基于改进SSD的钢材表面缺陷检测  

Steel surface defect detection based on improved SSD

作　　者：阎馨[1] 杨月川 屠乃威[1] YAN Xin;YANG Yuechuan;TU Naiwei(School of Electrical and Control Engineering,Liaoning Technical University,Huludao 125105,China)

机构地区：[1]辽宁工程技术大学电气与控制工程学院,葫芦岛125105

出　　处：《现代制造工程》2023年第5期112-120,共9页Modern Manufacturing Engineering

基　　金：国家自然科学基金项目(61601212,71771111);辽宁省教育厅辽宁省高等学校基本科研项目(LJ2017QL012);辽宁工程技术大学博士启动基金项目(14-1102)。

摘　　要：工业生产过程中,钢材表面缺陷的检测对于钢材的质量控制发挥着十分重要的作用,针对钢材表面缺陷检测中存在的检测精度低、检测速度慢等问题,提出了一种钢材表面缺陷检测的改进SSD算法。在所提算法中,采用Transformer多头注意力机制模块代替原SSD结构中的Conv5_1层,以提高小目标检测的能力;原SSD结构中的Conv7操作替换为Involution算子操作,以减少运算的参数量;对网络结构进行特征融合处理,以更全面地检测特征图中所包含的信息。利用NEU-DET数据集进行实验,实验结果表明改进后的SSD算法是有效的,可以高效检测到钢材表面的小目标缺陷,相比改进前平均检测精度提高了4.5%,检测速度提高了13.6%。In the process of industrial production,the detection of steel surface defects plays a very important role in the quality control of steel.Aiming at the problems of low detection accuracy and slow detection speed in the detection of steel surface defects,an improved SSD detection method of steel surface defects was proposed.In the proposed method,Transformer multi head attention mechanism module was used to replace Conv5_1 layer in the original SSD structure to improve the ability of small target detection;the Conv7 operation in the original SSD structure was replaced by the Involution operator operation to reduce the amount of parameters of the operation;the network structure was processed by feature fusion to more comprehensively detect the information contained in the feature map.The experiment was carried out with NEU-DET data set.The experimental results show that the improved SSD method is effective and can effectively detect small target defects on the steel surface.Compared with the previous improvement,the average accuracy is improved by 4.5%and the detection speed is improved by 13.6%.

关 键 词：钢材表面缺陷检测 改进SSD算法 注意力机制 Involution算子 特征融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]