改进SSD算法在工件表面缺陷检测中的应用  

Application of Improved SSD Algorithm in Surface Defect Detection of Workpieces

作　　者：周茂军[1] 胡江涛 王俊杰 彭德政 马沁怡[1] 王雅君[1] ZHOU Maojun;HU Jiangtao;WANG Junjie;PENG Dezheng;MA Qinyi;WANG Yajun(School of Mechanical Engineering and Automation,Dalian Polytechnic University,Dalian 116034,China)

机构地区：[1]大连工业大学机械工程与自动化学院,大连116034

出　　处：《组合机床与自动化加工技术》2024年第8期139-144,共6页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：国家重点研发计划项目(2022YFD2100603);辽宁省教育厅高校基本科研项目(LJKMZ20220888)。

摘　　要：为能够更为高效地检测工件表面加工缺陷,提出了一种MobileNetV3-SSD目标检测算法,它能够有效实现工件表面缺陷检测。此算法通过改进SSD的骨架网络为MobileNetV3-Large,有效减少网络参数量和计算量,通过结合自底向上的特征金字塔网络的改进算法,使网络能够捕捉不同尺度和形状的目标信息,将语义分割辅助任务与多层特征融合相结合,进一步提升检测性能。使用PASCAL VOC数据集和Kolektor工业缺陷数据集验证该算法,在PASCAL VOC数据集上的mAP达到77.3%,较SSD算法提高0.4%,在Kolektor数据集上的FPS达到了105,较SSD算法提高112%。验证了MobileNetV3-SSD算法相较于传统SSD网络,具有更好的检测精度和检测速度。A MobileNetV3-SSD object detection algorithm is proposed to efficiently detect surface defects in workpieces.This algorithm improves the skeleton network of SSD by using MobileNetV3-Large,effectively reducing the network parameters and computational complexity.By combining a bottom-up feature pyramid network and an improved algorithm,the network is able to capture target information of different scales and shapes.Furthermore,the algorithm incorporates semantic segmentation as an auxiliary task and fuses multi-level features to further enhance the detection performance.The method is validated using the PASCAL VOC dataset and the Kolektor industrial defect dataset,achieving an mAP of 77.3%on the PASCAL VOC dataset,which is a 0.4%improvement over the SSD algorithm.The FPS on the Kolektor dataset reaches 105,which is a 112%improvement over the SSD algorithm.These results validate that the MobileNetV3-SSD algorithm outperforms the traditional SSD network in terms of detection accuracy and speed.

关 键 词：目标检测 MobileNet SSD算法 缺陷检测 

分 类 号：TH164[机械工程—机械制造及自动化] TG66[金属学及工艺—金属切削加工及机床]