Optimization of Vegetable Production in Hydroculture Environments Using Artificial Intelligence: A Literature Review

Abstract

This review analyzes the role of artificial intelligence (AI) and automation in optimizing vegetable production within hydroculture systems. Methods: Following the PRISMA methodology, this study examines research on IoT-based monitoring and AI techniques, particularly Deep Neural Networks (DNNs), K-Nearest Neighbors (KNNs), Fuzzy Logic (FL), Convolutional Neural Networks (CNNs), and Decision Trees (DTs). Additionally, Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) models were analyzed due to their effectiveness in processing temporal data and improving predictive capabilities in nutrient optimization. These models have demonstrated high precision in managing key parameters such as pH, temperature, electrical conductivity, and nutrient dosing to enhance crop growth. The selection criteria focused on peer-reviewed studies from 2020 to 2024, emphasizing automation, efficiency, sustainability, and real-time monitoring. After filtering out duplicates and non-relevant papers, 72 studies from the IEEE, SCOPUS, MDPI, and Google Scholar databases were analyzed, focusing on the applicability of AI in optimizing vegetable production. Results: Among the AI models evaluated, Deep Neural Networks (DNNs) achieved 97.5% accuracy in crop growth predictions, while Fuzzy Logic (FL) demonstrated a 3% error rate in nutrient solution adjustments, ensuring reliable real-time decision-making. CNNs were the most effective for disease and pest detection, reaching a precision rate of 99.02%, contributing to reduced pesticide use and improved plant health. Random Forest (RF) and Support Vector Machines (SVMs) demonstrated up to 97.5% accuracy in optimizing water consumption and irrigation efficiency, promoting sustainable resource management. Additionally, LSTM and RNN models improved long-term predictions for nutrient absorption, optimizing hydroponic system control. Hybrid AI models integrating machine learning and deep learning techniques showed promise for enhancing system automation. Conclusion: AI-driven optimization in hydroculture improves nutrient management, water efficiency, and plant health monitoring, leading to higher yields and sustainability. Despite its benefits, challenges such as data availability, model standardization, and implementation costs persist. Future research should focus on enhancing model accessibility, interoperability, and real-world validation to expand AI adoption in smart agriculture. Furthermore, the integration of LSTM and RNN should be further explored to enhance real-time adaptability and improve the resilience of predictive models in hydroponic environments.

1. Introduction

Currently, the global population is growing at a rate three times higher than the figures recorded in the mid-twentieth century, and estimates project that by the year 2050 [1,2], it will reach 9.6 billion inhabitants [3,4]. This population growth, along with rapid industrialization, has significantly reduced the amount of arable land, raising serious concerns about the sustainability of agriculture and the ability to meet the increasing demand for food [5,6]. National Food Security ensures that each person has continuous access to an adequate supply of food, with high levels of nutrients essential for an active and healthy life [7]. However, childhood malnutrition and micronutrient deficiencies remain significant issues, especially among children under 5 years old, who suffer from anemia and vitamin A deficiency [8,9]. The agricultural industry has undergone a significant transformation as a result of the introduction of a number of cutting-edge technologies brought about by Agriculture 4.0, including automation using robotics and nanotechnology, improved production with synthetic proteins, cellular agriculture and genetic engineering, as well as applying the benefits of artificial intelligence and the blockchain to financial issues [10]. These advances have driven technological progress in agriculture by eliminating dependence on soil in crop production, instead allowing direct delivery and transport of nutrients through water. This facilitates precise control over the growth and development of plants, while ensuring optimal conditions without problems [11,12,13].

Hydroculture presents itself as a viable and advanced alternative to addressing these challenges. It includes three types of systems, known as aquaponics, aeroponics, and hydroponics, with the latter being the most widely used due to its efficiency in overcoming multiple challenges. Hydroponics incorporates various techniques such as the Nutrient Film Technique (NFT), Deep Water Culture, and Wick and Drip, all recognized as optimal solutions for achieving precise control over the nutritional environment [14,15,16]. The NFT, widely used around the world for the efficient absorption of nutrients in a controlled environment, offers several advantages, including significant savings in nutrient solution, reduced harvesting time, and space optimization through vertical farming techniques in confined areas [17,18]. This agricultural technique has proven to be viable and effective for production in rooftops, terraces, and urban residences [19,20,21,22]. Additionally, these soilless cultivation systems have managed to reduce the accumulation of heavy metals in crops like rosemary by up to 71% [23]. Therefore, continuous monitoring of the various parameters of these systems is essential, leading to the implementation of IoT sensors that are remotely controlled through smart devices [24]. It is important to highlight that one of the key elements in the automation of hydroculture systems is the accuracy of these sensor readings, which allows for precisely applying nutrient doses [25]. However, one of the main challenges in nutrient dosing is the ion balance, which plants absorb at varying rates and needs periodic adjustment. Ionic interference-induced data distortions can be corrected using Machine Learning (ML), and Deep Learning (DL) approaches can be used to improve the accuracy of data realignment [16,26]. Artificial intelligence, through advanced machine learning and deep learning techniques, has proven highly effective in estimating NPK macronutrient levels (nitrogen, phosphorus, and potassium, respectively) present in water, even within aquaponic systems [27]. The application of these technologies can enhance productivity without compromising crop quality, as previously mentioned. In this context, our systematic literature review aims to address this gap in optimizing vegetable production in hydroculture systems using intelligent dosing devices. In addition to IoT-based automation, artificial intelligence (AI) has significantly improved the efficiency of hydroculture system management by optimizing the real-time regulation of parameters such as pH, total dissolved solids concentration, temperature, and electrical conductivity, among others; some machine learning and deep learning models have demonstrated improvements in the accuracy of crop growth prediction, nutrient dosing, and light intensity management, thereby contributing to the sustainability of systems [28,29], resulting in improved production, reduced costs, and minimized resource waste [30].

This review is structured around the following research questions. RQ1: What artificial intelligence models are most effective in optimizing vegetable production in hydroculture systems? RQ2: What are the main benefits of using artificial intelligence models in vegetable production? To explore these questions, studies published between 2020 and 2024 were analyzed and synthesized. The primary databases consulted were SCOPUS, IEEE, MDPI, and Google Scholar, with SCOPUS serving as the main source of research. This process involved targeted keywords and specific criteria, utilizing the PICOC (Population, Intervention, Comparison, Outcomes, and Context) and PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) frameworks to select pertinent articles on dosing devices, vegetable cultivation, nutrient solutions, and artificial intelligence applications. This approach offers valuable insights and guidance for future research and practices in nutrient solution management, particularly within NFT, emphasizing resource control to enhance efficiency and sustainability.

2. Theoretical Background

This section provides a theoretical foundation that briefly explains key concepts and terms used to enhance the understanding of hydroculture systems and their optimization in vegetable production.

2.1. Hydroculture

Hydroculture refers to a cultivation method where plants grow without soil, obtaining essential nutrients through a controlled aqueous solution. This system enhances plant growth efficiency, reduces water usage, and minimizes environmental impact. Various hydroculture techniques (see Figure 1), including hydroponics, aeroponics, and aquaponics, are designed to optimize nutrient absorption and environmental conditions, significantly improving agricultural productivity. The following sections will explore the fundamental aspects of hydroculture, including the role of precision agriculture technologies, sensors, high-intensity lighting [31], and nutrient solutions to optimize plant growth.

2.1.1. Aeroponics

A finely atomized mist of fertilizer solution is sprayed onto the roots of the crop suspended in the air; this soilless growing method greatly maximizes the use of water and agricultural resources while promoting efficient nutrient uptake [32].

2.1.2. Aquaponics

Aquaponics is a sustainable and soil-free production system wherein fish farming and plant cultivation coexist in a closed environment, creating an efficient ecological balance. In this process, fish waste is converted into essential nutrients for plants through the action of nitrifying bacteria, while plants filter and purify the water, returning it in optimal condition to the fish habitat [33]. In this way, aquaponics integrates fish farming and plant cultivation into a symbiotic cycle that minimizes waste and optimizes resource use, promoting a sustainable and highly efficient model [34].

2.1.3. Hydroponics

Hydroponics is a method of cultivation that uses water containing nutrients [25,26], as opposed to traditional cultivation using soil [35], being a response to concerns such as soil erosion, climate change, soil infertility, and poor access to water resources [36]. There are three groups of hydroponic techniques, stationary, recirculating, and aeroponics, each suited for different types of crops, which are described below.

Deep Water Cultivation (DWC)

This technique uses rectangular tanks filled with water, covered by a sheet of technopor that keeps the roots in contact with the liquid and provides them with daily oxygen, either manually or through air pumps, promoting efficient plant growth and making it suitable for both small and large-scale production systems [35,37]; the electrical conductivity of the nutritional solution, oxygenation, salinity, pH, light intensity, temperature, photoperiod, and ambient humidity are important environmental parameters to consider [38,39].

Drip System (DS)

This technique enables efficient and economical water use by distributing it evenly at regular intervals without stressing the plants; the system uses less water by releasing only the necessary amount of water and nutrients at a slow rate, while any excess is returned to the container [37,38].

Wick System (WS)

This is a technique that retains nutrients in the reservoir and transports them into the system through a wick or thread via capillary movement, using growing mediums such as coconut fiber, vermiculite, perlite, and pro-mix, among others [37].

Flow and Reflux

Also referred to as flooding and draining, this system maintains a nutrient-rich solution at specific intervals, using a medium that allows for slow water flow, making it suitable for plants in arid regions as it easily accommodates adjustments to the watering cycle [37,38].

Dutch Cube System (DCS)

This is a solution for monitoring hydroponic crops in limited urban spaces using wireless sensors to record pH, EC, temperature, and humidity. It enables remote nutrient management, optimizing real-time control and minimizing risks in plant growth [25].

Nutritive Film Technique (NFT)

This is a closed system that recirculates water with nutrients continuously for 24 h through PVC pipe channels, where plants are placed in lids with openings that allow their roots to be immersed in the flow of nutrient solution, allowing constant and healthy growth of plants [25,40]; being supplied again circularly [17,35,41]. There are hybrid systems based on the nutrient film technique, as detailed below.

• IoT-based Nutrient Film Technique (I-NFT)

The primary feature of an irrigation system, which is a hybrid of the NFT and RF, is the inclination or slope between the pipes that must be between 5% and 15% for the water to flow by gravity [35].

2.

NFT Aquaponics

This technique requires constant observation and regulation of temperature, pH, ammonia levels, and nutrients to preserve the environment [42].

2.2. Sensors Used in Precision Farming

In modern agriculture, optimizing plant production in hydroculture systems requires advanced technologies that enhance environmental control and resource efficiency. One key component is precision farming sensors, which enable real-time data collection on critical parameters such as soil moisture, temperature, pH, electrical conductivity, and water content. These sensors provide essential insights for decision-making in agricultural management [43,44]. The use of pH and electrical conductivity sensors connected to the IoT enables real-time monitoring and automated nutrient dosing. For example, in the analyzed studies, the integration of sensors with AI reduced the error in nutrient solution adjustment to just 3% using Fuzzy Logic [45]. This system minimizes human intervention and optimizes nutrient absorption, improving agricultural productivity.

Although widely applied in industrial crops like wheat and maize, the use of these sensors in high-value crops such as saffron remains under-researched, with a need for further standardization. Their integration with IoT technologies enhances production efficiency and resource management, reducing costs and optimizing input usage [46,47]. This technological synergy plays a crucial role in hydroculture by ensuring precise control over nutritive solutions, light intensity, and environmental conditions, ultimately improving plant health and yield.

The following sections will explore how these components—hydroculture techniques, precision farming sensors, and nutritive solutions—interconnect to optimize plant production, highlighting their role in the broader framework of sustainable and efficient agriculture.

2.3. Application of Temporal Learning Models and Kalman Filters

Hydroculture has evolved with the integration of advanced monitoring and control technologies based on artificial intelligence (AI). Among the most promising approaches, temporal learning models such as Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM), and Kalman Filters have demonstrated a significant impact on optimizing soilless crop growth. These models enable the analysis of key variables over time, including pH, electrical conductivity, humidity, and nutrient concentration, facilitating real-time decision-making and improving efficiency, particularly in hydroponic systems.

