web of Science (WoS), IEEE Xplore, and Springer. This overlap is illustrated in the Martin’s Web application [34], which presents a Venn diagram demonstrating the intersections of citations between various databases for specific topics and subtopics. Figure 2 illustrates the overlap between Google Scholar, Scopus, and Web of Science in the fields of “Engineering & Computer Science”, “Automation & Control Theory”, “Artificial Intelligence”, and “Sustainable Energy”.

During the search process, a structured search equation was formulated using relevant keywords and synonyms to capture the core aspects of the topic. The search equation was defined as (“SMART FARMING” OR “GREENHOUSE CONTROL” OR “SMART AGRICULTURE”) AND (“CYBER PHYSICAL SYSTEM” OR “CPS”). This equation was used in the database mentioned before. This result was obtained from 100 Scopus papers. These articles form the primary literature field and cover the period from 2018 to 2024.

To incorporate earlier studies and establish a historical perspective, an additional search was conducted in Google Scholar, resulting in 53 articles published between 2008 and 2017.

It is important to note that despite the wide coverage of high-quality research ensured by the selected databases, certain biases may persist. One potential limitation is publication bias, as studies reporting significant or positive results are more likely to be published than those with inconclusive findings. Furthermore, regional bias may be present, as the selected literature could disproportionately reflect research from regions with higher academic activity in this field, potentially overlooking advancements in areas with fewer scholarly publications. However, efforts were made to mitigate these biases by incorporating studies from multiple sources, spanning a diverse range of publication years, and applying rigorous selection criteria to ensure a comprehensive and balanced review.

The temporal scope of this review was set from 2008 to 2024 to capture the evolution of the CPS in smart agriculture. The starting point was chosen considering the initial emergence of CPS concepts applied to agriculture, allowing an analysis of their development over time. However, greater emphasis was placed on studies published from 2018 onward, as this period reflects a growing trend in research and technological advances in CPS within the agricultural sector. This approach ensures that the review highlights the most relevant and up-to-date contributions while maintaining a historical perspective on the field’s evolution.

Exclusion criteria were established to limit the volume of literature and select the most crucial articles for research. Only articles, conference papers, journal-type sources, and publications in English were considered. Following this process, a file was created that contains the information from all articles that met the established criteria. This allowed for the verification of duplicate documents, resulting in the removal of 12 duplicate records.

Subsequently, quality criteria were established to further narrow the number of documents. Each document was carefully reviewed, and those that were most closely aligned with the research question were selected. Peer review is an exclusion method applied during the final stage of selection. This method is used solely to eliminate articles that do not align with the main focus of the investigation, minimizing the risk of information bias that could favor the investigation. Consequently, the data set comprises 108 documents for analysis.

Table 1. Summary of the descriptive information.

Description	Results
MAIN INFORMATION
Timespan	2008:2024
Sources (journals, books, etc.)	86
Documents	108
Annual growth rate %	9.05
Document average age	3.81
Average citations per doc	53.01
References	5061
DOCUMENTS TYPES
Articles	35
Conference paper	68
Review	5
DOCUMENTS CONTENTS
KeyWords Plus (ID)	989
Author’s keywords (DE)	367
AUTHORS
Authors	401
Authors of single-authored docs	5
AUTHORS COLLABORATION
Single-authored docs	5
Co-authors per doc	4.34
International co-authorships %	25

provides a summary of the literature database, including the total number of articles, the total number of authors involved, the average age of the documents at the time of writing of this article, and the average number of citations per document. It also specifies the time range considered for the creation of this database.

Figure 3 illustrates the review framework, emphasizing the document selection process and the removal of duplicate records. The selection followed strict eligibility criteria, focusing on studies related to smart agriculture, CPS, ML, or DT applied in the agricultural sector.

4. Bibliometric Analysis

Bibliometric analysis extracts the most relevant information from the trends and quantitative data of the research topic using mathematical tools [35]. The Bibliometrix package from RStudio 2023.12.1 was used to perform this analysis because of its ease in developing mapping and analyzing information based on keywords, author keywords, sources, institutions, etc. All data presented in this section come from the bibliographic database found in Section 3.

