基于深度学习的汽车油箱盖区域检测  

Area Detection of Automobile Fuel Tank Cap Based on Deep Learning

作　　者：李锁 杨凯悦 由金池 廖伟 李悦宁 LI Suo;YANG Kaiyue;YOU Jinchi;LIAO Wei;LI Yuening(Collge of Mechanical Engineering,Shenyang Ligong University,Shenyang 110159,China;Engineering Training Center of DUT,Shenyang Aerospace University,Shenyang 110136,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]沈阳理工大学机械工程学院,辽宁沈阳110159 [2]沈阳航空航天大学工程训练中心,辽宁沈阳110136 [3]中国科学院大学,北京100049

出　　处：《长江信息通信》2023年第6期13-16,共4页Changjiang Information & Communications

基　　金：辽宁省教育厅面上青年人才项目(LJKZ0258);辽宁省科技厅博士科研启动基金计划项目(2022-BS-187)。

摘　　要：针对传统的目标检测方法油箱盖形状识别准确率低的问题,提出融合注意力机制和深度学习的汽车油箱盖区域检测方法。首先,通过添加注意力模块对Yolov5算法进行改进,加强提取检测对象的特征,提升Yolov5算法对小目标对象的检测准确率;然后,在C3模块以及主干网络中最后一层分别添加SE模块以及CBAM模块;最后,通过Loss以及mAP训练结果对比得出在哪个位置添加何种模块对目标检测准确率提升效果更好。实验结果表明,在C3部分融合了SE模块之后,损失更加小接近0,提升了检测能力,mAP提高了0.5%,基于改进算法后进行目标检测,置信度提升了2%左右。Aiming at the low accuracy of the traditional target detection method for the shape recognition of the fuel tank cap,a new method for the detection of the fuel tank cap region based on attention mechanism and deep learning is proposed.First of all,Yolov5 algorithm is improved by adding attention module to enhance the detection accuracy of Yolov5 algorithm for small target objects;Then,SE module and CBAM module are added to C3 module and the last layer of backbone network respectively;Finally,through the comparison of the training results ofLoss and mAP,we can find out which module is added at which position to improve the accuracy of target detection better.The experimental results show that after the SE module is partially fused in C3,the loss is smaller and closer to 0,and the detection ability is improved.The mAP is improved by 0.5%.The target detection based on the improved algorithm increases the confidence by about 2%.

关 键 词：目标检测 Yolov5算法 注意力机制 深度学习 

分 类 号：TP24[自动化与计算机技术—检测技术与自动化装置]