A key example of this application is the study by Tipwong et al. [48], who developed a soluble nutrient detection system based on RNN. Their implementation achieved 98% accuracy in detecting and adjusting nitrogen, phosphorus, and potassium levels, optimizing fertilizer supply and reducing resource waste. In a complementary study, Venkatraman et al. [49] demonstrated that using RNNs and LSTM for lettuce cultivation under the Nutrient Film Technique (NFT) improves nutrient absorption predictions and enables system adjustments for optimal growth. Similarly, Thakur et al. [50] explored the use of LSTM and RNNs for the automated control of a hydroponic spinach system, achieving significant optimization in regulating environmental parameters. Additionally, Baek et al. [51] applied LSTM models to energy consumption management in hydroponics, enhancing the efficiency of water and ventilation control.

On the other hand, Kalman Filters have been successfully implemented in various hydroponic systems to improve the accuracy of monitoring and predicting key variables. Winursito et al. [52] applied this technique in IoT sensors used for precision agriculture, achieving a 66.49% reduction in data variability and providing more reliable measurements for decision-making. Similarly, Shadrin et al. [53] employed an Extended Kalman Filter (EKF) to assess plant growth dynamics in hydroponics, enabling precise real-time adjustments to optimize crop development. Putra Kunsina et al. [54] proposed an optimized version of the Kalman Filter using Genetic Algorithms, enhancing the robustness of the model in calibrating IoT sensors. Gudepu et al. [55] integrated Kalman Filters with machine learning algorithms to improve climate prediction and optimize hydroponic crop management. Finally, Lubis et al. [56] explored the application of Kalman Filters within a Digital Twin model for precision agriculture, facilitating the monitoring and control of environmental conditions in controlled hydroponic systems.

In conclusion, the incorporation of temporal learning models in hydroponics has led to significant improvements in predicting and controlling critical variables for crop growth. The combination of RNNs, LSTM, and Kalman Filters not only enables early anomaly detection but also optimizes resource usage and enhances agricultural production efficiency. These advancements represent a crucial step toward the automation and sustainability of hydroponic systems, ensuring more efficient and resilient production in the face of current environmental and economic challenges.

2.4. Nutritive Solution: Essential Components for Plant Growth

In hydroculture systems, ensuring optimal plant growth requires precise nutrient management [11]. One essential component is the liquid fertilizer solution, formulated with specific compositions to enhance crop growth and prevent nutrient deficiencies [57,58]. Each plant requires a carefully balanced supply of nutrients, as they rely on approximately thirteen essential minerals, each in different amounts [17,38,59].

These nutrients are classified into macronutrients and micronutrients, both of which play crucial roles in plant development. Macronutrients, such as nitrogen, phosphorus, and potassium, are needed in larger quantities to support fundamental growth processes, while micronutrients, including iron, zinc, and manganese, are required in smaller amounts but are equally vital (see

Table 1. Macronutrients and micronutrients in hydroponics: keys to optimal plant growth.

Macronutrients:
Plants require large quantities of essential nutrients for optimal growth and development [60]. These nutrients are divided into primary nutrients (nitrogen, phosphorus, and potassium) and secondary nutrients (calcium, magnesium, and sulfur) [61]. These elements play key roles in protein formation, DNA synthesis, energy production, and plant structural strengthening.
Nutrient	Function
Nitrogen (N)	Promotes vegetative growth and is essential for the formation of chlorophyll and proteins.
Phosphorus (P)	Stimulates root and flower development, essential for photosynthesis and energy transfer.
Potassium (K)	Regulates the opening and closing of stomata, improves resistance to water stress, and strengthens stems.
Calcium (Ca)	Strengthens cell structure and enhances the absorption of other nutrients.
Magnesium (Mg)	Participates in enzyme activation and chlorophyll synthesis.
Sulfur (S)	Essential for protein production and chlorophyll formation.
Micronutrients:
Plants need essential nutrients in very small quantities, yet they are crucial for their metabolism and physiological functions [62]. They include iron, zinc, copper, manganese, boron, molybdenum, and chlorine [63]. Despite their low concentration in plant tissues, these elements play a role in enzymatic processes, photosynthesis, hormone regulation, and the absorption of other nutrients.
Nutrient	Function
Iron (Fe)	Essential for chlorophyll synthesis and enzymatic processes and promotes photosynthesis.
Manganese (Mn)	Involved in enzyme activation and photosynthetic processes.
Zinc (Zn)	Contributes to root and leaf growth and development.
Copper (Cu)	Plays a role in carbohydrate and protein metabolism.
Boron (B)	Regulates sugar transport and cell wall formation.
Molybdenum (Mo)	Key in nitrate reduction and amino acid synthesis.
Chlorine (Cl)	Participates in osmotic balance and photosynthesis.

).

The effectiveness of liquid fertilizers in hydroculture depends on their integration with precision farming sensors and IoT technologies, which monitor environmental conditions and adjust nutrient delivery in real time. This synergy ensures that crops receive optimal nutrition while minimizing resource waste, reinforcing the efficiency of hydroculture techniques. The following sections will explore these interconnections, highlighting how they collectively contributes to the optimization of plant production in controlled environments.

There are various nutrient solutions designed for this purpose, with AB Mix being the most used due to its balance and effectiveness.

2.5. AB Mix

The type of fertilizer used in hydroculture environments is packaged in two parts, A and B [64,65]. Providing a proper balance of macronutrients (nitrogen, phosphorus, potassium, calcium, magnesium, and sulfur) and micronutrients (iron, manganese, zinc, copper, boron, molybdenum).

Similarly, it is essential to consider water and environmental parameters, as they are crucial for maintaining optimal growth conditions, ensuring that plants receive the right balance of water, nutrients, and environmental factors for healthy development.

2.6. Water and Environmental Parameters

In hydroculture systems, optimizing plant growth requires precise control of water and environmental parameters.

Table 2. Parameters in Hydroculture Systems.

Parameter	Description
Electrical Conductivity (EC)	Measures the ability of water to conduct electricity, indicating the concentration of dissolved salts and nutrients. Optimal levels vary per plant species [37,40].
Total Dissolved Solids (TDSs)	Represent the total amount of organic and inorganic substances dissolved in water, crucial for maintaining nutrient balance [40,66].
Dissolved Oxygen (DO)	Determines the oxygen availability in water, essential for plant roots, beneficial microorganisms, and aquatic systems [32].
Potential of Hydrogen (pH)	Defines the acidity or alkalinity of the nutrient solution, affecting nutrient solubility and absorption. The ideal pH for hydroculture is around 6.0 [67,68,69,70].
Environmental Temperature and Humidity	Key factors in crop growth, affecting transpiration, water uptake, and nutrient absorption. Monitored using IoT technologies for optimal control [71].
Water Temperature	Influences nutrient solubility, microbial activity, and plant metabolism, playing a critical role in overall system efficiency [72].

shows key factors such as pH, temperature, electrical conductivity, and dissolved oxygen levels, which directly influence nutrient absorption and crop development [35].

Smart monitoring with IoT and microcontrollers such as Raspberry Pi and ESP32 has been proven to improve efficiency in hydroponics. In an optimization experiment on NFT (Nutrient Film Technique) systems, IoT-controlled temperature and humidity sensors enabled automatic irrigation time adjustments, reducing water consumption by 30% and enhancing nutrient absorption in lettuce crops [18].

Real-time monitoring of these parameters, enabled by sensors and IoT technologies, ensures a stable environment, improving efficiency and productivity. Their integration with liquid fertilizer solutions optimizes nutrient delivery, reinforcing the effectiveness of hydroculture techniques. The following section outlines the most critical parameters and their role in enhancing crop performance.

To facilitate the understanding of the following subsections,

Table 3. Summary of AI, IoT, and big data applications in hydroculture.

Section	Main Focus	Technologies Used	Key Results
Section 2.6 Application of Artificial Intelligence in the Optimization of Parameters in Hydroculture	Application of AI in hydroculture to optimize crop growth	Machine Learning (DNN, RF), IoT Sensors	DNN achieved 97.5% accuracy in predicting plant growth, reducing fertilizer waste by 15%.
Section 2.7 Smart Sensor Monitoring with AI-powered Insights	Use of IoT sensors and AI for environmental monitoring in hydroculture	IoT Sensors (Raspberry Pi, ESP32), Machine Learning (SVM, k-NN)	50% reduction in temperature and humidity variability in tomato crops, increasing yield by 25%.
Section 2.8 Optimization of Nutrient Solutions, Energy, and Water in Hydroculture Powered by Machine Learning and Deep Learning Algorithms	Efficient management of nutrients, water, and energy using AI	ML/DL Algorithms (ANN, SVM, FL), Sensors	Nutrient absorption optimized by 15–25% and water consumption reduced by 30%.
Section 2.9 Autonomous Control Systems in Hydroculture Powered by Machine Learning and Deep Learning Algorithms	Implementation of AI-based autonomous control	Reinforcement Learning, Neural Networks	IoT and ML systems automatically adjust pH and EC, reducing fertilizer consumption by 30%.
Section 2.10 Computer Vision for Crop Growth Assessment	AI applications for detecting pests and nutrient deficiencies in hydroponic crops	CNN Neural Networks	CNN achieved 97.44% accuracy in detecting nutrient deficiencies using RGB images.
Section 2.11 Integration of IoT, Big Data, and AI in Modern Hydroculture	Convergence of technologies for advanced crop management in hydroculture	Apache Spark, IoT, Big Data	Apache Spark and Big Data optimized monitoring and decision-making in aquaponic systems.

provides a structured summary of the key applications of AI, IoT, and big data in hydroculture. Each row presents the focus of the section, the technologies involved, and the key results achieved in previous studies.

2.7. Application of Artificial Intelligence in the Optimization of Parameters in Hydroculture

Artificial intelligence has transformed the management of hydroculture systems, specifically hydroponics, aeroponics, and aquaponics, enabling automated resource optimization and increased agricultural efficiency. However, most existing studies have been developed in hydroculture environments, suggesting greater maturity in AI implementation in this system compared to the other two. The data collected by pH, EC, and temperature sensors are processed using machine learning models such as Deep Neural Networks (DNNs). In a comparative study, DNN achieved 97.5% accuracy in predicting plant growth and enabled the automation of nutrient solution adjustments, reducing fertilizer waste by 15% [60].

This method, also known as soilless agriculture, has proven to be highly efficient by maximizing the use of resources such as water and nutrients, facilitating cultivation in controlled environments, and reducing environmental impact compared to conventional agriculture. Within hydroculture systems, hydroponics is based on the management of nutrient solutions, where plant roots are in direct contact with water and essential nutrients. Aeroponics takes this concept further by allowing roots to remain suspended in the air and be periodically misted with a fine nutrient-rich spray. Finally, aquaponics combines hydroponics with fish farming, creating an integrated ecosystem where fish waste acts as a natural fertilizer for plants, while plants help purify the water within the system.

Artificial intelligence has revolutionized the management of these three systems, with a predominant focus on hydroponics, where it has been widely applied to optimize factors such as water usage, nutrient dosing, and disease detection in crops. Through advanced Machine Learning and Deep Learning models, such as Artificial Neural Networks (ANNs), Fuzzy Logic (FL), and optimization algorithms, it has become possible to analyze large volumes of real-time data and precisely adjust critical variables such as electrical conductivity (EC), water temperature, dissolved solids concentration, and pH levels [72].

The use of IoT sensors connected to AI systems has enabled continuous monitoring and automated decision-making, particularly in hydroponics, where DNN models can predict plant nutritional needs based on historical data and current environmental conditions, adjusting the nutrient solution in real time [73]. Similarly, computer vision has proven effective in assessing crop growth, allowing for adjustments in nutrient dosing and water supply to maximize system efficiency [71].

While AI implementation has shown promising results in hydroponics, research in aquaponics and aeroponics remains in its early stages. Recent studies have demonstrated that in hydroponic systems, AI can reduce water consumption by 30–50%, minimize fertilizer use through optimized dosing, and enhance early disease detection, thereby increasing yield and sustainability [74]. In aquaponics, the integration of Apache Spark and Kafka has enabled agricultural performance optimization through advanced real-time data analysis technologies, although the number of studies in this area remains limited [75].

