Improved detection of yolov4 sunflower leaf diseases

Abstract

In order to improve the timeliness and accuracy of the automatic detection and recognition of sunflower leaf diseases in Inner Mongolia, this study used an improved Yolov4 model to detect and recognize the three diseases of sunflower leaf verticillium wilt, powdery mildew and rust. First, use the K-means clustering algorithm to cluster sunflower disease samples to generate a new anchor box size, making the anchor box of the Yolov4 network model more suitable for the detection of smaller target lesions on sunflower leaves; secondly, define the three network functions of MobileNetV1, MobileNetV2 and MobileNetV3, obtain three effective feature layers corresponding to each MobileNet network, and use these three effective feature layers to replace the effective feature layer of the original Yolov4 backbone network CSP-Darknet53 as the backbone feature extraction network. The experimental results show that the accuracy and recall of the improved method proposed in this paper are significantly improved compared with the original Yolov4 model. Therefore, the deep learning model trained by the improved Yolov4 algorithm has better robustness and is more suitable for sunflower leaf disease detection.