Plant Disease Classification of Basal Bulb Rot in Shallots Using Vision Transformer

Abstract

Early detection of plant diseases is critical for sustainable agriculture and reducing crop losses. This study presents a real-time monitoring system integrating IoT and machine learning for the early detection of basal bulb rot disease in shallots. The system combines image data captured by an ESP32-CAM and soil pH data from a sensor to provide timely alerts to farmers. The images undergo preprocessing using Gaussian filtering and histogram equalization, while pH data is smoothed using a moving average filter. Features such as color, texture, shape, and pH dynamics are extracted and analyzed using a hybrid classification model comprising MobileNetV2 for image-based disease identification and Random Forest for soil pH classification, fused at the decision level. The models were optimized for edge deployment using TensorFlow Lite and field-tested under solar-powered conditions. Experimental results demonstrate a hybrid model accuracy of 93.7%, with recall and specificity of 94.1% and 92.8%, respectively. The system responds within 450 milliseconds, making it suitable for real-time applications. This solution offers a low-cost, scalable, and accurate method for precision agriculture, reducing dependence on manual inspections and enabling proactive disease management.