4.1. Co-Word Analysis

The Bibliometrix package offers the possibility of conducting a comprehensive, conceptual, and structural analysis of the compiled bibliographic database. This tool allows for the identification of key concepts and terms that are most frequently repeated in the titles, abstracts, and keywords of each article [35]. As a result, the analysis reveals trends and the current state of the art in the investigated topic.

The analysis of co-word or co-occurrence keywords identifies the principal keywords in the bibliographic database. The word cloud presents the most frequent keywords, and the size of each word is determined by the frequency of term occurrences. This type of graph is very useful because of the ease and speed with which the most relevant topics in a field of research can be identified.

To enhance the analysis, two “word cloud” graphics were generated based on the KeyWords Plus and authors’ keywords. During their development, it became evident that the authors often use different terms or synonyms to refer to the same research fields, resulting in inconsistencies in the representation of keywords. To address this, a subdatabase was created that consolidates all synonyms and variations used to refer to the same topics, ensuring a more cohesive and accurate analysis.

In addition, another database was created to exclude words that did not contribute to the analysis derived from the word clouds. Terms such as “review”, “graphs”, “prolog”, or “surveys” were omitted, as they do not provide meaningful contributions without the appropriate context.

Figure 4 shows the word cloud of keywords. In this graph, the most frequent word is CPS, as expected. Other frequent terms include smart agriculture, intelligent farming, IoT, and embedded systems. These terms are closely related to CPS, and in certain case studies, they can refer to the same or very similar concepts. In addition, trending techniques such as AI and ML are highlighted. These techniques, which have been research fields for the past few years, have been used in recent studies to improve the implementation and performance of CPS [36,37].

Figure 5 shows a word cloud obtained from the authors’ keywords. The term CPS is the predominant element, highlighting terms such as smart farming, smart agriculture, ML, IoT, and AI. One of the main reasons for selecting as much literature data as possible was to allow this comparative analysis between authors’ keywords and KeyWords Plus. The authors’ keywords provide an overview of the latest trends that CPS has been involved in, such as digital twins, wireless networks, blockchain, and neural networks, fields that are very promising for research. In contrast, the keyword graph provides an overview of the more established fields of study in which CPS has been involved, such as crops, automation, and embedded systems. These are important fields, but they are already in a more mature state than those shown in the authors’ keywords graph.

To establish the relationship between the KeyWords Plus, a co-occurrence network was created. In the development of the analysis and for a better understanding, the terms “smart agriculture” and “smart agriculture” are treated as synonyms. Similarly, the term “embedded systems” was considered synonymous with the term “CPS”.

The co-occurrence network highlights the strongest connections between the KeyWords Plus in the database. A key challenge in its development was determining the optimal number of nodes and establishing an appropriate attraction force for the terms to ensure that the graph produced reliable, accurate, and unbiased results. Ultimately, the graph identifies three primary nodes, representing the central axis of the research, the most prominent technologies, and examples within the application field. This analysis is shown in Figure 6.

The node network establishes CPS as the central focus of the research database. The principal node in the co-occurrence network is represented by the red node, which highlights the strong associations of CPS with foundational concepts (e.g., automation, information management, decision support systems, and the physical world), application fields (e.g., animals, crops, smart agriculture, and water management), and emerging techniques (e.g., ML, AI, and DT). These connections underscore the development, versatility, and applicability of CPS in the agricultural industry.

The red node represents the network with the strongest correlations to CPS, it is unsurprising that “smart farming” emerges as the most prominent topic. Furthermore, themes such as “virtual reality”, “physical world”, “decision making”, “real-time systems”, “cyber security”, “security systems”, “risk detection”, and “anomaly detection” are evident and exhibit strong correlations. These relationships are discussed in more detail in Section 2.

It is worth highlighting the presence of management topics such as “water management”, “energy use”, “decision making”, and “information management”. These topics within the main network denote the importance that CPS has taken on within resource management applications. These themes within the main network highlight the significant role that CPS has assumed in management and resource optimization applications during the production process.

This is exemplified by Radini’s research [38], which focuses on the energy required for urban water treatment. By implementing a CPS framework, his study demonstrates improvements in system performance and a significant reduction in the energy needed for the process.

