Classification of Plant Disease Pests Based on Smart Edge Devices Using Deep Learning Methods

Rice contributes about 70 % of the food supply in human civilization with an increased reliance on rice farming in developing nations. Farmers everywhere struggle economically when their rice production is compromised due to pest and disease factors. Current methods of pest identification, including but not limited to plant pathology as in classification (speed, objectivity, reliability, validity, etc.), continue to be subjective and slow for farmers where time truly is money. The study investigates a method for pest and disease detection that utilises IoT technologies and deep learning methodology (YOLOv8) via Convolutional Neural Networks (CNN) in rice farming including possibilities for smart edge computing devices (e.g., high-resolution cameras and Raspberry Pi microcontrollers) and a dataset with 4647 images for training. Ultimately the results show high performance in general, suggested with an mAP mean Average Precision of mAP 0.936 for moth detection, for precision values in range of 0.679-0.705, and for recall metrics in range of $0.64-0.713$. In fact, it achieved an accuracy that will allow the farmers applicable means to quickly classify more than one pest, pest treatment options both in the field and post-harvest and assist in monitoring systems for farms to implement sustainable farming and environmental benefits.