基于改进YOLOv7模型的PPR管铜嵌件缺陷检测方法  

Defect Detection Method of Copper Insert in PPR Pipes Based on Improved YOLOv7 Model

作　　者：胡波 刘玉良 郑伟 HU Bo;LIU Yu-liang;ZHENG Wei(Zhejiang College of Security Technology,Zhejiang Wenzhou 325016,China;School of Information Engineering,Zhejiang Ocean University,Zhejiang Zhoushan 316022,China)

机构地区：[1]浙江安防职业技术学院,浙江温州325016 [2]浙江海洋大学信息工程学院,浙江舟山316022

出　　处：《机械设计与制造》2024年第11期77-82,共6页Machinery Design & Manufacture

基　　金：浙江安防职业技术学院重点科研项目(AF2022Z03);浙江省公益技术应用研究计划项目(2015C31072);浙江省教育厅一般科研项目(Y202249514)。

摘　　要：针对无规共聚聚丙烯(Polypropylene Random Copolymer,PPR)管铜嵌件缺陷复杂、多样性、尺度差异大的检测难点,提出了一种改进k均值聚类算法(K-means Clustering Algorithm,K-means)结合多尺度特征融合和注意力机制融入的改进YOLOv7(You Only Look Once Version 7)模型检测方法。先用K-means++聚类生成各类缺陷集的平衡标签,再将平衡标签作为第二步K-means聚类的初始中心,使得聚类获得的初始锚框更适合本缺陷样本集。通过融入坐标注意力(Coordinate Attention,CA)机制使模型在检测复杂缺陷时更关注其重要特征,减少冗余特征的干扰。针对缺陷尺度差异性大的特点,采用多尺度特征融合增加了极小尺度检测模块。实验结果表明,改进后的YOLOv7模型比原始模型平均精度均值(Mean Average Precision,mAP)提升了2.29%,和其他主流检测算法相比,在维持检测速度基础上提升了检测精度。改进后的模型mAP达到89.21%,检测速度达到57帧/秒,具有较好的工业应用前景。To solve the defect detection for copper insert in polypropylene random copolymer(PPR)pipes due to complexity,diversity and great difference in scale,an improved you only look once version 7(YOLOv7)model detection method was proposed based on the improved K-means clustering algorithm(K-means)combined with multi-scale feature fusion and attention mechanism integration.Firstly,K-means++ clustering was used to generate balance labels of various defect dataset,and then the balance labels were taken as the initial center of the K-means clustering,making the clustering initial anchor frame more suitable for the defect sample dataset.By integrating coordinate attention(CA)mechanism,the model paid more attention to the important features when detecting complex defects and reduced the disturbance caused by redundant features.In view of the great difference of defect scale,multi-scale feature fusion module was adopted to add a minimum scale detection module.The experiment results show that the improved yolov7 improves the mean average precision(mAP)by 2.29% compared with the original mode.What's more,compared with the other mainstream detection algorithms,the improved YOLOv7 model improves the detection accuracy on the basis of maintaining the detection speed.The mAP of the improved model reaches 89.21% and detection speed reaches 57 frame/s,which reveals the improved model has a prospect in industrial application.

关 键 词：缺陷检测 YOLOv7 改进K-MEANS 多尺度特征融合 坐标注意力机制 

分 类 号：TH16[机械工程—机械制造及自动化] TP391[自动化与计算机技术—计算机应用技术]