In his review, Inderwildi [39] highlights various applications that underscore the impact of technologies such as big data, ML, and IoT in the industrial sector. He demonstrates how the integration of these innovative technologies within a CPS framework has led to energy optimization in production and supply processes, thus enhancing economic viability and reducing environmental impact.

The green node centers on IoT technology and its strong correlations with various aspects. It highlights trending application fields in recent years, including food supply, agricultural productivity, agricultural technology, and Industry 4.0, among others.

Examples of the close relationship between IoT systems and CPSs can be seen in works such as Zamora’s [40], Sahi’s [37], or Agarwal’s [41], which focuses on developing an application that uses IoT to monitor and manage crop information in real time. Similarly, Haris’ [42] study presents the development of a vertical farming system based on CPS architecture, while Dineva’s [43] research presents the results of implementing a livestock monitoring application. These studies emphasize the effective integration of IoT and CPS in data management applications, showcasing their efficiency and adaptability across diverse agricultural contexts.

The smallest node, the blue node, associates the concept of greenhouses with one of its main research areas: humidity control. Controlling environmental conditions within greenhouses through the implementation of novel controllers remains an attractive niche for researchers. For instance, Potdar’s [17] study focuses on the development and simulation of a fuzzy temperature controller, while Kang’s [44] research highlights a semi-autonomous greenhouse developed based on CPS architecture. Reviews of the literature by Bersani [7] and Duarte [45] further exemplify the focus of the blue node, presenting a comprehensive database of applications for internal greenhouse climate control.

4.2. Thematic Evolution

The thematic evolution of KeyWords Plus offers a historical perspective on the development of the terms most frequently used in the study of the implementation of CPSs in smart farming and greenhouses, making it easy to identify which research topics have persisted over time [35]. In addition, the graph highlights the most significant research topics over the years and their connection to emerging trends. KeyWords Plus are employed to gain a complete understanding of keywords associated with the content of documents. The dimensions of the boxes in Figure 7 reflect the frequency of occurrences of keywords and associated topics.

Figure 7 illustrates the progression of the terms most frequently used in CPS research, as revealed by the co-occurrence network from 2008 to 2024 (April). The graph is segmented into three distinct periods.

The initial period, which spans from 2008 to 2017, is characterized by the frequent occurrence of “greenhouses” and “CPS” as KeyWords Plus in research articles. During this time, “CPS” began to draw significant attention from researchers, largely due to its recent introduction within the context of Industry 4.0. In contrast, “greenhouses” had already been well established as a critical area of research in previous decades, driven by growing concerns over a potential global food crisis.

From 2018 to 2020, technical and management terms such as “energy utilization”, “food supply”, and “big data” gained a greater prominence. Furthermore, there was a notable shift in the focus of research articles, as the term “greenhouse” was increasingly aligned with “big data”, indicating a growing emphasis on management techniques aimed at improving greenhouse efficiency. During this period, the term “CPS” emerged as the most frequently cited keyword, underscoring its growing significance.

In the final period, 2021–2024, “CPS” remains a central term. Still, the scope of research has expanded to include terms such as “agricultural robots”, “learning systems”, “decision-making”, and “agricultural productivity. These terms represent specific tools for applying a “CPS” in agriculture. Furthermore, terms such as “food supply”, “greenhouse”, and “big data” have become increasingly associated with CPS. In particular, the term “greenhouse” is now related to Industry 4.0, while “energy utilization” continues to show consistency with trends from the previous period.

Thematic evolution demonstrated that previously unrelated fields have converged under the CPS concept. This indicates that CPS now has a wider range of applications and is increasingly recognized as an important niche field of research.

4.3. Thematic Map

Figure 8 presents a thematic map of the database, divided into four quadrants based on the degree of development and relevance of the research topics.

Quadrant 1 (Motor Themes): This quadrant includes four keywords, “CPS”, “smart agriculture”, “IoT”, and “food supply”, which demonstrate the highest relevance and significant development in research, with “CPS” being the most prominent. Additionally, this quadrant partially includes keywords like “energy utilization”, “convolutional neural networks”, “open source software”, “open systems”, “machine learning”, “agricultural robots”, “AI”, and “cyber security”.

