Plant Leaf Disease Detection Incorporating IoT and XAI-Enhanced Deep Learning

Plant diseases represent a significant global threat, affecting both food supply and economic growth. Despite technological advances, early plant disease recognition remains difficult for small-scale farmers because of the lackings of infrastructure and resources. Thus, early disease detection using machine learning, deep learning, and computer vision techniques can improve the prediction process and make it easier. Previous studies that were implemented for plant-specific purposes, require high-resolution images and lack real-time detection methods. Therefore, to address these issues, we proposed a deep learning and IoT-based system that uses convolutional neural networks (CNN) and other supporting models, such as InceptionNetV3, VGG16, and EfficientNetB5, with a fine-tuning process to achieve higher accuracy in predicting 38 diseases over 14 different plant species in lower resolution, noisy data. Besides, to facilitate the system, we incorporated an ESP32CAM microcontroller and a server host that can be used to detect diseases in a real-time environment, which makes our research study robust and meaningful. We also implemented Explainable Artificial Intelligence (XAI), particularly Grad-CAM, along with our CNN model for visualizing the infected region in a plant leaf. The individual-trained models achieved an average accuracy of around 99%. We believe, our study can contribute to the agricultural sector by providing real-time disease detection, smart decision support systems, and low-cost solutions for farmers, mainly in resource-limited territories.