基于改进Faster_RCNN模型的织物瑕疵目标检测算法
Fabric Defect Target Detection Algorithm Based on Improved Faster_RCNN Model
摘要: 采用卷积神经网络技术来取代传统的人力织物瑕疵检测，可大大降低织物瑕疵的发现和修复的成本，从而极大地改善生产的效率和质量。本文提出了用n × n网格代替原始Faster_RCNN模型中的RPN层，并使用FPN结构调整网格尺寸，最终生成较少的候选区域，以减少对于后续的数据传递以及处理的负担，节省检测的时间。用已经收集整理好的织物瑕疵数据集进行模型的训练、学习，最后将经过多次训练以及参数调整的模型与当前通用的目标检测模型在相同条件下进行织物瑕疵目标检测实验对比。试验表明使用n × n网格代替RPN的改进Faster_RCNN模型织物瑕疵目标检测的总体map值为74.1%，单张图片的检测速度为230.7 ms，相比于原始Faster_RCNN模型检测精度稍稍下降，但检测速度提高了约4倍。说明用n × n网格代替RPN层可以在保证一定织物瑕疵目标检测精度的前提下，加快了检测的效率。
Abstract: The use of convolutional neural network technology to replace the traditional manual fabric defect detection can greatly reduce the cost of fabric defect detection and repair, thus greatly improving the efficiency and quality of production. In this paper, an n × n mesh is proposed to replace the RPN layer in the original Faster_RCNN model, and the FPN structure is used to adjust the mesh size. Fi-nally, fewer candidate regions are generated to reduce the burden of subsequent data transmission and processing, and save the detection time. The collected fabric defect data set was used to train and learn the model. Finally, the fabric defect target detection experiment was compared between the model after multiple training and parameter adjustment and the current common target detec-tion model under the same conditions. Experiments show that the overall map value of the im-proved Faster_RCNN model for fabric defect target detection by using n × n mesh instead of RPN is 74.1%, and the detection speed of a single image is 230.7 ms. Compared with the original Fast-er_RCNN model, the detection accuracy is slightly reduced, but the detection speed is increased by about 4 times. It shows that using n × n mesh instead of RPN layer can speed up the detection effi-ciency under the premise of ensuring a certain accuracy of fabric defect target detection.
文章引用：张天鹏, 王会敏. 基于改进Faster_RCNN模型的织物瑕疵目标检测算法[J]. 应用数学进展, 2023, 12(5): 2593-2602. https://doi.org/10.12677/AAM.2023.125260

参考文献

[1] Shi, M., Fu, R., Guo, Y., et al. (2011) Fabric Defect Detection Using Local Contrast Deviations. Multimedia Tools & Applications, 52, 147-157. [Google Scholar] [CrossRef] 
[2] Shi, M., Fu, R., Huang, S., et al. (2009) A Method of Fabric Defect Detection Using Local Contrast Deviation. Proceedings of 2009 2nd International Congress on Image and Signal Processing, Tianjin, 17-19 October 2009, 1-5. [Google Scholar] [CrossRef] 
[3] Xia, D., Jiang, G., Li, Y. and Ma, P. (2016) Warp-Knitted Fabric Defect Segmentation Based on Non-Subsampled Contourlet transform. Journal of the Textile Institute Proceedings & Abstracts, 108, 239-245. [Google Scholar] [CrossRef] 
[4] Tolba, A.S. (2011) Fast Defect Detection in Homogeneous Flat Surface Products. Expert Systems with Applications, 38, 12339-12347. [Google Scholar] [CrossRef] 
[5] Li, M., Cui, S. and Xie, Z. (2015) Application of Gaussian Mix-ture Model on Defect Detection of Print Fabric. Journal of Textile Research, 36, 94-98.
[6] Allili, M.S., Baaziz, N. and Mejri, M. (2014) Texture Modeling Using Contourlets and Finite Mixtures of Generalized Gaussian Distributions and Applications. IEEE Transations on Multimedia, 16, 772-784. [Google Scholar] [CrossRef] 
[7] 杨晓波. 基于GMRF模型的统计特征畸变织物缺陷识别[J]. 纺织学报, 2013, 34(4): 137-142.
[8] Krizhevsky, A., Sutskever, I. and Hinton, G.E. (2017) ImageNet Classification with Deep Convolutional Neural Networks. Communications of the ACM, 60, 84-90. [Google Scholar] [CrossRef] 
[9] Ren, S., He, K., Girshick, R., et al. (2017) Faster R-CNN: Towards Re-al-Time Object Detection with Region Proposal Networks. IEEE Transactions on Pattern Analysis and Machine Intelli-gence, 39, 1137-1149. [Google Scholar] [CrossRef] 
[10] Liu, W., Anguelov, D., Erhan, D., et al. (2016) SSD: Single Shot MultiBox Detector. In: Leibe, B., Matas, J., Sebe, N. and Welling, M., Eds., ECCV 2016: Computer Vision-ECCV 2016, Lecture Notes in Computer Science, Vol. 9905, Springer, Cham, 21-37. [Google Scholar] [CrossRef] 
[11] Redmon, J. and Farhadi, A. (2018) YOLOv3: An Incremental Improvement. Computing Research Repository. ArXiv abs/1804.02767.
[12] Jia, J., Jing, J., Zhang, H., et al. (2012) Fabric Defect Detection Using Gabor Filter and Defect Classification Based on LBP and Tamura Method. Journal of the TextileInstitute, 104, 18-27. [Google Scholar] [CrossRef] 
[13] Lin, T., Dollar, P., Girshick, R., et al. (2017) Feature Pyramid Networks for Object Detection. Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, 21-26 July 2017, 936-944. [Google Scholar] [CrossRef] 
[14] He, K., Zhang, X., Ren, S. and Sun, J. (2016) Deep Residual Learn-ing for Image Recognition. Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, 26 June-1 July 2016, 770-778.
[15] Xu. Y.Z., Yu, G.Z., Wang, Y.P., Wu, X. and Ma, Y. (2017) Car De-tection from Low-Altitude UAV Imagery with the Faster R-CNN. Journal of Advanced Transportation, 2017, Article ID: 2823617. [Google Scholar] [CrossRef] 
[16] Han, S., Liu, X., Mao, H., et al. (2016) EIE: Efficient In-ference Engine on Compressed Deep Neural Network. Proceedings of 2016 ACM/IEEE 43rd Annual International Sym-posium on Computer Architecture (ISCA), Seoul, 18-22 June 2016, 243-254. [Google Scholar] [CrossRef] 

为你推荐