基于SVDD与VGG的纽扣表面缺陷检测  

Detection of button surface defects based on SVDD and VGG

作　　者：樊鑫江 佟强 杨大利[1] 侯凌燕[1] 梁旭 FAN Xin-jiang;TONG Qiang;YANG Da-li;HOU Ling-yan;LIANG Xu(Computer School,Beijing Information Science and Technology University,Beijing 100101,China)

机构地区：[1]北京信息科技大学计算机学院,北京100101

出　　处：《计算机工程与设计》2024年第3期918-924,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(6177010360)。

摘　　要：为解决纽扣表面缺陷检测中人工效率低下,且无需对纽扣表面瑕疵进行分类的问题,提出一种基于DEEP SVDD与改进VGG16的纽扣表面缺陷检测模型。在VGG16中增加BN层加快网络收敛;为提升网络特征提取能力引入SE注意力模块;使用全局平局池化替代全连接层,减少模型参数量,使模型更加健壮。实验结果表明,改进后的模型在DEEP SVDD中的两种方法软边界及一类方法的AUC值分别提升7.7%、5.9%,均高于96%,单张检测时间仅4.5 ms,模型性能满足实际要求。To solve the problem of low manual efficiency in button surface defect detection and no need to classify button surface defects,a button surface defect detection model based on DEEP SVDD(DEEP support vector data description)and improved VGG16(visual geometry group)was proposed.BN(batch normalization)layer was added to VGG16 to accelerate network convergence.SE(squeeze-and-excitation)attention module was introduced to improve the ability of network feature extraction.The full connection layer was replaced by global average pooling to reduce the number of model parameters and make the model more robust.Experimental results show the AUC values of the improved model in DEEP SVDD for two methods.AUCs of Soft boundary method and the first class method are increased by 7.7%and 5.9%respectively,both of which are higher than 96%.The single sheet detection time is only 4.5 ms.The performance of the model meets the actual requirements.

关 键 词：纽扣表面检测 深度支持向量数据描述 VGG16网络模型 注意力机制 全局平均池化层 批量归一化 深度学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]