基于DCGAN和改进YOLOv5的织物疵点检测算法  

Fabric defect detection algorithm based on DCGANand improved YOLOv5 Model

作　　者：张天宇 李子熠 李鸿强[1] ZHANG Tianyu;LI Ziyi;LI Hongqiang(Hebei University of Architecture,Zhangjiakou,Hebei 075000)

机构地区：[1]河北建筑工程学院,河北张家口075000

出　　处：《河北建筑工程学院学报》2024年第1期222-229,共8页Journal of Hebei Institute of Architecture and Civil Engineering

摘　　要：针对传统的织物疵点检测算法的精度低、速度慢、漏检率高等难点,提出了基于深度卷积生成式对抗网络和改进YOLOv5的织物疵点检测算法。首先,将公开的天池阿里云织物疵点图像通过DCGAN网络进行数据增强,建立各类疵点样本较为完善的数据集;其次,为了提高模型的检测精度,在YOLOv5模型上引入CBAM注意力机制模块,让模型检测更能聚焦在缺陷区域,从而降低漏检率;再次,将Mish激活函数替换Leaky ReLU函数,增强模型的泛化能力。最后通过改进模型与原模型的对比实验,得出结论,提出的模型相比其他深度学习模型具有更好的检测度和鲁棒性。In view of the difficulties of traditional fabric defects detection algorithms such as low precision,slow speed and high missing rate,a fabric defects detection algorithm based on deep convolution generative adversation network and improved YOLOv5 was proposed.Firstly,the public fabric defects images of Tianchi Alibaba Cloud were enhanced through DCGAN network to establish a relatively complete data set of all kinds of fabric defects samples.Secondly,in order to improve the detection accuracy of the model,CBAM attention mechanism module is introduced into the YOLOv5 model to make the model detection more focused on the defect area,so as to reduce the missed detection rate.Thirdly,the Mish activation function was replaced with the Leaky ReLU function to enhance the generalization ability of the model.Finally,through the comparison experiment between the improved model and the original model,it is concluded that the model proposed in this paper has better detection and robustness than other deep learning models.

关 键 词：织物疵点检测 YOLOv5 深度卷积生成式对抗网络 CBAM 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]