摘要
针对目前人工在汽车空调出风口缺陷检测任务中效率低、漏检率高的问题,构建一种YOLO-CSD模型对其进行更高效的检测。首先,在YOLOv5主干网络中引入坐标注意力机制(coordinate attention,CA),增强网络对汽车空调出风口缺陷的特征提取能力;其次,引入空间转深度(Space-to-Depth,SPD-Conv)下采样模块,替换网络中原有的跨步卷积和池化层,减少特征提取过程中细微特征的丢失;最后,将耦合检测头替换成解耦头(Decoupled Head),缓解分类任务与回归任务的冲突问题,从而提高缺陷检出率。在自制的汽车空调出风口缺陷数据集上,改进后模型的平均精度(mAP)达到93.6%,提高了3.4个百分点,平均每张图片检测时间29.4 ms。YOLO-CSD在汽车空调出风口缺陷检测中有明显效果,为构建汽车空调出风口自动化质检系统奠定了基础,同时也为其他镀铬产品的表面缺陷检测提供了新的参考。
Aiming at the current problem of low efficiency and high leakage rate of manual labor in the task of detecting defects in automotive air conditioning vents,a YOLO-CSD model is constructed to detect them more efficiently.First,coordinate attention(CA)is introduced into the YOLOv5 backbone network to enhance the feature extraction capability of the network for automotive air-conditioning vent defects.Second,the(Space-to-Depth,SPD-Conv)downsampling module is introduced to replace the original stepwise convolution and pooling layer in the network to reduce the loss of subtle features in the process.Finally,the coupled detection head is replaced by the decoupled head to alleviate the conflict problem between the classification task and the regression task,thus improving the defect detection rate.The average accuracy(mAP)of the improved model reaches 93.6%on the homemade automobile air conditioner vent defect data set,which is 3.4 percentage points higher,and the average detection time per image is 29.4 ms.YOLO-CSD has obvious effects in the detection of defects on automobile air conditioner vents,which lays the foundation for the construction of automated quality inspection system for automobile air conditioner vents,and also provides a new reference for surface defects detection of other chrome-plated products.
作者
徐海福
郑刚
仇梁
XU Haifu;ZHENG Gang;QIU Liang(Engineering Institute of Advanced Manufacturing and Modern Equipment Technology,Jiangsu University,Zhenjiang 212013,China)
出处
《自动化与仪表》
2024年第7期87-91,100,共6页
Automation & Instrumentation
关键词
汽车空调出风口
YOLOv5
镀铬面缺陷检测
深度学习
automotive air vent trim rings
YOLOv5
detection of defects on chrome-plated surfaces
deep learning