Quadrant 2 (Basic Themes): This quadrant partially includes four keywords: “machine learning”, “agricultural robots”, “AI”, and “cyber security”. Although these topics are not yet highly developed, they are exhibiting an upward trend and represent emerging areas of research interest.

Quadrant 3 (Niche Themes): This quadrant includes terms such as “energy utilization”, “convolutional neural networks”, “open source software”, and “open systems”, which are primarily techniques used for application development.

Quadrant 4 (Emerging or Declining Themes): This quadrant contains terms related to emerging research areas, such as “virtual reality” and “event modeling”.

The current systematic review employed various analyses conducted using the Bibliometrix package in RStudio. Although these analyses were performed automatically, their results could be adjusted by modifying the variables provided by Bibliometrix. To address potential biases, the review incorporated the ROBINS (Risk of Bias in Non-Randomized Studies) tool, which established the criteria for bias assessment. Furthermore, the recommendations from the work published by Ciapponi [46] were implemented. Through this process, the authors collectively evaluated and approved the results, ensuring their accuracy and minimizing bias.

5. CPS in Smart Agriculture

Greenhouses are used by engineers and researchers as a tool for controlling agricultural factors. In the absence of an efficient agricultural system, the application of water, fertilizers, and pesticides is a waste of time, money, and effort. As a result, it is essential to have an efficient and intelligent system that can identify plant stress early on. In addition, it avoids crop loss, has minimal operational costs, and ensures human food security.

Changes in the environment, such as temperature, humidity, light intensity, and the appearance of new diseases, cause stress in crops, which can lead to a loss of yield. Unfortunately, the way things are handled now is not working well enough. However, different sensors have been used to track things like stress, temperature, and humidity in the agricultural environment. Sensors, on the other hand, cannot work on their own. An effective CPS implementation requires seamless integration between sensors, computational systems, and agricultural robots to analyze data and enable informed decision making.

However, the implementation of a CPS in agriculture presents several challenges. From a technical implementation perspective, the interoperability of heterogeneous devices (including different equipment brands and communication protocols), the reliability of data transmission in remote areas, and the real-time processing of large datasets pose significant obstacles. In addition, ensuring the cybersecurity of these interconnected systems is crucial, as vulnerabilities could compromise the integrity of the data and the functionality of the system.

From a financial perspective, adopting a CPS requires substantial investment in hardware, software, and infrastructure, which may be prohibitive for small- and medium-scale farmers. Maintenance costs, along with the need for specialized personnel to manage and interpret the data, further increase the economic burden. These limitations must be carefully considered when assessing the feasibility of CPS implementation in the agricultural sector.

Researchers are now using technology such as cyber-physical solutions (CPSo) [7], agricultural robots, and sensors to build systems to tackle challenges in precision agriculture (PA). Therefore, the inclusion of technology in agricultural processes aims to optimize its production, which is achieved through the continuous monitoring and management of resources such as water and fertilizers [47]. Hence, the following paragraphs show some of the research works in which CPSs have been applied in agriculture, in addition to briefly describing their contributions.

Sisyanto [48] discusses an example of CPSs. A smart hydroponic farming system that can be monitored online via Telegram Messenger has been developed. The prototype is designed using Raspberry Pi 3. The design that is created can monitor important parameters in the hydroponics system, such as light intensity, room temperature, humidity, pH, nutrient temperature, and electrical conductivity (EC).

Mehdipour [49] discusses the design of “smart pest control” (SPeC). The author proposes a monitoring system of the cultivation fields where potential pests may exist, with a special focus on rats; this CPS consists of detecting the body temperature of the pests, acquiring data through a network sensor, and later making the data processing system send an alert message to the manager to monitor the field.

Irfan [50] proposes intelligent field monitoring. In his article, SPeC is shown to track the behavior and activities of the rat. A new approach for animal detection was developed using ToxTrac and the NS2 simulator and compared to traditional expert/farmer control methods. The results obtained by applying this method were more efficient in the resources used (chemicals and pesticides), and the simulation showed a high degree of effectiveness.

Antonopoulos [51] proposes a CPS-based solution that can be used in any type of crop. The proposed scheme consists of a network of wireless sensors that determine the conditions of the crop in real time, establishing its requirements. In this way, the use of water and fertilizers is optimized.

