基于改进YOLOv5s 的经编织物缺陷检测. (Chinese)

Aiming at the problems of low detection accuracy and high rate of missed detection and false detection of warp knitted fabrics，an improved YOLOv5s algorithm model CSC-YOLOv5s was proposed. Firstly， ConvNeXtBlock module was used to replace CSP module in the backbone extraction network to enhance the feature extraction capability of the backbone network and reduce loss of detailed information during feature extraction. Secondly，SC-PANet network structure was proposed and SimAM attention mechanism module was used to enhance the model attention to the defect area of the warp knitted fabric during feature fusion，and to improve the detection accuracy of small defects，combined with the Content-Aware ReAssembly upsampling operator to improve the upsampling layer and improve the precision of feature fusion. Finally，loss function was improved to accelerate the model convergence. The experimental results showed that the mAP value of the CSC-YOLOv5s algorithm on the self-built warp knitted fabric data set was 90.6%，the recall rate was 85.9%，which was 5.5 percentage points and 5.9 percentage points higher than the original YOLOv5s algorithm，the overall performance of the improved algorithm was better. [ABSTRACT FROM AUTHOR]

针对经编织物缺陷检测精度低、漏检和误检率高等问题，提出改进的YOLOv5s 算法模型CSC⁃ YOLOv5s。首先使用ConvNeXtBlock 模块替换主干提取网络中的CSP 模块，增强主干网络的特征提取能力，减 少特征提取时细节信息的丢失;其次提出SC⁃PANet 网络结构，引入SimAM 注意力机制模块，增强模型在特征 融合时对经编织物缺陷区域的关注，提高小尺寸缺陷的检测精度，结合Content⁃Aware ReAssembly 上采样算子 改进上采样层，提升特征融合精度;最后改进损失函数，加速模型收敛。试验结果表明，CSC⁃YOLOv5s 算法在自 建经编织物数据集上mAP 值为90.6%、召回率为85.9%，比原始YOLOv5s 算法分别提高5.5 个百分点和5.9 个 百分点，改进后的算法整体性能较好. [ABSTRACT FROM AUTHOR]

Copyright of Cotton Textile Technology is the property of Cotton Textile Technology Editorial Office and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)