关键词: 水稻叶片, 病害检测, YOLOv7, 数据增强, CARAFE

Abstract: In order to solve the problem of difficult and effective detection of rice diseases, a rice disease detection algorithm called DNC-YOLOv7 was proposed, focusing on key diseases such as bacterial blight, rice blast, and brown spot. Firstly, in response to the shortcomings of YOLOv7's original upsampling module in extracting semantic information of rice diseases, the NC (Nearest CARAFE) upsampling module was introduced. This module significantly improved the ability of the network model in restoring the details of rice leaf images, so that the model could more accurately capture and identify disease features. Secondly, in order to further enhance the feature extraction and fusion capabilities of the model, the DFPN structure was proposed to improve the neck design of the original model. Finally, Mixup and Mosaic techniques were used to enhance the original dataset to enhance the model's generalization ability and robustness. The experimental results show that the average detection accuracy of DNC-YOLOv7 algorithm on the dataset is significantly improved from the original 83.4% to 93.2%, which is 9.8% higher than the traditional YOLOv7 algorithm.

Key words: rice leaves, disease detection, YOLOv7, data augmentation, CARAFE