Context receptive field and adaptive feature fusion for fabric defect detection

Textile defect detection is a key link in measuring textile quality. An accurate and efficient textile detection method is essential for textile production processes. In the textile production process, the problem of large variation in textile defect shape scale and edge uncertainty makes textile defect detection very difficult. We propose a textile defect detection method based on context receptive field and adaptive feature fusion. Firstly, we introduce an improved context receptive field block (CRFB) in the backbone network CSPDarknet53. It makes full use of local and global context information to enhance the extraction capability of the backbone network for textile defect features. Secondly, a deconvolution-based adaptive feature fusion network (DAFF) is designed to improve the transfer efficiency of shallow localization information and feature scale invariance. Finally, exponential distance IoU (EDIoU) is proposed to optimize the calculation of bounding box loss and adaptively weighting the bounding box gradient to improve the detection accuracy of the model. The experimental results on ZJU-Leaper and Tianchi datasets show that the mAP of this paper reaches 42.5% and 61.5%, which are 2.9% and 3% higher than the baseline YOLOv5s, respectively. Meanwhile, the model in this paper outperforms most existing networks in terms of detection accuracy and speed.

Automated summary

This study proposes a textile defect detection method based on context receptive field and adaptive feature fusion. By improving the backbone network and designing an adaptive feature fusion network, this method achieves good performance in textile defect detection.