Next-gen agriculture: AI links sensors, analytics and robots in farming revolution

Artificial intelligence (AI) is rapidly shifting the agricultural sector from labor-intensive and experience-based farming toward data-driven systems that can detect crop stress, forecast yields, guide irrigation and operate farm machinery with limited human intervention, according to a new review published in Applied System Innovation.

The study, titled "Agricultural Intelligence: A Technical Review Within the Perception–Decision–Execution Framework", maps how AI is being applied across three linked stages of smart agriculture: perception, decision making and execution.

The authors state that AI is becoming the foundation of a closed-loop farming system in which sensors and drones collect field data, algorithms turn that data into management decisions, and robots or autonomous machines carry out precise field operations. This shift could help address mounting pressure on global food systems from population growth, climate change, resource scarcity and the environmental costs of conventional agriculture.

AI-powered perception is transforming crop monitoring

The first major shift identified in the review is the rise of AI-enabled perception systems. These technologies allow farms to collect and interpret large volumes of data on crop health, soil status, pest outbreaks, water stress and nutrient deficiencies. The study highlights unmanned aerial vehicles (UAVs), satellite imagery, sensor networks, RGB cameras, multispectral imaging, hyperspectral imaging, thermal infrared sensors and LiDAR as core tools in this transformation.

Deep learning has become especially important in crop disease and pest detection. Earlier systems depended heavily on controlled datasets and laboratory images, but more recent research has moved toward field-based models that can work under complex lighting, background interference and crop occlusion. The review notes that models such as convolutional neural networks, YOLO-based detectors, transformers and generative adversarial networks have improved disease classification, pest detection and lesion segmentation.

In one area, pest and disease detection, the review shows how AI is shifting agriculture from manual inspection to rapid, image-based diagnosis. UAVs equipped with RGB, multispectral, hyperspectral and thermal sensors can monitor fields and orchards at scale, while AI models identify disease symptoms or pest presence. Some systems reported high accuracy in detecting early infections, while lightweight models are being developed for real-time use on mobile or edge devices.

The authors also point to strong progress in crop growth monitoring. UAV-based systems can estimate plant height, canopy coverage, biomass, leaf area index, chlorophyll content and canopy temperature. These measures allow farmers to track crop development without destructive sampling. The review shows a clear movement from simple visual observation to more advanced physiological monitoring, where AI can detect stress before symptoms become obvious to the human eye.

Water and nutrient stress assessment is another fast-growing area. By combining thermal infrared imaging, hyperspectral data and machine learning, AI systems can identify drought stress, nitrogen deficiency and combined stress conditions. Some reviewed studies reported accuracy above 90 percent for water and nutrient stress classification, suggesting that AI could support more precise irrigation and fertilization.

Notably, the review warns that perception systems face persistent barriers. Many models depend on high-quality annotated datasets, which are expensive and difficult to build in agriculture. Performance can drop sharply when models are moved across crop varieties, regions, seasons or climates. Edge deployment also remains difficult because many advanced models require computing resources that are not available on low-cost farm devices.

Data-driven decisions are replacing experience-based farm management

The second stage of the framework is decision making, where AI turns field data into practical recommendations. This layer is central to the transition from traditional farming judgment to model-based agricultural management.

AI decision systems increasingly combine meteorological records, soil measurements, remote sensing imagery and market data. These data streams support yield prediction, pest and disease warnings, irrigation planning, fertilization prescriptions, planting layout decisions and market-linked production strategies.

Yield prediction has seen major advances through multi-source data fusion. The review describes models that combine satellite imagery, UAV imagery, weather data and soil information to estimate crop output at field and regional scales. Long short-term memory models are used to capture time-series patterns, while transformers can identify critical growth-stage features. CNN-based models process spatial imagery, and fusion models integrate RGB, multispectral, thermal infrared, optical and radar data to improve robustness.

Pest and disease early warning systems are also becoming more data-driven. These models combine historical outbreak records, temperature, humidity, precipitation, field survey data and remote sensing indicators to forecast risk. The review notes that time-series models can predict outbreak probability over short future windows, while random forest models can generate risk maps that support targeted pesticide application.

