Early Detection of Paddy Crop Diseases Using Drone Images and Machine Learning–Deep Learning Techniques

Abstract

Global food security is seriously threatened by paddy diseases, which have the ability to reduce yearly harvests by 20–40%. An automated approach for targeted pesticide distribution and real-time paddy disease detection is presented in this research. A high-resolution camera-equipped drone (Drone 1) takes live field images and sends them to a cloud platform over Wi-Fi. A hybrid deep learning model that combines EfficientNet-B7 and an Improved Swin Transformer is used to extract disease features from images after they have been denoised using a Modified Non-Local Means Filter (M-NLMF). The leaf state is then divided into four groups by an Extreme Learning Machine (ELM) 4 classes. Once the disease is confirmed, a second drone (Drone 2) finds the GPS-tagged diseased area on its own and exclusively sprays the afflicted plants with pesticide. The technology drastically lowers chemical waste and environmental impact while achieving 99.30% classification accuracy.