摘要
提出了一种基于改进的YOLOv5模型算法的蚕茧缺陷检测的智能分选方法。首先,搭建视觉测量系统,实时高效采集高分辨率的蚕茧图片。其次,利用空域的方式对样本图像进行特征增强;使用K-means++聚类算法,对本文数据集进行尺度分类,以更好地适应各种目标尺度的检测;此外,对YOLOv5模型检测部分进行修改,使用挤压激励(SE)模型,在前3个特征尺度的特征输出部分中添加该模块,以提高模型的有效特征响应。最后,对检测结果进行对比。实验结果表明:改进后的方法对特征检测的平均精度均值(mAP)提高了5.5%,其缺陷的检测精度为96.6%、检测速率为43 fps,能够实现对蚕茧缺陷的精确检测,实现蚕茧的高效精确智能分选。
An intelligent sorting method for cocoon defect detection based on an improved YOLOv5 model algorithm is proposed.Firstly,a visual measurement system is established to collect high-resolution cocoon images high efficiently in real-time.Secondly,feature enhancement is performed on the sample image using spatial methods;K-means++clustering algorithm is used for scale classification of the dataset in this article in order to better adapt to the detection of various target scales.In addition,using the squeeze exception(SE)model,the detection section of the YOLOv5 model is modified and this module is added to the feature output section of the first three feature scales to improve the effective feature response of the model.Finally,the detection results are compared.The experimental results show that the improved method to the mean average precision(mAP)of feature detection is improved by 6.5%,with a defect detection precision of 96.6% and a detection rate of 43 fps.It can achieve precision detection of cocoon defects and efficient and intelligent sorting of cocoons.
作者
姜阔胜
武松梅
任杰
JIANG Kuosheng;WU Songmei;REN Jie(School of Mechanical Engineering,Anhui University of Technology,Huainan 232001,China;Modern Fashion College of Anhui Vocational and Technical College,Hefei 230011,China)
出处
《传感器与微系统》
CSCD
北大核心
2023年第11期143-146,共4页
Transducer and Microsystem Technologies
基金
2023年度安徽省高校科研项目—重点项目(2023AH051426)
国家重点研发计划资助项目(SQ2020YFB130256)
2022年度安徽省高校科研项目(2022AH040274)
安徽省教育厅省级质量工程纺织品设计教学团队(2021jxtd155)。