This automated approach significantly enhances the operational efficiency of hydroculture systems, allowing producers to manage multiple variables without constant manual intervention. However, since most analyzed studies focus on hydroponics, further research is needed in aeroponics and aquaponics to establish full AI applications in these systems. The following sections will explore in detail the specific techniques used to optimize plant production in hydroculture through AI, distinguishing applications according to the type of system.

2.8. Smart Sensor Monitoring with AI-Powered Insights

The monitoring of the hydroculture environment using artificial intelligence (AI) plays a crucial role in optimizing real-time control of key parameters such as pH, electrical conductivity (EC), temperature, humidity, total dissolved solids (TDSs), dissolved oxygen, and light intensity, ensuring an optimal environment for plant growth [73,76]. The integration of Machine Learning (ML) and Deep Learning (DL) algorithms enhances data analysis and decision-making, enabling automated monitoring systems to adjust conditions dynamically and reduce manual intervention [77,78,79].

From a Machine Learning perspective, algorithms such as Random Forest (RF), Support Vector Machines (SVMs), Decision Trees (DTs), and k-Nearest Neighbors (k-NNs) analyze sensor data patterns to predict environmental fluctuations and prevent imbalances in the nutrient solution [80,81]. Additionally, Fuzzy Logic (FL) and Expert Systems provide rule-based automation, adjusting irrigation and fertilization according to predefined thresholds, thereby improving resource efficiency and reducing waste [82,83].

In contrast, Deep Learning models, such as Deep Neural Networks (DNNs) and Convolutional Neural Networks (CNNs), process large volumes of data collected from IoT sensors to detect complex trends in hydroculture environments [84,85]. These models enhance the identification of anomalies in water quality, temperature variations, and nutrient imbalances, leading to self-regulating hydroculture systems that optimize environmental parameters with minimal human intervention [86].

The automation of hydroculture monitoring is further enhanced through IoT-integrated microcontrollers such as Raspberry Pi and ESP32, which facilitate continuous data acquisition and cloud-based real-time processing [87,88]. These systems automatically trigger corrective actions, such as adjusting nutrient concentrations, modifying light exposure, or activating ventilation and irrigation systems, ensuring stable and optimized growth conditions [89,90]. The integration of these systems with IoT and automation based on microcontrollers such as Raspberry Pi, ESP32, and wireless sensors enable efficient remote monitoring, optimizing resource use and enhancing responsiveness to environmental changes [91,92].

In a study applied to hydroponic greenhouses, the integration of IoT sensors with Machine Learning algorithms reduced temperature and humidity variability in tomato crops by 50%. This optimized photosynthesis and increased production by 25% [72]. This type of integration not only improves crop efficiency but also reduces the water footprint and promotes a more sustainable agricultural model.

2.9. Optimization of Nutrient Solutions, Energy, and Water in Hydroculture Powered by Machine Learning and Deep Learning Algorithms

The optimization of nutrient solutions, energy, and water in hydroponic, aeroponic, and aquaponic systems through Machine Learning (ML) and Deep Learning (DL) algorithms has significantly improved precision in nutrient formulation and adjustment, as well as the efficient management of resources [93,94,95,96]. These models utilize key parameters such as pH, electrical conductivity (EC), and total dissolved solids (TDSs) to enhance crop performance [97,98,99,100,101]. Approximately 65% of the analyzed studies focus on resource optimization, with 41% dedicated to nutrient solutions, 13% to energy efficiency, and 11% to water optimization, highlighting the importance of these technologies in improving the efficiency and sustainability of hydroculture.

To achieve this optimization, supervised and unsupervised learning models analyze large volumes of data collected in real time by IoT sensors. Among the most commonly used algorithms are k-Nearest Neighbors (k-NNs), Random Forest (RF), Deep Neural Network (DNN), Decision Trees (DTs), Support Vector Machines (SVMs), Linear Regression (LR), Convolutional Neural Network (CNN), Artificial Neural Network (ANN), Multiple Linear Regression (MLR), Linear Support Vector Regression (LSVM), Backpropagation Neural Network (BPNN), Classification and Regression Trees (CARTs), Partial Least Squares Regression (PLSR), Random Forest Regression (RFR), and Support Vector Regression (SVR) [45,60,102,103,104]. These models enable the identification of nutrient consumption patterns, the prediction of optimal nutrient absorption, and the adjustment of nutrient dosage based on environmental factors [105,106,107,108].

Regarding energy optimization, some of the most commonly used algorithms include Artificial Neural Network (ANN), Fuzzy Logic (FL), Deep Neural Network (DNN), and Decision Trees (DTs) [109,110,111,112,113]. These models enhance energy consumption efficiency in LED lighting systems, climate control, and circulation pumps by adjusting their usage according to the crop’s needs and minimizing resource waste [114,115,116].

Similarly, in water usage optimization, some of the most commonly used algorithms include Decision Trees (DTs), Random Forest (RF), Support Vector Machine (SVM), Support Vector Regression (SVR), eXtreme Gradient Boosting (XGBoost), Adaptive Network-based Fuzzy Inference System (ANFIS), and Deep Neural Network (DNN) [30,59,117]. These models analyze moisture levels, evapotranspiration rates, and plant water consumption, enabling precise irrigation management through techniques such as smart irrigation and water recirculation [118,119,120,121]. This contributes to the sustainability and efficiency of hydroponic, aeroponic, and aquaponic systems.

2.10. Autonomous Control Systems in Hydroculture Powered by Machine Learning and Deep Learning Algorithms

Autonomous control systems in hydroponics, aeroponics, and aquaponics, powered by machine learning and deep learning, have revolutionized agricultural production by enabling precise and efficient automation, optimizing resource utilization, and maximizing crop yields [16,40]. However, most of the studies analyzed focus on hydroponic environments, where AI applications have been widely implemented, while research in aeroponics and aquaponics remains at its earlier stages. Through the integration of reinforcement learning algorithms, Fuzzy Logic systems, supervised learning (including classification and instance-based regression), and supervised regression, these systems process large volumes of real-time data and automatically adjust key parameters such as pH, electrical conductivity (EC), temperature, and dissolved oxygen, ensuring an optimal environment for plant growth [41,64].

Most of these applications have been tested in hydroponic systems, while their implementation in aeroponics and aquaponics still requires further validation to achieve similar efficiency levels. Among the most widely used algorithms are Quadratic Regression (QR), Fuzzy Logic (FL, MAMDANI, FLC), k-Nearest Neighbors (k-NNs), Deep Neural Networks (DNNs), and Multiple Linear Regression (MLR) [73,122]. These models enable automation, significantly reducing human intervention by continuously monitoring environmental conditions and autonomously regulating nutrient solutions, optimizing water and fertilizer consumption while minimizing dosing errors. Additionally, they allow for immediate responses to environmental fluctuations, proactively adjusting conditions before deficiencies or excesses occur, thereby ensuring a stable and continuous production cycle [123,124]. Their ability to operate uninterruptedly enhances efficiency in time and resource management, allowing farmers to focus on the strategic aspects of production while simultaneously increasing sustainability and profitability [125,126].

In aquaponic systems, the integration of Apache Spark and Kafka has improved agricultural performance through real-time data processing, although more research is needed to further consolidate AI applications in this field [75].

Additionally, the integration of these models with smart sensors and cloud-based analytics platforms has made vegetable production in hydroponic environments increasingly autonomous, minimizing human intervention and enhancing productivity. While AI-driven automation in aeroponics and aquaponics presents promising opportunities, further research is required to refine and validate its application in these environments.

2.11. Computer Vision for Crop Growth Assessment

Computer vision has emerged as a key technology for evaluating crop growth in hydroculture systems, enabling automated and precise monitoring through advanced deep learning techniques. This approach leverages convolutional neural networks (CNNs) and image-processing algorithms to identify visual patterns in crops, facilitating real-time monitoring and anomaly detection without human intervention. Abbasi et al. [127] employed YOLOv5 in aquaponic systems for classifying leafy green crops, achieving an 82.13% accuracy in disease detection and growth assessment. Similarly, Xu et al. [128] applied ConvNeXt-Nano-Adjust (CNNA) in tomato cultivation, attaining 98.96% accuracy by integrating global attention mechanisms and multi-scale feature fusion to enhance the identification of color, texture, and structural damage patterns in affected leaves. On the other hand, Kamlesh & Patil [129] optimized CNNs through hierarchical mixed pooling, reaching 95.41% accuracy in detecting tomato diseases within hydroculture systems, surpassing advanced models such as EfficientNet-Lite, MobileNet V2, and ResNet 50 due to lower validation loss. Additionally, Aadhitya et al. [130] implemented ResNet9 and CNN models on the New Plant Village dataset, combining RANSAC segmentation and edge extraction, achieving 99.53% accuracy in disease identification and treatment recommendations. Farooqui et al. [131], integrated AI and IoT into a smart greenhouse using sensors and computer vision to automate monitoring and disease detection in plants, using CNN and MASK-RCNN trained on 10,000 images and achieving 91% accuracy and a 93% F1-score, thus optimizing production with minimal human intervention.

The application of computer vision in detecting chemical imbalances in hydroculture has significantly advanced through various studies, utilizing image processing and machine learning models to identify deficiencies and anomalies in plant health. Sathyavani et al. [132] used CNN models integrated with IoT devices to process 3,000 images of cilantro, tomato, pepper, and chili leaves, diagnosing deficiencies in nitrogen, phosphorus, potassium, iron, zinc, and calcium, with cloud-based processing for automated reporting and superior accuracy compared to other deep learning models. Buakum et al. [133] developed a two-stage deep learning model (parallel-VaNSAS) to detect abnormalities in Centella asiatica leaves, integrating U-Net, Mask R-CNN, and DeepLabV3++ for segmentation and ShuffleNetV2, SqueezeNetV2, and MobileNetV3 for classification, processing 14,860 images to enhance detection accuracy with decision fusion strategies such as UWA, DE, PSO, and VaNSAS. Similarly, Xu et al. [134] utilized deep convolutional neural networks (DCNNs) to classify nutrient deficiencies in rice, evaluating models like Inception-v3, ResNet-50, NasNet-Large, and DenseNet-121, where DenseNet-121 achieved 97.44% accuracy, outperforming SVM and HOG-based techniques in detecting 10 types of deficiencies based on color, texture, and damage patterns. Ahsan et al. [135], used Deep Learning with CNN, specifically VGG16 and VGG19, to determine nutrient concentration in hydroponic lettuce using RGB images, achieving an accuracy between 87.5% and 100% depending on the model and cultivar, demonstrating its effectiveness in automated crop monitoring. Cho et al. [136], used an Artificial Neural Network (ANN) and two-point normalization (TPN-ANN) to predict the concentration of NO3, K, Ca, and Mg in hydroponic solutions, improving the accuracy of ion-selective electrodes (ISEs).

The evaluation of crop growth and yield estimation through computer vision has advanced significantly, enabling greater accuracy in detecting and analyzing plant development. Zhao et al. [137] applied computer vision for the detection and counting of maize seedlings using LW-YOLOv7, an optimized version of YOLOv7, which identifies the presence, location, and density of seedlings in drone-captured images. This approach calculates the crop emergence rate, considering attributes such as shape, size, color, and contrast with the soil and surrounding vegetation. Additionally, the model integrates GhostNet, CBAM, and SIoU to enhance feature extraction and seedling localization, significantly reducing segmentation and counting errors, ensuring greater robustness in environments with shadows, lighting variations, and weed interference. The implementation of these techniques has optimized crop monitoring and yield prediction, enabling precise agronomic adjustments based on automated image analysis [138,139].

