基于SE-MultiResNet50算法的辣椒病害种类及程度分级检测  

Grading detection of chili pepper diseases species and degree based on the SE-MultiResNet50 algorithm

作　　者：唐源 鲁茂悦 李丽平[2] 唐有万[2] 陈阳洋 李昱瑾 Tang Yuan;Lu Maoyue;Li Liping;Tang Youwan;Chen Yangyang;Li Yujin(College of Computer Science and Cyber Security,Chengdu University of Technology,Chengdu,610059,China;Chengdu Academy of Agriculture and Forestry Sciences,Chengdu,611130,China)

机构地区：[1]成都理工大学计算机与网络安全学院,成都市610059 [2]成都市农林科学院,成都市611130

出　　处：《中国农机化学报》2024年第12期259-266,274,共9页Journal of Chinese Agricultural Mechanization

基　　金：四川省科技计划项目(2021YFN0117)。

摘　　要：在实际的辣椒种植环境中,由于其复杂背景,辣椒叶片病害的识别难度较大。目前,关于受害程度分级检测和辣椒病害分级缺乏公开的数据集。以成都市农林科学院辣椒种植基地的叶片为研究对象,采用U2-Net对叶片进行分割,生成具有不同复杂背景的合成图像,从而丰富数据集。针对常见的辣椒细菌性斑点、白粉病和病毒病3种病害以及健康叶片,提出一种SE-MultiResNet50检测模型。该模型在全由复杂背景图像组成的测试集上表现出色:辣椒病害种类的识别准确率达到91.05%,受害严重程度分级的准确率为92.08%。结果表明,该检测模型在复杂背景下具有较高的识别精度,成功实现对辣椒病害种类分类和受害严重程度分级的智能识别。同时,提供一种新的数据集扩充方法,为相关领域的研究提供新的思路和途径。In the actual pepper planting environment,due to its complex background,the identification of pepper leaf diseases has always been a challenging problem.Currently,there is a lack of publicly available datasets for severity grading detection and classification of chili pepper leaf diseases.This study focuses on leaf samples from the chili pepper plantation base of the Chengdu Academy of Agriculture and Forestry Sciences,by utilizing U 2-Net for leaf segmentation to generate synthetic images with diverse complex backgrounds,thereby enriching the dataset.Addressing common chili pepper diseases such as bacterial spot,powdery mildew,viral diseases,and healthy leaves,a SE-MultiResNet50 detection model is proposed.This model performs remarkably well on a test set comprised entirely of images with complex backgrounds:achieving a recognition accuracy of 91.05%for chili pepper disease types and 92.08%for severity grading.Experimental results of this study demonstrate that the detection model exhibits high recognition accuracy under complex backgrounds,successfully achieving intelligent identification of chili pepper disease types and severity grading.Additionally,a novel dataset augmentation method is provided,offering new insights and avenues for research in related fields.

关 键 词：辣椒 病虫害 分级检测 注意力机制 ResNet50 

分 类 号：S436.418[农业科学—农业昆虫与害虫防治]