关键词: 水稻叶片, 病害检测, 深度学习, 图像分割, 目标检测

Abstract: The early and accurate detection and identification of rice leaf diseases are necessary in agricultural production. In order to reduce the impact of background noise on the identification of rice leaf diseases and improve the accuracy of disease detection, a rice leaf disease detection method based on the DeepLabv3—Faster R—CNN network is proposed in this study. The rice leaf image to be detected is first subjected to image segmentation by using the DeepLabv3 network, so as to obtain preliminary results of background and leaf segmentation. Then, the leaf part of the image, which has undergone background segmentation, is restored for detection, thereby avoiding the influence of background noise on the detection results. The detection part is mainly implemented by the Faster R—CNN network, and by combining the feature pyramid and the CBAM (Convolutional Block Attention Module) method, the model's ability to detect multi‑scale rice leaf disease targets is improved. By training the model on a public dataset, the proposed method in this study achieves an average detection accuracy of 98.1% for white leaf spot, blast disease, brown spot disease and yellow stunt disease.

Key words: rice leaf, disease detection, deep learning, image segmentation, object detection