2.12. Integration of IoT, Big Data, and AI in Modern Hydroculture

The advancement of Industry 4.0 has revolutionized agriculture through emerging technologies such as the Internet of Things (IoT), big data, artificial intelligence (AI), and Digital Twins (DTs), enhancing efficiency in hydroponic, aeroponic, and aquaponic cultivation systems. The implementation of Digital Twins enables the creation of virtual replicas of agricultural systems, facilitating real-time monitoring, scenario simulation, and resource optimization [140,141,142]. The IoT has established itself as a key tool in precision agriculture, with sensors collecting data on environmental and soil parameters, which are processed through cloud-based platforms to improve decision-making [75,143]. Similarly, AI and Deep Learning (DL) have transformed agricultural management through convolutional neural networks (CNNs) and classification techniques, enabling the detection of nutrient deficiencies in hydroponic crops and the optimization of plant growth [74,144]. In this context, the use of big data and tools like Apache Spark and Apache Kafka has significantly improved data processing and analysis, allowing for the prediction of anomalies in aquaponic systems and the optimization of agricultural performance [75]. The integration of these technologies not only enhances productivity and sustainability but also minimizes human intervention, fostering a smarter and more automated approach to modern agriculture.

3. Methodology

This systematic review aims to identify methodologies for optimizing vegetable production in hydroculture systems through IoT and AI integration by following a three-phase process of planning, execution, and reporting, aligned with PICOC and PRISMA guidelines. The review process was supported by the use of Zotero (version 7.0.15) for reference management, Microsoft Excel 365 for the selection and exclusion of studies, and draw.io (online version, accessed on December 26, 2024) for the development of the PRISMA flow diagram. See Supplementary Materials for additional details.

3.1. Planning the Review

This phase involves designing a strategy to identify empirical studies that will effectively address the research questions.

Table 4. Questions and objectives of the research work.

	Questions	Objectives
RQ1:	What artificial intelligence models are most effective in optimizing vegetable production in hydroculture systems?	Identify the types of artificially intelligent models implemented in hydroculture systems that optimize vegetable production.
RQ2:	What are the main benefits of using artificial intelligence models in vegetable production?	Identify the main benefits of using artificial intelligence models in improving efficiency and sustainability in hydroculture systems for vegetable production.

below describes the research questions that serve as the focus of this systematic review.

For an overview of the research, and to be clear about the search terms, the PICOC method approach (see

Table 5. PICOC method.

Criteria	Description
Population	Academic publications on optimization of vegetable production and harvesting in vertical environments.
Intervention	Optimization with IoT and Deep Learning, Optimization IoT and CNN. Techniques for production optimization. Available datasets related to vegetable production. Validation of appropriate techniques for crops.
Comparison	Control group with traditional hydroculture system, applying the AquaCrop simulator program.
Outcome	Optimization of vegetable production while maintaining nutritional quality.
Context	Apply to retail producers and homes.

) was considered for the systematic literature review.

In

Table 6. List of keywords, synonyms, and relationships.

Keyword	Synonyms	Related to
Optimization with Machine Learning	Optimization, ML	Population
Optimization of Nutrient Solution	Nutrient Solution
Resource Optimization in Hydroculture	Efficiency, Automation, Sustainability, Optimization in Hydroculture
Optimization of Hydroculture System.	Optimization, Doser
Internet of Things with Deep Learning	IoT, DL	Intervention
Internet of Things with Machine Learning	IoT, ML
Control Group with Traditional Hydroculture System	Hydroculture	Comparison
Validation with Other Datasets	Dataset
Quality of Vegetable Nutrient Solution	Optimization	Outcomes

, the keywords, synonyms, and relationships with the criteria of the PICOC method are presented, allowing the organization and simplification of the research problem in a search query in the process of the systematic literature review.

For the search criteria, the search string, intervention criteria, exclusion criteria, and search mode were considered (see

Table 7. Search criteria.

Key	Criteria
Search string	ALL((“Doser” OR “Dispenser” OR “Nutrient Solution” OR “Optimization” OR “Efficiency” OR “Automation” OR “Sustainability” OR “Improved” OR “Increased”) AND ((“Monitoring" OR “Control”) AND “Real-Time”) AND (“Hydroculture” OR “Hydroponic” OR “Aquaponic” OR “Aeroponic” OR “Soilless”) AND (“Technique” OR “Method” OR “Model”) AND (“Machine Learning” OR “Deep Learning” OR “Artificial Intelligence”) AND (“Pest” OR “Disease”) AND (“Vegetables” OR “Lettuce” OR “Plants”) OR “Dataset”)
Intervention criteria	Academic Journals (Peer-reviewed); Filtering by Keyword: Optimization, Production, Deep Learning, Machine Learning, Artificial Intelligence and Hydroculture; Language: English and Spanish; Full Text; Publication Date: 2020–2024.
Exclusion criteria	Articles older than five years.
Search mode	Applied to equivalent words.

). The following words were considered in the search string: Optimization, Automation, Efficiency, Sustainability, Improvement, Harvesting, Collection, Production, Productive, Vegetables, Lettuce, Dispenser, Doser, Technique, Method, Model, Aquacrop, and Nutrient Solution. In the inclusion criteria were synonyms, English and Spanish language, open access and full text; in the exclusion criteria were age no older than five years; and the search mode applied to the title, abstract and keywords.

To perform the data extraction, the metadata of the information extracted from the research articles were determined (see

Table 8. Relevant fields for data extraction.

Fields	Descriptions
Reference	Provides information on titles and citations of research articles.
Publication	Indicates the different types of publication, e.g., whether the paper is presented at a conference or published in a journal.
Year	Year of publication of the article.
Dataset	Contains detailed information on the dataset used in the analysis. Lists the names of distributed public datasets or automatically generated datasets.
Main Idea	Indicates the main idea of the study.
Major Contributions	New contributions by the author in related fields.
Gaps	Mention the limitations, shortcomings and problems of the study (if any) in the research.
Methodology	List of the different methods used by researchers in their articles. For example, using different cultivation techniques, Machine Learning models or Deep Learning models, Fuzzy Logic methods, etc.
Results	Provide the results obtained from the research. For example, accuracy comparison of proposed Machine or Deep Learning models with other Machine or Deep Learning models.
Machine Learning Models	Identify the list of Machine Learning models used in the articles.
Deep Learning Models	Identify the list of Deep Learning models used in the articles.
Other Models	Relates to another model that is not a machine and uses a model based on Deep Learning.

), which allowed the assertive choice of related works.

The bibliometric and editorial databases SCOPUS, IEEE, MDPI and Google Scholar were searched for journal and conference articles. The following search phrases were employed: (“Doser” OR “Dispenser” OR “Nutrient Solution” OR “Optimization” OR “Efficiency” OR “Automation” OR “Sustainability” OR “Improved” OR “Increased”) AND ((“Monitoring” OR “Control”) AND “Real-Time”) AND (“Hydroculture” OR “Hydroponic” OR “Aquaponic” OR “Aeroponic” OR “Soilless”) AND (“Technique” OR “Method” OR “Model”) AND (“Machine Learning” OR “Deep Learning” OR “Artificial Intelligence” OR “LSTM” OR “Neural Network” OR “Kalman”) AND (“Pest” OR “Disease”) AND (“Vegetables” OR “Lettuce” OR “Plants”) OR “Dataset”. The final search was conducted in 2024 and found 619 studies. A total of 539 studies were further evaluated after 80 duplicate studies were found and removed. Unpublished research was excluded because, in the absence of a peer-review procedure, it did not merit quality evaluation. The appropriateness of each article for the study was assessed. Articles that didn’t fit the requirements for eligibility were eliminated.

3.2. Selection Criteria

The initial selection of studies was carried out based on their titles, abstracts, and keywords, with a systematic review of all articles to exclude irrelevant studies leading to the removal of 103 irrelevant studies and a resulting set of 436 studies that were then examined, during which the abstract of each study facilitated the exclusion of 307 studies classified as review articles, reports, dissertations, work-in-progress publications, or those not aligning with the objectives of the current research, resulting in a total of 129 articles, which were subsequently selected according to specific criteria including empirical studies, studies specifying the technology used in the production process, studies describing the applied algorithm, studies reporting the results obtained, and studies published in the last five years, ultimately leading to the selection of 72 scholarly articles for detailed examination and inclusion in the analysis, while the reference lists of each article were reviewed for additional studies but none were identified, with the PRISMA 2020 flow diagram detailing the study selection process presented in Figure 2.

3.3. Performing the Review

A data extraction framework was developed to systematically gather pertinent information for addressing the research questions, encompassing elements such as the title, year of publication, type of technology employed, type of algorithm utilized, results of the implemented model, country of study, dataset, key concepts, reported limitations, and principal findings or identified gaps for further investigation (see Table 8); researchers individually reviewed each article and employed content analysis to extract the relevant data.

3.4. Review Report

This step relates to the procedure for documenting and clearly and transparently presenting the review’s findings. This step is essential to ensure the review’s reproducibility and dependability so that readers may evaluate the study’s review findings’ rigor and legitimacy. The outcomes of this stage are shown in the section that follows.

4. Results

This section provides an overview of recent research focused on the identification and categorization of hydroculture systems using advanced artificial intelligence techniques, such as Machine Learning and Deep Learning, integrated with IoT and Big Data. These technologies are applied to optimize the management of nutrients, energy, and water, as well as to enable automatic parameter adjustment in production. Additionally, they enhance the early detection of diseases and pests, prevent chemical imbalances in nutrients, assess crop conditions, and allow for accurate estimation of agricultural yield. Furthermore, this section details the use of next-generation sensors designed to ensure precise measurements of environmental conditions, monitor growth parameters, and perform real-time tracking of plant health. Thanks to mass data storage, these technologies enable more efficient and precise management of hydroculture systems.

• RQ1: What artificial intelligence models are most effective in optimizing vegetable production in hydroculture systems?

This first research question seeks to identify and evaluate the most effective and widely used artificial intelligence models in hydroculture systems with various techniques, with the objective of optimizing vegetable production and maximizing efficiency and sustainability.

Table 9. Classification of AI models for nutrient optimization.

