Tomato brown rot disease detection using improved YOLOv5 with attention mechanism

Abstract

Brown rot disease poses a severe threat to tomato plants, resulting in reduced yields. Therefore, the accurate and efficient detection of tomato brown rot disease through deep learning technology holds immense importance for enhancing productivity. However, intelligent disease detection in complex scenarios remains a formidable challenge. Current object detection methods often fall short in practical applications and struggle to capture features from small objects. To overcome these limitations, we present an enhanced algorithm in this study, building upon YOLOv5s with an integrated attention mechanism for tomato brown rot detection. We introduce a hybrid attention module into the feature prediction structure of YOLOv5s to improve the model’s ability to discern tomato brown rot objects in complex contexts. Additionally, we employ the CIOU loss function for precise border regression. Our experiments are conducted using a custom tomato disease dataset, and the results demonstrate the superiority of our enhanced algorithm over other models. It achieves an impressive average accuracy rate of 94.6% while maintaining a rapid detection speed of 112 frames per second. This innovation marks a significant step toward robust and efficient disease detection in tomato plants.