摘要
在玉米生长过程中,灰斑病、叶锈病和叶斑病是几种常见的玉米叶片病害。目前,深度学习成为玉米病害识别的重要方法。为提高玉米叶片病害的识别准确率,将在传统目标检测网络模型YOLOv5s的基础上进行优化改进,对玉米叶斑病、叶锈病、灰斑病3种叶片病害图像和正常玉米叶片图像进行识别。首先,通过镜像翻转、图像增强及亮度调节方法等对病害图像进行预处理操作,增强数据集并提高网络鲁棒性,将YOLOv5s网络模型原有的Bottleneck CSP模块替换为CBAM注意力机制模块,并与原始的YOLOv5s网络模型进行对比实验。实验结果表明,该检测方法对玉米叶片病害识别的平均准确率为95.6%,识别精度较原始YOLOv5s网络模型有所提升,可为玉米叶片病害识别提供有效的技术支持。
During the growth process of corn,gray spot disease,leaf rust,and leaf spot disease are several common corn leaf diseases.Currently,deep learning has become an important method for identifying corn diseases.To improve the recognition accuracy of corn leaf diseases,this article will optimize and improve the traditional object detection network model YOLOv5s,and recognize three types of leaf disease images of corn leaf spot,leaf rust,and gray spot,as well as normal corn leaf images.Firstly,the disease images are preprocessed using methods such as mirror flipping,image enhancement,and brightness adjustment to enhance the dataset and improve network robustness.Then,the original Bottleneck CSP module of the YOLOv5s network model is replaced with a CBAM attention mechanism module and compared with the original YOLOv5s network model.The experimental results show that the average accuracy of the detection method used in this article for identifying corn leaf diseases is 95.6%,which is improved compared to the original YOLOv5s network model and can provide effective technical support for identifying corn leaf diseases.
作者
凌慕菲
杨冬风
Ling Mu-fei(Heilongjiang Bayi Agricultural University,Daqing,Heilongjiang 163711)
出处
《农业灾害研究》
2023年第8期126-128,共3页
Journal of Agricultural Catastrophology