基于改进YOLOv5s的小麦赤霉病检测方法  

A Wheat Fusarium Head Blight Detection Method Based on Improved YOLOv5s

作　　者：高君 张正华[1] 邵明[1] 廖瑾薇 蔡涵杰 GAO Jun;ZHANG Zhenghua;SHAO Ming;LIAO Jinwei;CAI Hanjie(School of Information Engineering(School of Artificial Intelligence),Yangzhou University,Yangzhou Jiangsu 225127,China)

机构地区：[1]扬州大学信息工程学院(人工智能学院),江苏扬州225127

出　　处：《信息与电脑》2023年第12期61-65,共5页Information & Computer

基　　金：“稻麦生长全程智能测报技术示范推广”(项目编号:2020-SJ-003-YD03);“智能化稻麦生产精准变量喷洒关键技术研究与平台研发”(项目编号:yzuxk202008);“小麦赤霉病单穗感染率测试仪的研发”(项目编号:2022111171827)。

摘　　要：针对目前小麦赤霉病检测依赖人工和实时性较低等缺点,提出一种基于改进YOLOv5s的小麦赤霉病检测算法。首先使用数据增强方式扩充样本集合,增强模型的训练效果,其次引入GhostConv卷积来代替原有Conv卷积,提升计算速度,再次引入卷积注意力机制模块(Convolutional Block Attention Module,CBAM)提升检测精度,最后引入TensorRT网络进行模型检测加速。实验结果表明:改进后的YOLOv5s模型经TensorRT加速后平均精度均值达到0.938,提升了2.6个百分点;单幅图像处理时间为77.8 ms,检测速度提升1.45倍。综上,本次提出的检测方法可以基本满足小麦赤霉病检测对精度和实时性的要求。Aiming at the shortcomings of the current wheat Fusarium head blight detection relying on manual and low real-time performance,a wheat Fusarium head blight detection algorithm based on improved YOLOv5s is proposed.Firstly,use data augmentation to enrich the sample collection to enhance the training effect of the model.Secondly,GhostConv convolution is introduced to replace the original Conv convolution to improve the calculation speed.Thirdly,Convolutional Block Attention Module(CBAM)was introduced to improve the detection accuracy.Finally,TensorRT network is introduced for model detection acceleration.The experimental results show that the average accuracy of the improved YOLOv5s model accelerated by TensorRT reaches 0.938,an increase of 2.6 percentage points.The processing time of a single image is 77.8 ms,and the detection speed is increased by 1.45 times.In summary,the detection method proposed in this paper can basically meet the requirements of accuracy and real-time detection of Fusarium head blight.

关 键 词：小麦赤霉病 YOLOv5s 卷积注意力机制模块(CBAM) TensorRT 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]