计算机应用

• 人工智能与仿真 •    下一篇

基于深度特征融合的无纺布低对比度浆丝缺陷检测方法

鲁永帅,唐英杰*,马鑫然   

  1. 北京印刷学院 信息工程学院,北京102600
  • 收稿日期:2021-05-21 修回日期:2021-09-09 发布日期:2021-09-09 出版日期:2021-10-29
  • 通讯作者: 唐英杰

Low contrast pulp filament defect detection method of non-woven fabric based on deep feature fusion

  • Received:2021-05-21 Revised:2021-09-09 Online:2021-09-09 Published:2021-10-29

摘要: 针对无纺布生产过程中产生的浆丝缺陷对比度较低、传统图像处理方法检测效果较差的问题,提出了一种基于卷积神经网络的无纺布低对比度浆丝缺陷检测方法。首先,对采集到的无纺布图像进行预处理,构建浆丝缺陷数据集;然后,利用改进的卷积神经网络以及多尺度特征采样融合模块构造编码器提取低对比度浆丝缺陷语义信息,在解码器中采用跳跃连接进行多尺度特征融合优化上采样模块;最后,通过构建的数据集训练网络模型,实现低对比度浆丝缺陷的检测。实验结果表明,所提方法可以有效定位并检出无纺布上的低对比度浆丝缺陷,其算法平均交并比、类别平均像素准确率能够达到77.32%、86.17%,单张样本平均检测时间为50ms,能够满足工业生产的要求。

关键词: 无纺布, 低对比度, 浆丝缺陷, 语义分割, 深度学习

Abstract: In order to solve the problems of low contrast of pulp filament defects in non-woven fabric production process and poor detection effect of traditional image processing methods, a new low contrast pulp filament defect detection method of non-woven fabrics based on convolutional neural network was proposed. Firstly, the collected non-woven fabric images were preprocessed to construct the defect data set of pulp filament. Then, an improved convolutional neural network and multi-scale feature sampling fusion module were used to construct an encoder to extract the semantic information of low contrast of pulp filament defects, and a skip connection was used in the decoder to achieve multi-scale feature fusion for optimizing the upsampling module. Finally, the low contrast defect detection of pulp filament was realized by training the network model of the constructed data set. The experimental results show that, the proposed method can effectively locate and detect the low contrast pulp filament defects on non-woven fabrics. The mean intersection over union and category mean pixel accuracy of the proposed method can reach 77.32% and 86.17%, and the average detection time of single sample is 50 ms, which can meet the requirements of industrial production.

Key words: non-woven fabric, low contrast, pulp filament defect, semantic segmentation, deep learning

中图分类号: