基于ECA-SSD模型的汽车零件缺陷检测

摘要/Abstract

摘要： 汽车零件对汽车外观、性能以及安全性都有重大影响。由于汽车零件数量大、体积小、对精度要求高，因此对零件检测的精度和速度都有一定的要求。本文利用图像处理技术，以SSD模型为基础，将其中的VGG模块用深度可分离卷积和线性瓶颈倒残差结构替换，并引入避免降维的局部跨通道交互有效的注意力机制ECA模块，在减少模型参数运算量的同时，适当增加通道以提高模型精度，并将注意力放在图像目标上，忽略背景带来的干扰，实现快速又准确的汽车零件缺陷检测。利用本文模型对上汽提供的汽车零件外壁缺陷进行检测，实验结果表明，模型大小仅为15.9 MB，mAP为94.64%，检测每张图片时间为0.013 s，满足汽车工业上的速度和精度的需求。对比性研究表明，本文模型检测精度和速度以及大小较其他目标检测算法VGG-SSD、MobileNetv2-SSD、YOLO v3等有一定的提高和改善。

关键词: 有效通道注意力(ECA), 深度可分离卷积, 倒残差, 缺陷检测, SSD(Single Shot Multibox Detector)

Abstract: Automobile parts have great influence on the appearance, performance and safety of automobile. Due to the large number of automobile parts, small volume and high accuracy requirements, there are certain requirements for the accuracy and speed of parts detection. Using image processing technology, based on SSD model, the VGG module is replaced by deep separable convolution and linear bottleneck inverse residual structure. An effective attention mechanism ECA module is introduced to avoid dimensionality reduction. At the same time the computational complexity of the model parameters is reduced, and the channel is increased to improve the accuracy of the model. And this paper focuses on the image target, ignores the interference of the background to achieve fast and accurate defect detection of automotive parts. In addition, the proposed model in this paper is used to detect the outer wall defects of automobile parts provided by SAIC. The experimental results show that the size of the model is only 15.9 MB, the mAPis 94.64%, and the detection time of each image is 0.013 s, which meets the requirements of speed and accuracy in the automotive industry. Compared with other target detection algorithms such as VGG-SSD、MobileNetv2-SSD and YOLO v3, the detection accuracy, speed and size of the proposed model are improved.

Key words: effective channel attention, depth separable convolution, inverted residual, defect detection, SSD(Single Shot Multibox Detector)

金文倩, 彭露露, 朱媛媛, 王笑梅. 基于ECA-SSD模型的汽车零件缺陷检测[J]. 计算机与现代化, 2022, 0(03): 82-90.

JIN Wen-qian, PENG Lu-lu, ZHU Yuan-yuan, WANG Xiao-mei. Defect Detection of Automobile Parts Based on ECA-SSD Model[J]. Computer and Modernization, 2022, 0(03): 82-90.

参考文献［23］

［1］	LECUN Y, BOTTOU L, BENGIO Y, et al. Gradient-based learning applied to document recognition［J］. Proceedings of the IEEE, 1998,86(11):2278-2324.
［2］	GIRSHICK R. Fast R-CNN［C］// Proceedings of the 2015 IEEE International Conference on Computer Vision. 2015:1440-1448.
［3］	REN S Q, HE K M, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks［J］. IEEE transactions on Pattern Analysis and Machine Intelligence, 2016,39(6):1137-1149.
［4］	LIU W, ANGUELOV D, ERHAN D, et al. SSD: Single shot multibox detector［C］// European Conference on Computer Vision. 2016:21-37.
［5］	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection［C］// Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. 2016:779-788.
［6］	LIN T Y, DOLLR P, GIRSHICK R, et al. Feature pyramid networks for object detection［C］// Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. 2017:2117-2125.
［7］	FU C Y, LIU W, RANGA A, et al. DSSD: Deconvolutional single shot detector［J］. Computer Vision and Pattern Recognition, arXiv preprint arXiv:1701.06659, 2017.
［8］	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition［C］// Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. 2016:770-778.
［9］	BISWAS D, SU H B, WANG C Y, et al. An automatic traffic density estimation using single shot detection (SSD) and MobileNet-SSD［J］. Physics and Chemistry of the Earth, Parts A/B/C, 2019,110:176-184.
［10］	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition［J］. Computer Vision and Pattern Recognition, arXiv preprint arXiv:1409.1556, 2014.〖HJ1.15mm〗
［11］	HENRIQUES J F, CARREIRA J, CASEIRO R, et al. Beyond hard negative mining: Efficient detector learning via block-circulant decomposition［C］// proceedings of the 2013 IEEE International Conference on Computer Vision. 2013:2760-2767.
［12］	HOWARD A G, ZHU M L, CHEN B, et al. Mobilenets: Efficient convolutional neural networks for mobile vision applications［J］. Computer Vision and Pattern Recognition, arXiv preprint arXiv:1704.04861, 2017.
［13］	SZEGEDY C, IOFFE S, VANHOUCKE V, et al. Inception-v4, inception-resnet and the impact of residual connections on learning［C］// Proceedings of the 31st AAAI Conference on Artificial Intelligence. 2017:4278-4284.
［14］	SANDLER M, HOWARD A, ZHU M, et al. Mobilenetv2: Inverted residuals and linear bottlenecks［C］// Proceedings of the 2018 IEEE Conference on Computer Vision and Pattern Recognition. 2018:4510-4520.
［15］	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need［C］// Proceedings of the 31st International Conference on Neural Information Processing Systems. 2017:6000-6010.
［16］	BAHDANAU D, CHO K, BENGIO Y. Neural machine translation by jointly learning to align and translate［J］. Computation and Language, arXiv preprint arXiv:1409.0473, 2014.
［17］	JADERBERG M, SIMONYAN K, ZISSERMAN A. Spatial transformer networks［C］// Proceedings of the 28th International Conference on Neural Information Processing Systems. 2015:2017-2025.
［18］	HU J, SHEN L, SUN G. Squeeze-and-excitation networks［C］// Proceedings of the 2018 IEEE Conference on Computer Vision and Pattern Recognition. 2018:7132-7141.
［19］	WANG F, JIANG M Q, QIAN C, et al. Residual attention network for image classification［C］// Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. 2017:3156-3164.
［20］	WANG Q L, WU B G, ZHU P F, et al. ECA-Net: Efficient channel attention for deep convolutional neural networks［C］// Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2020:11534-11542.
［21］	NEUBECK A, VAN GOOL L. Efficient non-maximum suppression［C］// Proceedings of the 18th International Conference on Pattern Recognition. 2006,3:850-855.
［22］	EVERINGHAM M, WINN J. The Pascal Visual Object Classes Challenge 2012 (voc2012) Development Kit［EB/OL］. ［2021-05-18］. https://www.researchgate.net/profile/John-Winn-4/publication/267296489_The_PASCAL_Visual_Object_Classes_Challenge_2010_VOC2010_Development_Kit_Contents/links/55e8554308ae65b638997bc3/〖JP+1〗The-PASCAL-Visual-Object-Classes-Challenge-2010-VOC 2010-Development-Kit-Contents.pdf
［23］	MARITE CURITE ACTIONS. Pattern Analysis, Statistical Modelling and Computational Learning［EB/OL］.(2003-05-02)［2021-05-18］. http://lear.inrialpes.fr/people/triggs/projects/pascal-rtn/pascal-rtn-final.pdf.

[1]	宁娟, 周庆华, 曾小为. 改进YOLOv7算法在西林瓶轧盖缺陷检测中的应用[J]. 计算机与现代化, 2023, 0(12): 82-86.
[2]	张嘉琪, 徐啟蕾. 基于NAM-YOLO网络的苹果缺陷检测算法[J]. 计算机与现代化, 2023, 0(10): 53-58.
[3]	叶思佳, 魏延, 杜韩宇, 邓金枝. 结合注意力机制的HRNet图像语义分割算法[J]. 计算机与现代化, 2023, 0(10): 65-69.
[4]	刘昱杉, 刘卫康, 刘庆华, 者甜甜, 王家晨. 基于YOLOX结合DeepSort的船载车辆行人检测算法[J]. 计算机与现代化, 2023, 0(08): 60-67.
[5]	罗伟, 刘思远, 徐健祥, 董天培. 基于改进YOLOv5s的太阳能电池缺陷检测算法[J]. 计算机与现代化, 2023, 0(07): 119-126.
[6]	叶力鸣, 陈蔚文. 一种结合语义分割和目标检测的级联式绝缘子缺陷检测方法[J]. 计算机与现代化, 2023, 0(06): 82-88.
[7]	任秋霖, 任德均, 李鑫, 闫宗一, 曹林杰, 唐洪. 基于卷积自编码器的医用玻璃瓶口缺陷检测方法[J]. 计算机与现代化, 2022, 0(08): 114-120.
[8]	金鑫, 曾思轲, 刘阳, 武楚涵. 基于改进YOLOv4的口罩佩戴检测算法[J]. 计算机与现代化, 2022, 0(01): 85-90.
[9]	黄子平, 黄继风, 周小平. 基于YOLOv3的改进仪表检测算法[J]. 计算机与现代化, 2022, 0(01): 77-84.
[10]	周璇, 朱苏磊, 何为. 基于显著性目标检测的弹着点定位算法[J]. 计算机与现代化, 2022, 0(01): 54-60.
[11]	李传栋, 邱磊, 于雁. 基于改进残差密集网络的心律失常自动分类[J]. 计算机与现代化, 2021, 0(11): 106-111.
[12]	郭兆明, 周庆华, 曾小为. 基于双目视觉的缺陷药片检测[J]. 计算机与现代化, 2021, 0(11): 56-60.
[13]	程凯强, 张旭, 寇旭鹏. 基于SSD的小目标检测改进算法[J]. 计算机与现代化, 2021, 0(07): 77-82.
[14]	刘亚东, 罗印升, 曹阳阳, 宋伟. 深度学习在手机数据接口缺陷检测中的应用[J]. 计算机与现代化, 2021, 0(06): 35-40.
[15]	张磊, 郎贤礼, 王乐. 基于图像融合与YOLOv3的铝型材表面缺陷检测[J]. 计算机与现代化, 2020, 0(11): 8-15.