Author/Ref.	Hydroculture Sector	Model/Method	Accuracy	Dataset	Benefits
Fitriani et al. [18]	Hydroponic	CNN	pH: 95% TDS: 96.65%	Unspecified dataset	Optimization of nutrient and pH control in hydroponics through automation and real-time monitoring.
Dhal et al. [24]	Hydroponic	Linear-SVM	75%	Hydroponic Heavy Metal Monitoring	Optimization of heavy metal control in hydroponics through IoT and artificial intelligence to enhance lettuce growth.
Taha et al. [27]	Aquaponic	RF, PLSR, BPNN	RF: 96%	Generated	Optimization of nutrient content detection in aquaponic plants using machine learning and spectroscopy.
Venkatraman et al. [49]	Hydroponic	LSTM, RNN	90–97%	Generated	Maximization of growth with reduced inputs.
Winursito et al. [52]	Hydroponic	Kalman Filter	80–90%	Generated	Optimization in adjustment of the nutrient solution.
Verma et al. [72]	Hydroponic	Lasso Regression, k-NN	Unspecified	Unspecified dataset	Optimization of tomato growth in hydroponics using Machine Learning to predict nutrient uptake and improve yield.
Zhang et al. [93]	Hydroponic	RF, SVM, BP	RF: 86.32%	Unspecified dataset	Accurate and rapid diagnosis of nutrient deficiencies in hydroponic lettuce using Machine Learning, improving nutrient management.
Helmy et al. [95]	Hydroponic	MLR	97.9%	Generated	Optimization of nutrient deficiency diagnosis in hydroponic lettuce using Machine Learning, improving nutrient management.
Surantha & Vincentdo [96]	Hydroponic	ANFIS	95%	Generated	Optimization of pH and nutrient control using ANFIS (MATLAB R2021a) and IoT (Arduino IDE 1.8.19), improving accuracy and stability.
Budiman et al. [97]	Hydroponic	RFR, LR, PR	RFR: 93.3%	Generated	Optimization of bok choy and water spinach growth in hydroponics using IoT and Machine Learning, improving nutrient management.
Wang et al. [98]	Hydroponic	CNN	Unspecified	Generated	Optimization of precise electrical conductivity (EC) control in nutrient solutions using PSO-BPNN-PID, improving water and fertilizer integration in agriculture.
Choudhury et al. [99]	Hydroponic Aeroponic	DNN	Unspecified	Generated	Optimized crop detection using IoT and Deep Neural Networks, enhancing the monitoring and management of hydroponic crops.
Kollu et al. [100]	Hydroponic	MLR	99%	Fertilizer Prediction	Optimized fertilizer recommendation in precision agriculture using IoT and multivariable regression, increasing nutrient use efficiency.
Mamatha & Kavitha [101]	Hydroponic	k-NN	93%	University of Agricultural Sciences, Bangalore	Optimized crop growth management in greenhouses using Machine Learning in hydroponics and vegetable production.
Priya et al. [102]	Hydroponic	LR	95%	Generated	Optimization of holy basil cultivation in hydroponics using AI and IoT, reducing harvest times and maximizing efficiency.
Rajendiran & Rethnaraj [103]	Aeroponic	LCGM-Boost	95.86%	Generated	Optimization, monitoring, and prediction of lettuce crop yield in aeroponics using IoT and the LCGM-Boost model, improving growth precision.
Saputra et al. [45]	Hydroponic	FL	Unspecified	Generated	Intelligent nutrient control in hydroponics using Fuzzy Logic, optimizing nutrient distribution based on TDS.
Sangeetha & Periyathambi [60]	Hydroponic	Proposed Model, Mask R-CNN, CNN, DNN	Proposed Model: 97.5–99.4%	Kaggle	Automatic nutrient estimation in hydroponics according to plant growth, optimizing nutrient supply and reducing waste.
Atmaja & Surantha [104]	Hydroponic	MFL	Adjustment: 100%	Generated	Automation of the hydroponic system based on the NFT with multistage Fuzzy Logic, improving growth monitoring and control.
Vincentdo & Surantha [105]	Hydroponic	ANFIS	>95%	Generated	Precise monitoring and control of pH and nutrients in hydroponics using ANFIS and IoT, optimizing conditions for plant growth.
Sumarsono et al. [106]	Hydroponic	Cubist, LM, GLM/GLMNET, SVM, k-NN, CART, RF, GBM	Cubist: 97.7%	Generated	Optimization of precise monitoring of soil pH using IoT and Machine Learning.
Amalia et al. [107]	Hydroponic	FL	Unspecified	Generated	Optimization of automation of nutrient mixing in hydroponics with Fuzzy Logic.
Dhal et al. [108]	Aquaponic	LDA, DT, k-NN, LSVM	Unspecified	Generated	Optimization of irrigation and nutrients in aquaponics using Machine Learning.

Source: Own elaboration based on data retrieved from Scopus, IEEE, MDPI, and Google Scholar. The search was conducted using the keywords listed in Table 4, in the search string row, covering the period from 2020 to 2024.

presents a comparative analysis of artificial intelligence models applied to nutrient optimization in hydroculture systems, highlighting their effectiveness in vegetable production. Convolutional Neural Networks (CNNs), used by Fitriani et al. [18], achieved 95% accuracy in pH and 96.65% in total dissolved solids (TDSs), while k-Nearest Neighbors (k-NNs), Decision Trees (DTs), Multiple Linear Regression (MLR), Linear Regression (LR), and Random Forest (RF), applied by Ibarra et al. [35], reached 97% accuracy.

Similarly, Partial Least Squares Regression (PLSR), Backpropagation Neural Network (BPNN), and Random Forest (RF), employed by Taha et al. [27], obtained 96% accuracy, whereas Fuzzy Logic (FL), used by Musa et al. [61], is presented as a reliable alternative but without reported accuracy. In terms of effectiveness, Deep Neural Networks (DNNs), k-NN, and Random Forest (RF) have proven to be the most precise models for nutrient optimization, enabling optimal adjustments based on key parameters such as pH, electrical conductivity, and temperature, significantly improving production in hydroculture systems with minimal human intervention. Venkatraman et al. [49] employed LSTM and RNN models to maximize nutrient absorption in hydroponics, achieving greater precision in system regulation for lettuce growth under the NFT. Complementarily, Winursito et al. [52] applied a Kalman Filter to IoT sensors within a hydroponic system, enabling more accurate measurements for real-time decision-making. These advancements highlight the importance of integrating artificial intelligence models in optimizing the management of hydroponic crops.

Table 10. Details of nutrient optimization types.

Author/Ref.	Optimization Type	Applied Method
Uptake Optimization
Fitriani et al. [18]	pH optimization for nutrient uptake.	CNN-based precise control in NFT.
Dhal et al. [24]	Heavy metal uptake optimization.	IoT and ML for real-time monitoring of Fe, Cu, Zn.
Taha et al. [27]	NPK uptake efficiency.	Spectroscopy and ML to estimate NPK in aquaponic plants.
Saputra et al. [45]	TDS and EC optimization.	Fuzzy Logic for nutrient adjustment.
Winursito et al. [52]	Optimization in adjustment of the nutrient solution.	Kalman filtering enables the adjustment of nutrient solutions in hydroponic environments.
Zhang et al. [93]	Nutrient deficiency detection.	Random Forest for classifying deficiencies in lettuce.
Helmy et al. [95]	Nutrient concentration control.	Linear regression for concentration adjustment in NFT.
Surantha and Vincentdo [96]	Optimization of pH and nutrients.	ANFIS for intelligent control in NFT
Wang et al. [98]	Electrical conductivity optimization.	PSO-BPNN-PID for nutrient solution adjustment.
Sumarsono et al. [106]	Soil pH optimization.	IoT and ML for pH analysis in precision agriculture.
Efficiency in Different Crops
Venkatraman et al. [49]	Maximization of growth with reduced inputs.	AI can predict growth rate.
Verma et al. [72]	Growth efficiency in tomatoes.	ML is used to predict growth rate in tomatoes.
Mamatha and Kavitha [101]	Greenhouse crop efficiency.	ML and IoT sensors for hydroponic monitoring.
Priya et al. [102]	Nutrient supply efficiency in basil.	IoT and ML for real-time nutrient adjustment.
Rajendiran and Rethnaraj [103]	Aeroponic crop growth efficiency.	IoT and LCGM-Boost for growth prediction.
Atmaja and Surantha [104]	Automated irrigation and nutrient efficiency.	Multi-step Fuzzy Logic for dynamic adjustment.
Vincentdo and Surantha [105]	Precision nutrient management in NFT.	ANFIS and IoT for nutrient adjustment.
Dhal et al. [108]	Optimized aquaponic nutrient supply.	ML and reinforced error estimation for nutrient adjustment.
Fertilizer Cost Optimization
Sangeetha and Periyathambi [60]	Automatic nutrient estimation for cost savings.	Arduino and sensors for nutrient adjustment.
Budiman et al. [97]	Nutrient efficiency based on physical parameters.	ML to correlate parameters with growth.
Choudhury et al. [99]	Optimized crop monitoring to reduce waste.	IoT and DNN for crop monitoring in NFT.
Kollu et al. [100]	Precision fertilization for cost reduction.	IoT and multilinear regression for optimal fertilization.
Amalia et al. [107]	Automated fertilization cost reduction.	Fuzzy Logic for automatic nutrient mixing.

Source: Own elaboration based on data retrieved from Scopus, IEEE, MDPI, and Google Scholar. The search was conducted using the keywords listed in Table 4, in the search string row, covering the period from 2020 to 2024.

of the document analyzes different approaches to optimizing nutrient use in hydroculture environments, categorizing them into three main groups: nutrient uptake, efficiency in different crops, and fertilizer cost optimization. Each of these categories highlights the use of technologies such as the Internet of Things (IoT), Machine Learning (ML), and artificial intelligence (AI) to improve nutrient management in crops. In nutrient uptake optimization, studies focus on enhancing nutrient assimilation by plants through the control of pH, electrical conductivity (EC), and heavy metal concentrations. AI models such as Convolutional Neural Networks (CNNs), Fuzzy Logic, and classification algorithms (such as Random Forest) adjust nutrient availability in real time, ensuring efficient absorption while preventing toxicity or deficiencies.

On the other hand, efficiency in different crops includes specific strategies to optimize nutrient supply according to the plant type. Predictive models have been developed for crops such as lettuce, tomatoes, and basil, utilizing IoT sensors and machine learning to monitor growth and automatically adjust nutrient levels. These innovations maximize the yield of each species while minimizing resource waste. Finally, in fertilizer cost optimization, studies aim to reduce agricultural input expenses without compromising productivity. To achieve this, automated nutrient estimation systems, precision fertilization models with multivariable regression, and advanced monitoring techniques are implemented to minimize excessive fertilizer use. These strategies have proven to be effective in improving the economic sustainability of hydroculture environments.

Table 11. Classification of AI models for energy optimization.

Author/Ref.	Hydroculture Sector	Model/Method	Accuracy	Dataset	Benefits
Montaño-Blacio et al. [4]	Hydroponic	ANN	89.37%	Generated	Optimization of intelligent monitoring and control of hydroponic crops using IoT and Neural Networks, improving nutrient and environmental management.
Baek et al. [51]	Aeroponic	LSTM	80–90%	Generated	Energy savings in automated hydroponic systems.
Tuan et al. [109]	Hydroponic	ANN, GPR, PLSR	ANN: 96%	Generated	Optimization of phosphate detection in hydroponic solutions using data fusion and Machine Learning, improving precision in nutrient control.
Nugroho et al. [110]	Hydroponic	DNN	81%	Generated	Optimization of predictive control in hydroponic lettuce crops using Deep Neural Networks, optimizing growth and system efficiency.
Arora et al. [111]	Hydroponic	ANN	Unspecified	Generated	Optimization of automation of nutrient dosing in hydroponics using Machine Learning, reducing waste and improving plant quality.
Adidrana et al. [112]	Hydroponic	k-NN, DNN, FL	k-NN: 91.2%	Generated	Optimization of simultaneous automation of nutrient control in hydroponics using IoT and k-NN, optimizing nutrient absorption.
Kondaka et al. [113]	Hydroponic	RFR, RP, DT	RFR: 92.87%	Generated	Optimization of smart hydroponic farming system based on Machine Learning to optimize plant cultivation through automated environmental parameter control.
Bin et al. [114]	Hydroponic	FL	Unspecified	Generated	Optimization of smart irrigation in cotton crops using Fuzzy Logic, reducing water waste and improving agricultural yield.
Khudoyberdiev et al. [115]	Hydroponic	FL	Unspecified	Generated	Optimization of energy consumption in hydroponics using Fuzzy Logic, reducing energy expenditure by 18% and improving resource efficiency.
Laktionov et al. [116]	Hydroponic	FL	Unspecified	Generated	Optimization of intelligent monitoring of gas composition and light in greenhouses using Fuzzy Logic, optimizing crop growth.

Source: Own elaboration based on data retrieved from Scopus, IEEE, MDPI, and Google Scholar. The search was conducted using the keywords listed in Table 4, in the search string row, covering the period from 2020 to 2024.

presents an overview of artificial intelligence models applied in energy optimization within hydroculture systems. The models employed include Artificial Neural Networks (ANNs), Deep Neural Networks (DNNs), Random Forest Regression (RFR), k-Nearest Neighbors (k-NNs), and Fuzzy Logic (FL), among others, with accuracies ranging from 81% to 96% depending on the methodology used. ANN-based models, such as those proposed by Montaño et al. [4] and Arora et al. [111], demonstrated significant efficiency in energy consumption management. Similarly, the study by Tuan et al. [109] integrated Gaussian Process Regression (GPR) and Partial Least Squares Regression (PLSR), achieving an accuracy of 96% in optimizing energy usage. Baek et al. [51] introduced the use of LSTM in aeroponics, achieving an accuracy range of 80 to 90%, which significantly reduced the energy consumption of automated hydroponic systems. These studies highlight how artificial intelligence contributes to the sustainability of soilless cultivation systems through optimized energy resource management.

Table 12. Classification of AI models for water optimization.

