基于改进型SSD的多类汽车车身冲压件识别算法  

Recognition Algorithm for Multi-Class Stamping Parts of Automobile Body Based on Improved SSD

作　　者：徐聪聪 胡习之[1] 姜立标[1] 李小军 XU Congcong;HU Xizhi;JIANG Libiao;LI Xiaojun(School of Mechanical and Automotive Engineering,South China University of Technology,Guangzhou 510640,China)

机构地区：[1]华南理工大学机械与汽车工程学院,广州510640

出　　处：《重庆理工大学学报（自然科学）》2020年第7期35-41,74,共8页Journal of Chongqing University of Technology：Natural Science

基　　金：广东省科技计划项目(201513010137002);广州市科技计划项目(201707020045)。

摘　　要：智能机器人在汽车生产制造中有着重要意义,准确地识别抓取任务中的目标是基于视觉引导的汽车车身冲压件抓取系统的基础。针对传统工件识别算法人工提取特征困难、通用性差、识别率不高且易受环境因素影响等问题,首先采用深度学习SSD网络模型对10类汽车车身冲压件进行识别。在此基础上,为了提高工件识别准确率,改善工件相互遮挡情况下识别差的问题,提出一种改进的SSD算法,引入残差网络,采用Resnet-50替换原SSD的基础网络VGG-16。实验结果表明:原始的SSD网络在自制的工件数据集评估集上的平均准确率均值m AP为92. 3%,改进后的SSD网络检测的平均准确率均值m AP为98. 3%,比原始的SSD网络提高了6%,基于Resnet-50改进的SSD模型具有更高的识别准确率、更好的遮挡识别效果以及更强的泛化性能。Intelligent robots play an important role in automobile manufacturing. Accurately identifying the target of grasping task is the basis of visual-guided grasping system for automobile body stamping parts. Aiming at the problems of difficult feature extraction, poor universality, low recognition rate and vulnerability to environmental factors in traditional workpiece recognition algorithm,the deep learning model SSD was used to identify 10 types of automobile body stamping parts. Furthermore,in order to improve the accuracy of workpiece recognition and the situation of poor target detection by occlusion between workpieces,an improved SSD algorithm is proposed. The residual network was introduced,and the basic network VGG-16 of the original SSD was replaced by Resnet-50. The experimental results show that the average accuracy of the original SSD model in the test set of self-made workpiece data set is 92. 3%,while the average accuracy of the improved SSD model is 98. 3%,which is 6% higher than the original SSD model. The improved SSD model based on Resnet-50 has higher recognition accuracy, better occlusion recognition effect and stronger generalization performance.

关 键 词：车身冲压件 改进的SSD模型 工件识别 残差网络 

分 类 号：TH165[机械工程—机械制造及自动化]