IMPROVED YOLOv8-BASED AUTOMATED DETECTION OF WHEAT LEAF DISEASES

Abstract

Stripe rust, leaf rust, and powdery mildew are important leaf diseases in wheat, which significantly affect the yield and quality of wheat. Their timely identification and diagnosis are of great significance for disease management. To achieve convenient identification of wheat leaf diseases based on mobile devices, an improved YOLOv8 method for wheat leaf disease detection is proposed. This method incorporates the CBAM (Convolutional Block Attention Module) attention mechanism module into the feature fusion network to enhance the network's feature expression ability. Experimental results show that the improved YOLOv8 model has an accuracy, recall rate, and mean average precision (mAP) of 95%, 98.3%, and 98.8% respectively for wheat leaf disease detection, with a model memory usage of 5.92 MB. Compared with the Faster R-CNN, YOLOv5, YOLOv7, and YOLOv8 models, the mAP has been improved by 66.76, 48, 13.2, and 1.9 percentage points respectively, and it also has the lowest model memory usage. The research demonstrates that the improved YOLOv8 model can provide an effective exploration for automated detection of wheat leaf diseases.