Radu [52] presents a precision agricultural management integrated system architecture based on CPS design technology. His proposal focuses on monitoring potato cultivation and providing a low-cost system to farmers with limited financial resources; moreover, his proposal shows innovative techniques in each of the layers that make up the architecture system, thus optimizing the use of water and chemicals (fertilizers and pesticides), causing a greater economic return for the farmer.

Et-taibi [53] presents a real-world application of the implementation of a CPS in agriculture. Based on the basic structure of a CPS, an automatic irrigation system is developed controlled by fuzzy logic, powered by a photovoltaic system that, thanks to a network of wireless sensors, monitors the variables of temperature and humidity of the cultivation soil and the level of the water tanks. The appropriate humidity ranges of the system implemented in the design of the fuzzy controller were established based on the knowledge of experts. The results obtained show an increase in the efficiency of the use of resources and highlight the low cost of the hardware used.

Kang [44] proposes a semi-autonomous system for greenhouse control based on the knowledge of experts combined with automatic learning. Within the work, the scheme of the system that it applies is presented. In addition, a simulation is performed in which the monitored temperature and a prediction are compared using an algorithm called lightGBM. Finally, Kang presents in his work a method of greenhouse temperature control based on the opening and closing of windows, obtaining good results as a temperature monitoring control.

Dusadeerungsikul et al. [54] presented a collaborative control protocol for robotic and cyberphysical systems (CCP-CPS). The CCP-CPS, which enables the application of robotics in a CPS system for smart and precision agriculture (PA), aims to monitor and identify stresses in greenhouse crops using a hyperspectral analysis method. In addition, the CCP-CPS assigns tasks to agents and identifies and resolves system conflicts and errors.

Currently, some proposals show a hybrid system. In these works, it is mentioned that the next step is to implement a CPS that manages crop production. In this study [55], an architecture for a future agricultural enterprise is proposed as a complex system. The architecture takes into account issues of sustainability and adaptability to changes in the market and the environment.

Yahata et al. [56] presents the beginning of a CPS for “smart agriculture”. In this study, two image detection methods are shown that allow automatic observation to capture pods and flowers. This article considers a section of a CPS where the condition of flowers and pods is included in big data on the growth status of agricultural plants and environmental information to analyze and extract valuable rules for proper cultivation.

In the world, there are very aggressive climates for crops; for this reason, researchers have had to use alternative methods of cultivation. Liu [57], in his work, implemented a micro-sprinkler system for gray jujube trees planted in an arid region. After experimenting for about two years, the researchers obtained promising results, and observed a significant reduction in DPV when air temperature decreased and relative humidity increased. This translates into obtaining higher-quality fruits. In addition, the study evidenced the optimization of water when using this technique.

Garro et al. [58] propose the design of a robotic greenhouse, which essentially involves the greenhouse and a mobile robot interacting with the surrounding physical environment (collecting data and acting on plants). Therefore, the most important contribution is to utilize real-world knowledge and generalize it through the utilization of tried-and-true methods when creating a CPS.

Guo et al. [59] proposed a CPS-oriented framework and workflow for the control of agricultural greenhouse stressors. The system is termed MDR-CPS, which stands for monitoring, detecting, and responding. Monitoring, identifying, and responding to the many types of stress that can be experienced are the primary goals of the MDR-CPS. In addition, MDR-CPS makes use of CCT (collaborative control theory) to implement CRP (collaborative requirements planning), deal with CEs (conflicts and errors), and enable a collaborative architecture for improved CPS interactions.

Inderwildi et al. [39] have published an extensive review article that examines the current influence digital technologies within CPS are having and the prospective impact of these technologies on the decarbonization of energy systems. An ad hoc calculation for chosen applications of CPSs and their subsystems not only evaluates the potential for economic effect but also calculates the potential for emission reduction. This analysis makes it abundantly clear that the digitalization of energy systems through the utilization of CPS radically modifies the marginal abatement cost curve (MACC) and opens new avenues for the transition to an energy system with a lower carbon footprint. In addition, the evaluation concludes that when CPS is joined with artificial intelligence (AI), decarbonization might advance at an unforeseen rate while simultaneously presenting unknown and perhaps existential hazards.

