基于改进YOLOv3的电容表面缺陷检测方法  

Detection method of capacitor surface defects based on improved YOLOv3

作　　者：李俊杰 周骅 唐纲浩 LI Junjie;ZHOU Hua;TANG Ganghao(College of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China)

机构地区：[1]贵州大学大数据与信息工程学院,贵阳550025

出　　处：《智能计算机与应用》2023年第3期235-241,共7页Intelligent Computer and Applications

基　　金：国家自然科学基金联合基金重点支持项目(U1836205);贵阳市科技计划项目《面向工业物联网的智能安全融合关键技术与应用》(筑科合同[2021]1-5号)。

摘　　要：基于在电容表面缺陷识别任务中所用方法效率低下、准确率低等问题,提出了一种改进的YOLOv3(you only look once)的电容表面缺陷检测方法。使用CIoU(Complete Intersection over Union)进行损失函数的计算,加速收敛、提高定位精度;通过Mosaic方法增强数据集,丰富检测样本;使用K-means聚类算法优化先验框,使先验框更适于解决电容表面缺陷检测问题;引入基于PANet(Path Aggregation Network)结构的特征融合层,增强语义信息与定位信息。实验结果表明,改进的YOLOv3算法的平均精度均值可以达到89.70%,与原始算法相比有4.79%的提升。相比于其他主流算法,该算法在钽电容表面缺陷检测中精度更高,且能满足实时检测的要求。Based on the problems of poor efficiency and low accuracy of the methods used in the task of capacitor surface defect recognition,an improved YOLOv3(you only look once)method for capacitor surface defect detection is proposed.In the research,use Complete Intersection over Union(CIoU)to calculate the loss function to speed up convergence and improve positioning accuracy;use Mosaic method to enhance the data set and enrich the detection samples;use K-means clustering algorithm to optimize a priori box,making a priori box more suitable for solving the detection of capacitance surface defects problem;introduce a feature fusion layer based on the Path Aggregation Network(PANet)structure to enhance semantic information and positioning information.Experimental results show that the average accuracy of the improved YOLOv3 algorithm can reach 89.70%,which is the improvement of 4.79%compared with the original algorithm.Compared with other mainstream algorithms,this algorithm has higher accuracy in surface defect detection of tantalum capacitors and can meet the requirements of real-time detection.

关 键 词：图像处理 损失函数 特征融合 深度学习 缺陷检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]