Artificial Intelligence Techniques for IoT-Based Soil Nutrition and Plant Disease Detection System for Smart Agriculture Using Multi-Layer Stacked Residual Coordinate Boosted Sooty Tern Attention Network: Trends and Challenges

Abstract

The rapid growth of global population and climate variability has intensified the need for efficient and sustainable agricultural practices. Smart agriculture, driven by Artificial Intelligence (AI) and the Internet of Things (IoT), offers innovative solutions for real-time monitoring of soil nutrition and early detection of plant diseases. IoT-based systems enable continuous data acquisition from environmental sensors, while deep learning models provide automated analysis and decision-making capabilities. Recent advancements in deep learning, particularly Convolutional Neural Networks (CNNs), Residual Networks (ResNet), and Vision Transformers (ViT), have significantly improved plant disease detection accuracy. These models automatically extract complex features from plant images and environmental data, outperforming traditional machine learning approaches. Studies indicate that deep learning models provide superior accuracy and efficiency in disease classification tasks due to automated feature learning and scalability. The integration of IoT with deep learning enables real-time monitoring, prediction, and management of agricultural systems. Advanced hybrid architectures, such as multi-layer stacked residual networks combined with coordinate attention and transformer-based attention mechanisms, enhance both spatial and contextual feature representation. These models improve detection performance by focusing on relevant regions and capturing global dependencies. This paper presents a comprehensive review of AI techniques for IoT-based soil nutrition monitoring and plant disease detection systems, focusing on developments from 2020 to 2023. It analyses deep learning architectures, hybrid attention models, and optimization techniques, and discusses key challenges such as data heterogeneity, computational complexity, and deployment constraints. Future directions, including edge computing, explainable AI, and federated learning, are also explored.