Article’s

The integration of Artificial Intelligence (AI), the Internet of Things (IoT), and satellite-based remote observation is reshaping contemporary agriculture, shifting it from an experience-driven manual discipline into a richly data-oriented field. Although standalone machine learning models targeting crop selection, disease identification, and harvest volume estimation have individually attained impressive accuracy, existing research exposes a critical fragmentation in their real-world deployment: these modules rarely communicate within a consolidated decision-support environment. This review systematically examines the architectural requirements for a unified Sense-Analyze-Act agricultural framework. It critically assesses the effectiveness of ensemble learning approaches, notably Random Forest classifiers, for matching soil nutrient profiles to suitable crops. Convolutional Neural Networks (CNNs), particularly MobileNetV2, for lightweight field-based plant disease surveillance [11], [14]; and Long Short-Term Memory (LSTM) networks for predicting temporal patterns in commodity prices and soil moisture levels . The transformative potential of Large Language Models (LLMs) combined with Retrieval-Augmented Generation (RAG) in making agronomic guidance accessible to resource-limited smallholder farmers is also thoroughly examined. This review highlights key obstacles to deployment, namely cross-domain generalization failures, algorithmic biases stemming from unrepresentative training data, and the absence of robust multimodal sensor fusion architectures. Finally, a forward-looking research agenda is proposed, emphasizing Federated Learning approaches and autonomous unmanned aerial vehicle (UAV) surveillance programs as pathways toward bridging the gap between demonstrated algorithmic promise and smallholder operational reality. Keywords: Precision agriculture, ensemble learning, convolutional neural networks, long short-term memory networks, multimodal artificial intelligence, smart farming, remote sensing, federated learning, retrieval-augmented generation, plant disease detection.