Duarte-Galvan et al. [45] present an interesting review of the advantages and disadvantages offered by the different control techniques implemented in greenhouses, from conventional controllers (PID, ON/OFF) and optimal controllers (MPC, fuzzy logic, artificial intelligence, genetic algorithms, networks of neurons). In his review, he highlights the advantages of optimal controllers and the promising results that could be achieved through their implementation. However, their primary drawback is the high computational demand that they require. In contrast, conventional controllers offer the advantage of easy implementation, though the simulations conducted show a noticeable decline in performance compared to optimal controllers.

In this section, we presented several of the most prominent research works identified in the selected literature database. These studies were categorized into four main fields of application: greenhouses, smart farming, Industry 4.0, and cyber security.

Within the classification performed, the field of “smart farming” stands out as the most prominent, being the category with the highest number of research papers, as shown in Table 2. This category includes documents focused on topics related to the agricultural industry.

Research works in this field often involve wireless sensor networks (WSN) designed to collect large volumes of data (big data) from cultivation areas. Examples include systems proposed to determine the viability of soils for exploitation [60] and the creation of virtual environments capable of simulating crop exposure conditions [61,62,63,64,65]. These advances contribute significantly to increasing production and reducing losses by addressing critical supply chain needs [66,67]. They enable precise monitoring of crop health through observation of anomalies in individual plant leaves [68,69,70,71], the detection of pests [49,50], and the prediction of potentially damaging environmental events, such as extreme cold temperatures [72].

Cyber security stands out as the second most represented category after smart farming, which aligns with the inherent vulnerabilities of CPS architectures. The centralization of information within these systems makes them attractive targets for cyber attacks [73,74]. Evaluating potential threats [75,76], assessing the effectiveness of existing security strategies, and proposing innovative approaches to improve cyber security within smart agriculture have become increasingly relevant research areas. Cyber attacks pose risks not only for individual devices, such as agricultural machinery [77], but also to entire supply chains [78], highlighting the critical need for robust cyber security measures as technology adoption accelerates in agriculture.

Table 2. CPS application fields.

APPLICATION
Greenhouse	[7,12,13,14,16,17,18,40,44,54,58,59,79,80]
Smart Farming	[42,53,60,61,66,67,68,81,82,83,84,85,86,87,88] [62,63,64,65,69,70,72,89,90,91,92,93,94,95,96] [36,37,38,41,43,71,97,98,99,100,101,102,103,104] [6,55,105,106,107,108,109,110]
Industry 4.0	[111,112,113,114,115,116,117,118,119,120,121] [22,26,27]
Cyber Security	[73,75,76,78,122,123,124,125,126,127,128] [41,74,77,94,95,96,100,129,130,131,132]

Table 2. CPS application fields.

APPLICATION
Greenhouse	[7,12,13,14,16,17,18,40,44,54,58,59,79,80]
Smart Farming	[42,53,60,61,66,67,68,81,82,83,84,85,86,87,88] [62,63,64,65,69,70,72,89,90,91,92,93,94,95,96] [36,37,38,41,43,71,97,98,99,100,101,102,103,104] [6,55,105,106,107,108,109,110]
Industry 4.0	[111,112,113,114,115,116,117,118,119,120,121] [22,26,27]
Cyber Security	[73,75,76,78,122,123,124,125,126,127,128] [41,74,77,94,95,96,100,129,130,131,132]

On the other hand, the “Industry 4.0” category encompasses research that ranges from reviews of innovative technologies and their emergence within the fourth industrial revolution to application studies specifically tailored to the farming industry. For example, Schoitsch [111] presents a comprehensive investigation tracing the evolution of the industrial sector from its origins, examining its societal impact, and outlining ethical guidelines for fostering a “Smart Society” or “Society 5.0”. Similarly, Bernhardt [115] offers a comparative analysis of Industry 4.0 application approaches, exploring how outcomes vary across regions. This analysis considers factors such as the maturity of technological infrastructure and business policies related to information sharing, providing insight into the differences and similarities between diverse regional contexts.