Author/Ref.	Hydroculture Sector	Model/Method	Accuracy	Dataset	Benefits
Aliar et al. [30]	Hydroponic	SVR+K-Means, MLR+K-Means, DT, SVM, GBRT	SVR+K-Means: 96%	Generated	Optimization of IoT-based smart farming using Machine Learning, improving production and reducing risks in crops.
Gudepu et al. [55]	Aquaponic	Kalman Filter, Dynamic Bayesian Networks	87-93%	Generated	Reduction of water consumption and improvement of yield.
Rahman et al. [59]	Hydroponic	RF, DT, SVM, k-NN, XGB	RF: 97.5%	Crop Recommendation Dataset	Optimization of monitoring and crop recommendation in hydroponics using AIoT, enhancing resource efficiency and maximizing yield.
De Los Santos et al. [117]	Hydroponic	DT	91%	Generated	Optimization of root development monitoring in hydroponic lettuce using Machine Learning to detect diseases.
Mohamed et al. [118]	Aeroponic	ANFIS	67%	Generated	Optimization of intelligent temperature control in aeroponics using IoT, improving nutrient absorption and lettuce growth.
Mokhtar et al. [119]	Hydroponic Aeroponic	XGB, SVR, RF, DNN	XGB: 94%	Generated	Optimization of hydroponic lettuce yield prediction using Machine Learning models, enhancing production accuracy.
Metin et al. [120]	Aquaponic	TFT	67%	Generated	Optimization of nutrient control in aquaponics using deep learning models, improving ecosystem management efficiency.
Venkatraman et al. [121]	Hydroponic Aquaponic	Unspecified	Unspecified	Generated	Optimization of water recirculation in aquaponics and hydroponics using Machine Learning, improving water quality and crop growth.

Source: Own elaboration based on data retrieved from Scopus, IEEE, MDPI, and Google Scholar. The search was conducted using the keywords listed in Table 4, in the search string row, covering the period from 2020 to 2024.

presents artificial intelligence models applied to water optimization in hydroculture systems. According to the reviewed studies, various machine learning approaches have proven effective in managing water resources efficiently for vegetable production. Aliar et al. [30] implemented SVM, MLR + K-Means, DT, SVR + K-Means, and GBRT, achieving 96% accuracy. Rahman et al. [59] used Random Forest (RF), Decision Trees (DTs), SVM, k-NN, and XGBoost, reaching 97.5% accuracy with the Crop Recommendation Dataset. De Los Santos et al. [117] applied Decision Trees (DTs) with 91% accuracy. Meanwhile, Mohamed et al. [118] employed an Adaptive Neuro-Fuzzy Inference System (ANFIS) but achieved a lower accuracy of 67%. Mokhtar et al. [119] tested SVR, XGBoost, RF, and DNN, where XGBoost reached 94% accuracy. Finally, Metin et al. [120] implemented Temporal Fusion Transformers (TFTs) with an accuracy of 67%. Gudepu et al. [55] explored the combination of Dynamic Bayesian Networks and Deep Learning Models, optimizing water recirculation and reducing waste in aquaponic systems. These studies emphasize the importance of integrating AI to maximize efficient water use and enhance the sustainability of hydroponic production.

Table 13. Classification of AI models for parameter adjustment optimization.

Author/Ref.	Hydroculture Sector	Model/Method	Accuracy	Dataset	Benefits
Agustian et al. [41]	Hydroponic	FIS Mamdani	Unspecified	Generated	Optimization of nutrient control in NFT hydroponics using Mamdani Fuzzy Logic, improving pH and TDS stability.
Thakur et al. [50]	Aeroponic	LSTM, (PSO, GA)	88–95%	Generated	Optimization and improvement of irrigation and nutrient control.
Kunsina et al. [54]	Hydroponic	Kalman Filter, Genetic Algorithms	92–98%	Generated	Noise reduction in sensors for more precise measurements.
Rahayu et al. [58]	Aquaponic	FL	Unspecified	Generated	Optimization of pH control and water recirculation in aquaponics using Mamdani Fuzzy Logic, reducing turbidity and improving water quality.
Helmy et al. [122]	Hydroponic	MLR	94.84%	Generated	Optimization of acidity control in NFT hydroponic nutrient solutions using multiple linear regression.
Casillas-Romero et al. [123]	Hydroponic	FL	Unspecified	Generated	Optimization of monitoring and pH regulation in urban hydroponic systems using Mamdani Fuzzy Logic, improving nutrient absorption.
Triantino et al. [124]	Hydroponic	FLC	Unspecified	Generated	Optimization of nutrient solution pH control in hydroponics using Fuzzy Logic, increasing stability and solution quality.
Aquino et al. [125]	Hydroponic	SVM, k-NN, NB, LDA	SVM: 100% k-NN: 100%	Generated	Optimization of healthy and chlorotic leaf detection in hydroponic lettuce using SVM.
Yang et al. [126]	Hydroponic	MLR, k-NN, SVM	MLR: Yellow Leaves: 98.33% Rotten Leaves: 97.91%	Generated	Optimization of abnormal leaf detection in hydroponic lettuce using image processing and Machine Learning.

Source: Own elaboration based on data retrieved from Scopus, IEEE, MDPI, and Google Scholar. The search was conducted using the keywords listed in Table 4, in the search string row, covering the period from 2020 to 2024.

presents various artificial intelligence models employed in parameter adjustment optimization within hydroculture systems, highlighting their effectiveness in improving crop production. Ban et al. [16] applied Quadratic Regression (RQ), achieving an accuracy of 98.3% using Yamazaki’s Nutrient Solution. Herman et al. [73] employed k-Nearest Neighbors (k-NNs), Deep Neural Networks (DNNs), and Fuzzy Logic (FL), with k-NN reaching 93.3% accuracy. Helmy et al. [122] utilized Multiple Linear Regression (MLR) and obtained an accuracy of 94.84%. Additionally, Aquino et al. [125] implemented Support Vector Machines (SVMs) and k-NN, both achieving 100% accuracy, demonstrating the models’ potential for optimizing parameter adjustments. Other contributions include those of Agustian et al. [41] and Triantino et al. [124], who explored Fuzzy Inference Systems (FISs) and Fuzzy Logic Control (FLC) [58], respectively, though without specifying accuracy values. Thakur et al. [50] explored the use of LSTM with Particle Swarm Optimization (PSO) and Genetic Algorithms (GAs) in aeroponics, achieving an accuracy of 88 to 95%, thus optimizing irrigation and nutrient absorption. Additionally, Kunsina et al. [54] implemented an optimized Kalman Filter with Genetic Algorithms in IoT sensors for hydroponics, reducing data noise and improving measurement accuracy within a range of 92 to 98%. These studies highlight the role of artificial intelligence in the dynamic adjustment of parameters to enhance productivity and efficiency in hydroponics.

Table 14. Classification of AI models for computer vision in disease and pest detection.

Author/Ref.	Hydroculture Sector	Model/Method	Accuracy	Dataset	Benefits
Shadrin et al. [53]	Aquaponic	Kalman Filter, CNN, Dynamic Bayesian Networks	90–98%	Generated	Early identification of diseases and pests.
Abbasi et al. [127]	Aquaponic	Yolov5s, Faster R-CNN	Yolov5s: 82.13%	Generated	Optimization of disease detection and management in leafy green crops in aquaponics using deep learning, improving crop health.
Xu et al. [128]	Hydroponic	ConvNeXt-Nano-Adjust (CNNA), MixNet, MobileNetV3, MobileVit, GhostNet, ShuffleNetV2	CNNA: 98.96%	PlantVillage	Optimization of pest and disease classification in tomatoes using lightweight CNNA.
Kamlesh & Patil [129]	Hydroponic	Proposed CNN, EfficientNet-Lite, MobileNetV2, ResNet-50	CNN: 95.41%	Kaggle (Tomato Leaf Disease Detection Dataset)	Optimization of early disease detection in plants using CNN and hierarchical pooling techniques, thus enhancing diagnosis accuracy.
Aadhitya et al. [130]	Hydroponic	ResNet-9	99.53%	New Plant Village Dataset	Optimization of disease diagnosis in agricultural plant leaves using ResNet and CNN, detecting diseases at early stages with high accuracy.
Farooqui et al. [131]	Hydroponic	VGG, MASK-RCNN	VGG: 93%	PlantVillage	Optimization of monitoring and control in automated greenhouses using IoT and Machine Learning, improving production and reducing manual labor.

Source: Own elaboration based on data retrieved from Scopus, IEEE, MDPI, and Google Scholar. The search was conducted using the keywords listed in Table 4, in the search string row, covering the period from 2020 to 2024.

presents various artificial intelligence models applied to disease and pest detection in hydroculture systems using computer vision techniques. The models covered include Faster R-CNN and Yolov5s, as used by Abbasi et al. [127], achieving an accuracy of 82.13%. Xu et al. [128] employed a range of models, including MixNet, MobileNetV3, MobileVit, GhostNet, ShuffleNetV2, and ConvNeXt-Nano-Adjust (CNNA), with CNNA attaining a remarkable 98.96% accuracy using the PlantVillage dataset. Kamlesh & Patil [129] implemented EfficientNet-Lite, MobileNetV2, ResNet-50, and a proposed CNN model, with the CNN achieving an accuracy of 95.41% on the Kaggle Tomato Leaf Disease Detection Dataset. Aadhitya et al. [130] used ResNet-9, which achieved 99.53% accuracy on the New Plant Village Dataset. Farooqui et al. [131] integrated AI and IoT into a smart greenhouse with sensors and computer vision, utilizing VGG and MASK-RCNN trained on around 10,000 images, obtaining an accuracy of 91% and an F1-score of 93%, demonstrating the effectiveness of these models in automating disease detection while minimizing human intervention. Additionally, Shadrin et al. [53] combined Kalman Filters, CNN, and Dynamic Bayesian Networks, achieving an accuracy of 90 to 98% in the early identification of diseases and pests in aquaponic systems. These advancements highlight the importance of using artificial intelligence techniques for automated disease detection in soilless cultivation, reducing reliance on manual inspections and improving efficiency in agricultural production.

Table 15. Classification of AI models for computer vision in chemical imbalance detection.

Author/Ref.	Hydroculture Sector	Model/Method	Accuracy	Dataset	Benefits
Sathyavani et al. [132]	Hydroponic	ResNet50, VGG16, VGG19, DenseNet121, DenseNet169, DenseNet201	ResNet50: 98.5%	Generated	Optimization of plant leaf nutrient detection using IoT and CNN, improving the accuracy of nutritional analysis.
Buakum et al. [133]	Hydroponic	Ensamble He-Meta (Proposed), U-Net, Mask R-CNN, DeepLabV3++, ShuffleNetV2, SqueezeNetV2, MobileNetV3	Proposed: 98.98%	ABL-1 ABL-2	Optimization of leaf abnormality detection in Centella asiatica using a two-stage Deep Learning model.
Xu et al. [134]	Hydroponic	DenseNet121, ResNet50, NasNet-Large, Inception-v3	DenseNet121: 97.44%	Generated	Optimization of nutrient deficiency diagnosis in rice using Deep Convolutional Neural Networks.
Ahsan et al. [135]	Hydroponic	CNN, VGG16, VGG19,	CNN: 97.9%	Generated	Optimization of nutrient concentration determination in hydroponic lettuces using Deep Learning models.

Source: Own elaboration based on data retrieved from Scopus, IEEE, MDPI, and Google Scholar. The search was conducted using the keywords listed in Table 4, in the search string row, covering the period from 2020 to 2024.

presents various AI models applied in hydroculture systems to detect chemical imbalances in plant health. Sathyavani et al. [132] employed Convolutional Neural Networks (CNNs) integrated with IoT devices to process images of plant leaves, diagnosing deficiencies in essential nutrients such as nitrogen, phosphorus, and potassium, achieving a 98.5% accuracy using ResNet50. Buakum et al. [133] implemented a deep learning model combining U-Net, Mask R-CNN, and DeepLabV3++, reaching an accuracy of 98.98% using an ensemble approach. Xu et al. [134] used DenseNet121 and ResNet50 to classify nutrient deficiencies in crops, achieving a 97.44% accuracy. Ahsan et al. [135] applied CNN models such as VGG16 and VGG19, demonstrating an accuracy of 97.9%. Finally, Cho et al. [136] used an artificial neural network (ANN) model, though the accuracy was not specified. These findings highlight the efficiency of AI-based models, particularly deep learning architectures, in accurately detecting and diagnosing chemical imbalances in hydroculture cultivation.