Precision water and nutrient management is another major decision-making application. AI models estimate crop demand by analyzing soil moisture, canopy temperature, weather conditions, soil nutrients and crop growth status. Optimization algorithms then produce irrigation and fertilization prescriptions that aim to balance yield, cost and environmental impact. In greenhouse systems, such approaches have reduced water and fertilizer inputs while improving yield.

The review also expands AI decision making beyond field operations into market-linked agricultural planning. Machine learning can help determine what to plant, where to plant and when to sell by integrating soil suitability, climate adaptability, historical profitability, subsidy policies, demand signals and market prices. The authors highlight the potential of large language models, reinforcement learning and graph neural networks to connect production planning with market demand.

However, the study makes clear that decision systems remain constrained by data quality and integration problems. Agricultural data are highly heterogeneous: weather data are time series, soil data may be sensor-based or laboratory-based, remote sensing images are spatial and often unstructured, and market reports may be semi-structured text. Aligning these sources across time and space remains a major technical challenge.

The review also warns that many AI models still lack explainability. Farmers may be reluctant to follow recommendations if they cannot understand why a model suggests a specific irrigation, fertilization or planting decision. This trust gap is particularly important because agricultural decisions carry direct economic risk.

Robots could close the farming loop, but deployment remains uneven

The third stage of agricultural intelligence is execution, where AI-guided decisions are translated into physical action. The review identifies agricultural robots, autonomous tractors, UAVs and robotic manipulators as key technologies in this layer.

Execution is the weakest but most critical link in the smart agriculture chain. The authors note that perception and decision research has advanced faster than autonomous field operation. Without reliable execution systems, AI-generated insights cannot fully translate into practical farm benefits.

Agricultural robots typically rely on modular architectures that include perception, decision, control, execution and communication modules. Ground robots and UAVs are used for different tasks depending on terrain, crop type and farm size. Wheeled platforms are suited to flat fields, tracked platforms can operate in muddy or mountainous terrain, rail-guided platforms are useful in greenhouses, multi-rotor UAVs support precision spraying and crop monitoring, and fixed-wing UAVs are more suitable for large-area surveys.

Autonomous navigation is central to robotic farming. The review discusses the role of real-time kinematic GPS, LiDAR, inertial measurement units, RGB cameras, multispectral cameras, soil sensors and ultrasonic sensors. These tools help robots locate themselves, detect obstacles, follow crop rows and carry out tasks safely in dynamic field environments.

Path planning remains one of the toughest technical problems. Agricultural environments are unstructured and unpredictable, with uneven terrain, slopes, mud, changing light, crop occlusion, temporary obstacles and weak GPS signals. The review highlights global path-planning algorithms for complex terrain, local navigation systems for orchards and greenhouses, and multi-robot coordination methods for large-scale operations.

The execution layer supports precision spraying, autonomous harvesting and robotic weeding. In spraying, AI can identify target areas and reduce chemical overuse. In harvesting, robotic arms and flexible grippers can pick fruits or crops with less damage. In weeding, machine vision and mechanical tools can target weeds while reducing herbicide dependence.

Despite these benefits, cost and reliability remain major barriers. High-precision sensors, robots and intelligent machinery can be expensive, making them difficult for small and medium farms to adopt. Harsh agricultural conditions such as dust, humidity, salinity, high temperatures and cold weather can also reduce equipment durability. Maintenance costs further limit adoption.

The review gives special attention to global inequality in agricultural AI. It warns that many AI systems are designed around large-scale, mechanized farming in technologically advanced regions. These models may fail in smallholder settings common in the Global South, where farms are smaller, cropping systems are more diverse, internet access is weaker and digital literacy may be lower.

The authors also raise concerns over data sovereignty and what they describe as digital colonialism. Agricultural data are valuable strategic assets, yet farmers in developing regions may become passive data suppliers for external technology companies without fair control or benefit sharing. The review states that locally adapted datasets, low-cost tools, offline functionality, local-language interfaces, community data cooperatives and South-South cooperation will be essential for inclusive smart agriculture.