Detection of diseases on common crop leaves in Vietnam using explainable YOLO architectures

Abstract

Agriculture is a key economic sector in Vietnam but is increasingly confronted with significant challenges from pests and diseases, which lead to substantial yield losses and negatively impact both GDP and exports. Reports indicate that prevalent diseases, such as rice blast, can result in yield reductions of 50-100%, while coffee dieback can decrease productivity by up to 75%. Early detection and diagnosis are critical for timely intervention to mitigate such damages. Nevertheless, traditional methods are predominantly manual, making them time-consuming, inefficient, and costly, particularly for large-scale applications. To address these challenges, this study proposes an automated solution based on deep learning, utilizing the YOLO real-time object detection architecture. The model was trained and evaluated on an extensive dataset of Vietnam’s strategic crops, including rice, coffee, and tea. The experimental results demonstrated high performance, with an accuracy of 0.987 and an mAP50 of 0.984. These findings confirm the model’s superior performance and its potential for developing intelligent, explainable, and scalable crop monitoring systems in Vietnam.