Accurate Detection of Pests and Diseases in Rice Fields Using Drone Image Recognition

Abstract

In recent years, with the development of technology, unmanned aerial vehicle (UAV) remote sensing has higher efficiency, flexibility, and resolution than manual monitoring. Combining with deep learning technology has become the mainstream direction for the intelligent identification of rice pests and diseases. This article uses drone image recognition to monitor the research status of rice field diseases and pests, summarizes the research progress of scholars, and summarizes the differences of the YOLO series, Convolutional Neural Network (CNN), U-shape Network (U-Net), and other models in rice disease and pest identification. The research points out the current challenges in datasets, normalization, model generalization, and cross-regional applications. In response to the above issues and challenges, this article proposes future development directions: jointly building standardized open source datasets for multiple regions, multiple time periods, and multiple categories, and uniformly labeling them with specifications; Implement lightweight deployment of models through techniques such as model pruning, quantification, and knowledge distillation; Combining multi-scale feature fusion and weakly supervised learning to enhance early disease and pest identification capabilities; Integrate multiple sources of data such as RGB, multispectral, meteorological, and crop growth period to build an integrated closed-loop system for monitoring, early warning, and precise prevention and control.