Among the practical applications developed by the authors are the works of Lestari [116] and Bapatla [117], which exemplify this approach by showing the development of classifiers for agricultural products such as potatoes and livestock. Using novel techniques, these authors effectively address the everyday challenges faced by farmers and ranchers, highlighting the transformative potential of smart agriculture technologies to improve agricultural practices and increase productivity.

In the first section of Table 2, research works are presented in which authors such as Temelkova [79] and Zamora [40] describe specific CPS applications designed for greenhouses. These studies include the definition of critical parameters and variables (e.g., humidity, temperature, water, lighting, pH) and detailed steps for development and implementation, and they highlight the advantages of CPS applications in the agricultural field. Similarly, the research by Gkoulis [80] highlights the incorporation of IoT into an event-based platform, tested through its implementation in greenhouse humidity control, further demonstrating the integration of emerging technologies in agricultural processes.

The development of CPSs has been significantly improved through their integration with emerging technologies. In this context, it was considered appropriate to establish a classification that reflects the volume of research incorporating CPS concepts alongside these advances.

Based on the bibliometric analysis conducted in this study, particularly the word cloud of authors’ keywords (Figure 5) and the co-occurrence network of KeyWords Plus (Figure 6), it was determined that ML, AI, and DT are the most recurrent emerging technologies within the analyzed literature. Specifically, ML appears in 15% of the documents, AI in 21%, and DT in 12%, highlighting their prominence in recent research. Although IoT also has a significant presence (63% of occurrences), it was categorized under the classical group, along with studies employing more traditional methodologies within the CPS framework. This decision was based on its extensive adoption in earlier research phases and its established role in CPS development. By integrating these quantitative measures, this classification ensures a data-driven approach, strengthening the robustness of the thematic grouping.

Table 3. CPS combination fields.

CPS
Classical	ML	AI	DT
[79,94,119,129] [38,43,120,131] [41,88,101,132] [77,102,108,121] [22,24,55,133] [28,29]	[36,71,100,130] [97,103,104,107] [65,69,84]	[37,85,93,98] [69,103,104,106] [64]	[95,96,99,105] [92]

presents four primary categories: classical, ML, AI, and DT. Although the classical approach remains the most widely adopted, ML and AI techniques have gained increasing attention in recent years, emphasizing their expanding potential. Although the DT category contains the fewest studies, its significance lies in its role during the testing phase and preimplementation stage of CPS, ensuring more reliable and efficient system deployment.

6. Some Remarks

Cyber-physical systems offer multifaceted benefits to modern agriculture, allowing real-time monitoring and control of processes to optimize resource use, increase efficiency, and reduce waste. Precision management of crop cultivation, data-driven decisions, and automation reduce costs, increase productivity, and promote sustainability. CPSs improve early detection of crop issues, reduces manual labor needs, and incorporates weather data for informed decisions. In addition, they allow remote monitoring, fuel research and development, and are adaptable to various farms. Early adoption of CPSs gives farmers a competitive edge by producing higher-quality products efficiently and sustainably, strengthening their position in the agricultural industry, ensuring food safety, and contributing to global food security.

The integration of novel technologies such as ML and AI within the perception layer of the CPS represents a promising and nascent field of research. This integration enables CPSs to fully realize their initial potential by enhancing their capacity for advanced data interpretation and decision making.

Incorporating AI into the core control mechanisms of a CPS facilitates the development of intelligent managers capable of making rapid and effective decisions in complex scenarios. This is particularly beneficial in highly nonlinear systems such as greenhouses, where traditional modeling and control strategies face significant challenges. AI’s adaptability allows for the creation of efficient controllers that can dynamically adjust to changing conditions. Furthermore, by effectively interpreting intricate data, AI can improve human–machine interaction (HMI) by offering a more user-friendly and intuitive interface, improving the overall usability of the system.

ML plays a critical role in processing large volumes of sensor data in real time, allowing the identification of behavioral patterns, event prediction, and rapid decision making. In agricultural applications, such as crop management, this capability can make the difference between achieving high-quality yields and suffering poor production, or even complete crop failure. Furthermore, ML’s predictive capabilities are invaluable for predictive maintenance within production processes, helping to optimize efficiency and reduce the risk of costly process interruptions.

