基于ResNet的改进多车型车门分类算法  

An improved multi-vehicle door classification algorithm based on ResNet

作　　者：王立君[1] 王玉龙 许奇珮 周丽娟[4] WANG Lijun;WANG Yulong;XU Qipei;ZHOU Lijuan(School of Applied Technology,Changchun University of Technology,Changchun 130102,China;School of Computer Science&Engineering,Changchun University of Technology,Changchun 130102,China;School of Mechatronic Engineering,Changchun University of Technology,Changchun 130012,China;School of Journalism&Communication,Changchun University of Technology,Changchun 130102,China)

机构地区：[1]长春工业大学应用技术学院,吉林长春130102 [2]长春工业大学计算机科学与工程学院,吉林长春130102 [3]长春工业大学机电工程学院,吉林长春130012 [4]长春工业大学新闻与传播学院,吉林长春130102

出　　处：《长春工业大学学报》2024年第6期530-536,共7页Journal of Changchun University of Technology

基　　金：吉林省教育厅基金资助项目(JJKH20230772CY)。

摘　　要：针对柔性焊装生产线中多车型车门分类算法的低精度问题,设计了一种基于改进ResNet18模型的多车型车门分类算法。为提升模型的特征提取和表达能力,算法模型引入了通道增强注意力模块和密集残差模块。通道增强注意力模块通过自适应强化重要特征通道,提升了模型对关键信息的关注能力;密集残差模块通过连接多个层的特征,有效减缓了梯度消失问题,促进了信息流动和再利用,进而增强了模型训练效果和准确性。结合这两个模块,算法模型显著提高了不同车型车门的识别精度及其泛化能力。经过一系列实验验证,改进的ResNet18模型在Top-1准确率达到99.01%,较原始ResNet18模型提升了1.8%。Aiming at the low-accuracy problem of the multi-model door classification algorithm in the flexible welding production line,this study designs a multi-model door classification algorithm based on the improved ResNet18 model.In order to improve the feature extraction and expression ability of the model,the algorithm model introduces a channel-enhanced attention module and a dense residual module.The channel-enhanced attention module improves the model's ability to pay attention to key information by adaptively strengthening the important feature channels,while the dense residual module effectively mitigates the gradient vanishing problem by connecting the features of multiple layers and promotes the flow and reuse of information,which in turn enhances the model training effect and accuracy.Combining these two modules,the algorithmic model significantly improves the recognition accuracy of doors of different car models and its generalization ability.After a series of experimental validations,the improved ResNet18 model achieves 99.01%in Top-1 accuracy,which is a 1.8%improvement over the original ResNet18 model.

关 键 词：机器视觉 图像识别 深度学习 分类模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TP311.5[自动化与计算机技术—计算机科学与技术]