Explainable AI for Smart Agriculture: Advances in Field Monitoring, Disease Detection, and Precision Resource Management

Abstract

With a large proportion of the population in rural areas depending on agriculture for their livelihoods, and the sector increasingly at risk and facing challenges such as climate change, irregular rainfall, land degradation, water scarcity and crop diseases, there is an urgent need to practice more efficient farming. Artificial Intelligence (AI), Internet of Things (IoT), drones and deep learning have proven useful in smart agriculture to monitor crops, identify diseases, predict crop yield and maximize resource use. Most AI models are also “black boxes,” However, which makes it challenging for farmers and agricultural professionals to understand the choices made, and reduces farmers' confidence and actual use. To address this, several approaches to interpreting AI's predictions, such as Explainable Artificial Intelligence (XAI), which uses techniques like SHapley Additive exPlanations (SHAP), Local Interpretable Model-agnostic Explanations (LIME), Gradient-weighted Class Activation Mapping (Grad-CAM) and attention mechanisms, have been proposed. In this survey, the recent progress and development of XAI in smart agriculture are summarized, such as crop monitoring, disease detection of plants, soil analysis, management of irrigation water, fertilizer recommendation, and federated learning. It also summarizes the existing research and necessary challenges (such as limited real-world validation, infrastructure limitations, and privacy issues) encountered, and proposes research directions. The study highlights the transformative power of integrating XAI with advanced AI technologies, paving the way for the creation of more transparent, reliable, and user-trusted smart agriculture systems, which will ultimately benefit farmers.