Table 16. Classification of AI models for computer vision in crop evaluation and yield estimation.

Author/Ref.	Hydroculture Sector	Model/Method	Accuracy	Dataset	Benefits
Tipwong et al. [48]	Hydroponic	ANN	85–95%	Unspecified	Accurate prediction of crop yield.
Zhao et al. [137]	Hydroponic	LW-YOLOv7, YOLOv7	LW-YOLOv7: 93.2%	Generated	Optimization of maize seedling detection using a lightweight model based on YOLOv7, improving recognition accuracy and speed.
Palacios et al. [138]	Hydroponic	MLP	98.93%	Generated	Optimization of quantitative production of hydroponic tomatoes using Artificial Neural Networks and digital image processing, reducing errors in productivity measurement.
Park et al. [139]	Hydroponic	YOLO V3, YOLO V2, TinyYOLO V3, TinyYOLO V2	YOLO V3: 98.27%	Generated	Optimization of monitoring and prediction of the optimal harvest timing in hydroponic strawberries using cloud AI and IoT-Edge, reducing quality loss and improving harvesting efficiency.

Source: Own elaboration based on data retrieved from Scopus, IEEE, MDPI, and Google Scholar. The search was conducted using the keywords listed in Table 4, in the search string row, covering the period from 2020 to 2024.

presents a comparative analysis of artificial intelligence models utilizing computer vision for crop evaluation and yield estimation in hydroculture systems. The table includes models such as YOLOv7, LW-YOLOv7 (Zhao et al. [137]), MLP (Palacios et al. [138]) and variations of YOLO (Park et al. [139]), with reported accuracy values ranging from 93.2% to 98.93%. These models enhance precision in estimating crop growth and productivity by leveraging advanced Deep Learning techniques. Notably, LW-YOLOv7 was optimized for maize seedling detection, demonstrating a 93.2% accuracy rate, while Palacios et al. [138]’s MLP model reached 98.93% accuracy, indicating its robustness in yield estimation. Tipwong et al. [48] incorporated a combination of Convolutional Neural Networks (CNNs) and Generative Adversarial Networks (GANs), achieving a 94.5% accuracy in improving lettuce growth evaluation in NFT systems. These studies highlight the impact of artificial intelligence on automating monitoring and optimizing productivity in hydroponics.

Table 17. Big Data, IoT and AI in smart agriculture: challenges and benefits.

Author/Ref.	Implemented Technologies	Application	Constraints/ Challenges	Findings/ Advantages	Precision
Lubis et al. [56]	Kalman Filter. Machine Learning. IoT.	Real-time data integration and automated decision-making.	The integration of IoT sensors faces interoperability issues, as the data come in different formats, making processing more difficult.	Optimizes the accuracy in measuring environmental variables, improving decision-making in hydroponics.	85–92%
Xu et al. [140]	Digital Twin. Big Data. IoT. AI.	Optimization of aquaponic systems.	Implementation and calibration costs of the model.	Improvement in monitoring efficiency and decision-making.	Unspecified
Reyes Yanes et al. [141]	Digital Twin. Big Data. IoT. AI.	Smart management of hydroponic crops.	Integration with existing monitoring systems.	Reduction in water and nutrient usage.	Unspecified
Jans-Singh et al. [142]	Digital Twin. Big Data. IoT. AI.	Urban integration of hydroponics with Digital Twins.	Quality and accuracy of collected data.	Optimization of space usage in urban agriculture.	Unspecified
Farooq et al. [143]	Big Data. IoT. AI.	Monitoring in smart greenhouses.	Data security and compatibility between IoT devices.	Higher accuracy in crop yield prediction.	Unspecified
Ezzahoui et al. [75]	Big Data. IoT. AI.	Optimization of aquaponics with Big Data.	High demand for processing and data storage.	Advanced automation and reduced human intervention.	Unspecified
Bhandari et al. [74]	Big Data. IoT. AI.	Automation and monitoring in hydroponics.	Difficulty in implementing AI in real-world environments.	Increased productivity with reduced resource waste.	15%
Abidi et al. [144]	Big Data. IoT. AI.	Classification of deficiencies in hydroponic lettuce.	Need for larger datasets to train models.	High accuracy in detecting nutritional deficiencies in crops.	96%

Source: Own elaboration based on data retrieved from Scopus, IEEE, MDPI, and Google Scholar. The search was conducted using the keywords listed in Table 4, in the search string row, covering the period from 2020 to 2024.

presents various smart agriculture technologies, outlining their applications, challenges, and benefits. Xu et al. [140] highlight the use of Digital Twins, Big Data, IoT, and AI for optimizing aquaponic systems, though they emphasize the high costs of implementation and calibration. Reyes Yanes et al. [141] discuss smart hydroponic crop management, focusing on integration challenges with existing monitoring systems but showcasing a significant reduction in water and nutrient usage. Jans-Singh et al. [142] explore the urban integration of hydroponics using Digital Twins, where data accuracy and precision pose challenges, yet spatial optimization in urban farming is a key advantage. Lubis et al. [56] utilized a combination of Kalman filtering, Machine Learning, and IoT for real-time data integration in hydroponics, achieving an accuracy level of 85 to 92% and optimizing the precision of environmental variable measurements to enhance decision-making. These studies highlight the key role of emerging technologies in transforming data-driven agriculture, ensuring more efficient and sustainable production.

Table 18. Sensors and parameters used in the studies.

Measured Parameters	Sensor	Description
pH	SEN0161 SEN00244	Measures the level of acidity or alkalinity of water using measurements based on acid–base reactions and is used to monitor and adjust the pH of water to keep it within an optimal range (typically between 5.5 and 7), which is essential for a healthy plant growth.
Ultrasonic	HC-SR04 JSN-SR04T	Measures distances using ultrasonic waves, emits an ultrasonic pulse, and calculates the distance to the object by measuring the time it takes for the echo to return, which is ideal for applications such as obstacle detection, liquid level measurement and robotics.
Water Temperature	DS18B20 DFR0024	Digital temperature sensor with a high accuracy of ±0.5 °C within a range of −10 °C to +85 °C which is used to measure water temperature in hydroculture systems; its function is crucial, as water temperature directly influences nutrient uptake by plants, helping to maintain optimal conditions for crop growth.
Air Temperature and Humidity	DHT11 DHT22	Measures ambient temperature and humidity, being a low-cost device that provides accurate data on weather conditions and is widely used in automated systems such as smart agriculture and IoT-based projects for environmental monitoring.
Barometric Pressure	BMP180	Measures atmospheric barometric pressure, which allows short-term weather forecasting and helps determine whether to continue or delay agricultural activities.
Infrared Temperature	MLX90614	Measures the temperature of the leaf to compare it with the ambient temperature, which allows automatic adjustment of the operating cycle of the electric pumps in the hydroculture system, optimizing energy consumption and ensuring suitable conditions for the crop.
Water Level/Soil Moisture	T1592 Float Switch	Detects and monitors the water level in a system; in applications such as automated hydroculture, it allows the water supply to be managed by detecting if the level is below a set value, ensuring that plants receive the right amount of water.
Intensity Luminosity	BH1750 TSL2561 LDR GL5528	Its function is to ensure that the plants receive an adequate amount of light, essential for photosynthesis and optimal growth; the predictive control system uses this light intensity data to drive the growth lamp, stabilising the light received within optimal ranges when natural light is not sufficient, such as at night or in low-light conditions.
Water Flow	YF5201 FS300A	Measures water flow in a hydroculture system using an internal vane mechanism that rotates with the flow of liquid, generating electrical pulses proportional to the volume of water passing through it; with an operating range of up to 80 mL/h, this sensor is ideal for accurately controlling the supply of water and nutrients in the system.
Water Quality	SEN0189	Measures the turbidity of the water, its clarity, or the level of suspended particles; this parameter is crucial for assessing water quality, ensuring the effectiveness of the filtration system and preventing diseases in the hydroculture system, thus contributing to the sustainability of the ecosystem.
Total Dissolved Solids	SEN0244	This element detects the electrical conductivity of the water, which is proportional to the concentration of dissolved solids, ensuring optimal levels for plant growth; this ensures precise control of water quality and improves the efficiency of hydroculture cultivation.
Electrical Conductivity	DFR0300-H	EC measurement is used to monitor the levels of nutrients dissolved in water in hydroculture systems. This ensures that plants receive adequate amounts of essential nutrients, optimizing their growth and guaranteeing productivity at each stage of development, such as germination, vegetation and flowering.
Leaf Temperature	SG-1000	Monitors leaf transpiration by measuring the temperature difference between the leaf and the environment, together with the relative humidity, allowing the detection of internal changes in the plants in real time, optimizing the control of irrigation cycles in hydroculture systems. By adjusting the cycles according to the state of the plant, energy consumption is reduced and the efficiency of the system is improved.
Rain Detection	FC-37	Monitors plant exposure to rainfall in outdoor crops. When it detects rain, it automatically adjusts watering settings based on soil sensor readings, reducing or stopping scheduled watering to avoid over-watering and ensure optimal conditions for plant growth, thus optimizing water usage and protecting crop health.
CO2 Level	Figaro’s CDM4161A TPS-2 PTM-48A CI-340 SGP30	They collect real-time data on CO2 concentration, which is crucial for regulating the process of plant photosynthesis; maintaining optimal CO2 levels improves crop growth and productivity. These IoT systems proccess information to automate ventilation and other necessary actions, thus optimizing the greenhouse environment and reducing manual intervention.
Air Quality Index	MQ-135 MQ-137	Detects the presence of harmful gases in the environment, generates real-time data that are processed by a microcontroller such as an Arduino and sent to an IoT platform for monitoring. This is essential for identifying pollution levels that could affect plant growth and reduce crop yields, it serves to alert on air quality conditions that require attention, helping to protect the growing environment and optimize productivity.
Nutrient Level, NPK (nitrogen–phosphorus–potassium)	RS485	Measures using specific ion sensors, which detect the concentrations of each element in the nutrient solution of a hydroculture system; the sensors work by measuring the difference in electrical potential between a detection electrode and a reference electrode, which is directly related to the concentration of the ions, allowing the adjustment of the nutrient solution in real time to ensure that the plants receive the essential nutrients in the right amount, thus optimizing their growth and the quality of the crop.
Dissolved Oxygen	SEN0237	Measures the amount of oxygen present in the water, essential for maintaining optimal conditions in hydro-culture systems; it operates through a signal conditioning circuit that uses a voltage regulator and a low-displacement operational amplifier to convert microvolt signals into a usable range of 0 V to 5 V. It integrates a temperature compensation module to correct for variations in the measurement, thus improving the accuracy of the data obtained.

Source: Own elaboration based on data retrieved from Scopus, IEEE, MDPI, and Google Scholar. The search was conducted using the keywords listed in Table 4, in the search string row, covering the period from 2020 to 2024.

presents a comprehensive overview of sensors utilized in smart agriculture, specifically within hydroculture systems. The table outlines the parameters measured, sensor descriptions, and their applications in optimizing crop growth. Among the key technologies highlighted are total dissolved solids (TDSs), electrical conductivity (EC), leaf temperature, CO₂ levels, and air quality sensors, which enable precise, real-time monitoring of environmental and nutritional conditions essential for plant development. Additionally, rain detection sensors are incorporated to automatically adjust irrigation systems, preventing water waste, while dissolved oxygen level sensors play a crucial role in maintaining optimal conditions for hydroculture cultivation. Integrating these advanced sensors with IoT and Big Data platforms significantly enhances monitoring efficiency and facilitates data-driven decision-making, ultimately improving agricultural productivity and sustainability.

• RQ2: What are the main benefits of using artificial intelligence models in vegetable production?