The integration of DT further enhances the feasibility of implementing CPSs in production environments. DT allows for the simulation and testing of a CPS under various scenarios, providing a clearer and more comprehensive understanding of the system’s viability before actual deployment. This approach significantly reduces risks and enhances the precision of the implementation process.

It is essential to highlight the enormous potential that CPSs offer in terms of sustainability, both for the present and the future. In light of the overpopulation crisis and the impending food crisis, resource management, especially the development of sustainable agricultural processes, has become not only a necessity but also a research priority.

Water is a highly valuable resource that is often overlooked in the production process. The proposals of authors such as Moses [91] provide an encouraging perspective, demonstrating efficient and responsible approaches to water management within the agricultural sector. Similarly, challenges in water management extend beyond agriculture, particularly in urban contexts where the focus is not only on consumption but also on wastewater treatment. Radini [38] addresses these issues, highlighting the current state and potential solutions to urban water challenges.

Another critical yet underutilized resource in agriculture is energy. This inefficiency is often due to the lack of attention paid to quantifying and managing energy use throughout the production process. By incorporating CPSs, data management capabilities enable the integration of energy measurement software, as evidenced by Puangpontip’s research [36]. Effective energy management not only optimizes resource use but also contributes to reducing carbon emissions, thus mitigating greenhouse gas production and combating global warming, as noted by Inderwildi [39].

In general, the integration of CPS with AI, ML, and DT in smart agriculture presents substantial benefits, equipping farmers with advanced tools and insights to effectively address the complexities of modern agricultural practices. These technologies improve decision making, optimize resource management, and increase productivity, making them a promising and still nascent field of research with immense potential for future development.

7. Conclusions

This study presents a systematic review and bibliometric analysis focused on the integration of artificial intelligence, machine learning, and digital twin technologies in the development of cyber-physical systems for smart agriculture. The findings indicate that these technologies play a fundamental role in improving agricultural efficiency, optimizing resource utilization, and improving crop resilience to environmental stress factors. Their ability to enable predictive analysis, real-time monitoring, and autonomous decision making underscores their importance for the advancement of smart agriculture.

Despite the evident benefits, the implementation of CPSs in smart agriculture presents several challenges. The technical complexities remain a major barrier, as the integration of AI, ML, and DT requires advanced computational models, reliable sensor networks, robust communication infrastructures, and skilled personnel for operation. In addition, the high costs associated with hardware, software, and maintenance limit accessibility, particularly for small- and medium-scale farmers. Standardization and interoperability between the physical and virtual components of a CPS remain critical areas requiring further research and development to ensure seamless integration into agricultural applications.

One of the most relevant applications of the CPS in smart agriculture is the control of agricultural ecosystems, particularly greenhouses. Greenhouses provide a highly controlled environment in which parameters such as temperature, humidity, solar radiation, $C{O}_2$ concentration, and irrigation can be precisely regulated, resulting in increased crop yields and improved resource efficiency. The implementation of CPSs in greenhouses enables real-time monitoring and adaptive control of environmental conditions, optimizing water and energy consumption, while improving productivity. Furthermore, the integration of DT technology allows for advanced simulation and predictive modeling of greenhouse operations, facilitating optimization of cultivation strategies before real-world implementation. These advances position greenhouses as key components in the transition toward sustainable and resilient agricultural systems.

Although this study does not discuss cybersecurity threats, it acknowledges that concerns such as data protection, system vulnerabilities, and potential cyber attacks pose risks to the large-scale adoption of CPS. Ensuring secure data transmission, safeguarding sensitive agricultural information, and implementing robust cybersecurity protocols will be crucial to fostering trust in these systems. Addressing these issues will be essential to ensure the long-term viability of CPS-based agricultural systems.

In general, this research contributes to a more comprehensive understanding of CPS applications in smart agriculture, highlighting both their transformative potential and the challenges that must be addressed to facilitate greater implementation. Identifying key technological trends and challenges provides valuable information for researchers and stakeholders looking to advance the development of Industry 4.0 in the agricultural sector. Future studies should explore strategies to improve scalability, affordability, and security in CPS, facilitating their wider adoption and maximizing their impact on global food production and sustainability.