Artificial intelligence models applied to hydroculture systems optimize the use of nutrients, water, and energy, enabling automatic parameter adjustment according to the specific needs of each crop. This enhances efficiency, reduces waste, improves sustainability, and minimizes carbon footprint. Furthermore, computer vision facilitates the early detection of diseases, pests, and chemical imbalances, as well as real-time crop condition monitoring and yield estimation, allowing for proactive intervention before these factors negatively impact production. Additionally, the integration of Big Data, combined with IoT and AI technologies, has proven essential in managing large-scale agricultural data, contributing to automation, resource efficiency, and sustainability while simultaneously addressing key implementation challenges. This not only enhances crop productivity and quality but also promotes sustainability by reducing environmental impact and maximizing resource utilization in limited spaces.

In Figure 3, the transformative impact of artificial intelligence on agriculture is illustrated. Based on the analysis of 72 articles from the systematic literature review, the key benefits of AI models in crop production were identified. Among the most significant are nutrient optimization 32% and automated parameter adjustment 12%, which significantly enhance resource efficiency. Additionally, AI drives energy optimization by 14%, Big Data and IoT integration by 11%, and water optimization by 11%, thus reducing waste and improving sustainability. It also strengthens early detection of chemical imbalances by 6% and disease and pest detection by 8%, enabling faster and more effective interventions. Meanwhile, the production estimation of 6% still presents opportunities for improvement, suggesting the need for further advancements in this area. Collectively, these findings demonstrate that AI not only optimizes efficiency in agriculture but also promotes sustainability and productivity, paving the way for a smarter and more resilient agricultural future.

4.1. Location of the Studies

Figure 4 shows the distribution of studies on optimizing vegetable production in hydroculture systems using artificial intelligence, which shows a significant concentration in Asian and developing countries, where the need for innovative agricultural solutions is crucial due to issues such as food security, water resource access, and high population density.

India leads the list with 17 studies, reflecting its growing interest in smart agriculture to address challenges such as water scarcity and rapid population growth.

Indonesia, with 16 studies, faces similar issues, where optimizing hydroculture crops is key to reducing dependence on increasingly limited agricultural land due to deforestation and urbanization. China, with 11 studies, has strongly invested in automation and AI in agriculture to improve efficiency and ensure sustainable production in a country with a massive food demand. The United States and Egypt, with three studies each, show interest in crop efficiency under controlled conditions, with Egypt being a particular case due to its reliance on the Nile and the need for agricultural solutions adapted to its desert climate.

Türkiye, Korea, Thailand, and Canada, with two studies each, have explored advanced production technologies in controlled environments, focusing on sustainability and energy efficiency, particularly in urban and extreme climate settings.

Finally, countries like the United Kingdom, Croatia, Hungary, Singapore, Saudi Arabia, the Philippines, Paraguay, Pakistan, Nepal, Morocco, Mexico, Japan, Ecuador, and Bangladesh (each with one study) represent isolated but significant efforts in developing smart agricultural technologies, often driven by the need to adapt production to adverse environmental conditions, water scarcity, or the necessity of improving food security in their respective regions.

4.2. Study Publisher Details

Figure 5 shows a chart representing the distribution of analyzed studies based on the database in which they were published, reflecting the relevance and impact of each source within the scientific community. Scopus, with 35% of the studies, is the most influential database in the review, highlighting its prestige and recognition in indexing high-impact scientific research.

Scopus is known for its rigorous selection of journals and conferences, ensuring the quality and validity of the studies included in this source. IEEE, representing 42% of the studies, is fundamental in engineering and technology, indicating that a significant portion of the reviewed literature has a technological focus, particularly on the use of artificial intelligence and the Internet of Things (IoT) in hydroculture.

MDPI, with 18%, is an open-access publisher that has gained relevance in recent years by facilitating unrestricted access to scientific research, allowing for the rapid dissemination of advancements in sustainability and agricultural optimization.

Finally, Google Scholar, with only 5%, is a valuable source for finding academic literature but does not have the same indexing and quality standards as the other databases. Its lower representation suggests that most of the analyzed studies come from sources with stricter peer-review processes. Overall, this distribution demonstrates that research on optimizing vegetable production in hydroculture using artificial intelligence is a growing area of interest within the scientific community, with publications in high-impact and reputable databases.

4.3. Publication Date

Figure 6 shows the distribution of research articles by year and highlights the growing interest in optimizing vegetable production in hydroculture systems using artificial intelligence. The data indicate a notable increase in publications in 2023 and 2022, with 23 and 20, respectively, suggesting that these years marked a peak in research activity. This surge may be attributed to the acceleration of technological advancements in agriculture, the growing demand for sustainable food production, and the increasing integration of AI-driven automation in hydroculture farming.

In contrast, earlier years, such as 2020 and 2021, show a lower number of publications (9), likely due to the initial stages of research adoption in this field and the impact of global disruptions such as the COVID-19 pandemic, which may have affected research funding and project timelines. The year 2024, with 11 publications so far, suggests continued but slightly reduced momentum, possibly due to the stabilization of research trends or a shift toward practical implementation rather than theoretical exploration.

The overall pattern underscores the increasing recognition of AI as an essential tool in agricultural optimization. The peak in 2022 and 2023 suggests an intensive phase of AI research, with subsequent years likely focusing on refinement and real-world implementation. This trend highlights the importance of continued research funding and collaboration within academia and industry to ensure that AI-driven hydroculture systems remain at the forefront of sustainable agricultural innovation.

5. Discussion

The literature review has identified Deep Neural Networks (DNNs), Fuzzy Logic (FL), Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM), k-Nearest Neighbors (k-NNs), and Random Forest (RF) as the most effective artificial intelligence models for optimizing vegetable production in hydroculture systems, each with specific applications. DNNs excel in predicting crop growth and optimizing nutrient solutions, while the FL method enables dynamic adjustments based on environmental changes. CNNs have proven highly accurate in detecting diseases and nutritional deficiencies, achieving over 98% accuracy in pest identification. Models like k-NN and RF have proven highly efficient in predicting water and energy consumption, leading to optimized resource use and enhanced sustainability in hydroculture systems. Other algorithms, such as Support Vector Machines (SVMs) and XGBoost, have been applied for regulating environmental parameters and predicting nutrient demand, thereby improving irrigation efficiency and nutrient solution management. The integration of these models with IoT sensors has enabled real-time monitoring and automated control. Additionally, RNN and LSTM have been instrumental in analyzing temporal patterns of nutrient absorption, electrical conductivity, and pH fluctuations, allowing for more precise long-term adjustments in hydroponic environments. Their ability to model sequential data makes them essential for predicting future crop requirements and automating nutrient delivery over time. The findings suggest that model selection should depend on the specific objective within the hydroculture system. While DNN and FL optimize nutrient management, CNNs are crucial for disease detection, and models like RF and k-NN facilitate water and energy management. RNN and LSTM, on the other hand, enhance predictive accuracy in dynamic environmental conditions, improving the adaptability of hydroponic systems.

The application of artificial intelligence in vegetable production has brought multiple benefits, improving nutrient management, optimizing water and energy consumption, reducing human intervention, and enabling early detection of diseases and pests. A key advantage is the automation of nutrient solution adjustments. Models like Deep Neural Networks (DNNs) and Fuzzy Logic (FL) optimize nutrient use, minimizing waste and enhancing crop growth. Additionally, algorithms such as Random Forest (RF) and Support Vector Machines (SVMs) have improved irrigation efficiency, minimizing water waste and lowering energy costs in hydroculture systems. Early disease detection has been another key benefit, enabling faster identification of pests and nutrient deficiencies, thus reducing pesticide use and improving crop quality. Furthermore, integrating artificial intelligence with IoT sensors has facilitated real-time monitoring of key variables, automating decision-making, and optimizing operational efficiency. The incorporation of RNN and LSTM has further strengthened AI applications in hydroculture by enabling predictive modeling of long-term trends in crop health and environmental conditions. These models ensure more precise adjustments to irrigation schedules and nutrient levels, enhancing overall sustainability. In conclusion, artificial intelligence has transformed hydroculture production by optimizing resource use, enhancing disease detection, and reducing manual intervention. Future studies should explore refining LSTM and RNN architectures for improved efficiency in hydroponic nutrient delivery and environmental control.

6. Limitations

Despite the great potential of artificial intelligence to optimize vegetable production in hydroculture systems, there are still limitations that must be addressed for its effective implementation. One of the main barriers is the dependence on large volumes of high-quality data, as many models require detailed information for training. However, the available datasets are often limited or specific to certain environmental conditions, making generalization difficult.

Another significant challenge is the lack of standardization in AI models, as studies use different algorithms and methodologies, making it difficult to directly compare their efficiency. Additionally, implementing these technologies involves high costs in IoT sensors, specialized hardware, and connectivity, which may be inaccessible to small-scale producers or regions with limited technological resources. The high computational demand of some models, such as deep and convolutional neural networks, presents another limitation, as their training and real-time execution require advanced infrastructure. Moreover, validation of these models in real production environments is still scarce, creating uncertainty about their performance under variable conditions such as climate, water quality, and crop types.

Furthermore, the computational complexity of Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) models presents an additional limitation, as these models require substantial processing power to analyze sequential data effectively. Their dependence on large and well-structured datasets for training can also pose challenges in real-world applications, where inconsistencies in sensor data and environmental variations may affect predictive accuracy.

There is also a lack of long-term studies that evaluate the real impact of artificial intelligence on the sustainability and productivity of crops. Furthermore, integrating these models with existing hydroculture systems can be complex due to compatibility issues between software, hardware, and digital platforms. Finally, although artificial intelligence offers significant advancements in optimizing hydroculture crops, its application faces challenges in terms of technology access, interoperability, costs, and validation under real conditions. To overcome these limitations, future research should focus on developing more accessible, efficient, and adaptable models for various agricultural production environments. Additionally, improving the efficiency of RNN and LSTM architectures, reducing their computational costs, and enhancing their integration with IoT-based monitoring systems will be essential for their broader adoption in hydroponic applications.

7. Conclusions

The most effective artificial intelligence models for optimizing vegetable production in hydroculture systems vary depending on their specific application. Deep Neural Networks (DNN) have achieved 97.5% accuracy in predicting crop growth, while Fuzzy Logic (FL) has demonstrated a 3% error margin in nutrient regulation, allowing for dynamic and efficient adjustment of the nutrient solution. In disease detection and nutritional deficiency monitoring, Convolutional Neural Networks (CNN) have achieved over 98% accuracy, enabling automated identification of pests and plant diseases through image analysis. Additionally, Random Forest (RF) and Support Vector Machines (SVMs) have demonstrated 97.5% and 96% accuracy in optimizing water consumption and irrigation, promoting efficient resource use. Furthermore, models such as k-Nearest Neighbors (k-NNs) and XGBoost have reached up to 94% accuracy in energy management and environmental parameter optimization. Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) models have also emerged as valuable tools for analyzing sequential data, optimizing nutrient absorption and improving hydroponic system adaptability. Their ability to predict long-term trends allows for more efficient nutrient scheduling and enhanced environmental control. While each model offers specific advantages, there is no single ideal model, and combining different approaches could enhance hydroculture system efficiency. The development of hybrid models integrating deep learning with classification and optimization techniques presents an opportunity to improve automation and crop performance further. Despite their effectiveness in experimental settings, challenges related to large-scale implementation remain due to the need for high-quality data, technological infrastructure, and operational costs.

Artificial intelligence has improved hydroculture production by optimizing nutrient management, reducing waste, and enhancing crop growth. It has also increased water and energy efficiency through models that predict water demand and optimize irrigation. Additionally, early disease detection using neural networks has reduced the need for pesticides, improving crop quality and sustainability. Integration with IoT sensors has facilitated real-time monitoring, automated decisions, and lowered operational costs. The incorporation of RNNs and LSTM further strengthens AI applications in hydroculture by providing advanced predictive capabilities and adaptive control mechanisms. These models contribute to real-time decision-making by capturing dynamic environmental interactions, enhancing system precision, and improving sustainability. Artificial intelligence has made hydroculture production more precise, efficient, and sustainable, although its implementation still faces challenges related to accessibility and costs. Future research should focus on refining RNNs and LSTM architectures to enhance their efficiency, reduce computational requirements, and integrate them seamlessly with IoT systems for broader adoption in smart agriculture.