Investigating the Effect of Attention Module Positioning on Fabric Defect Detection Performance Using the Convolutional Block Attention Module

Abstract

The detection of fabric defects is a critical step in ensuring product quality within the textile industry. However, existing object detection methods often struggle to effectively capture the subtle features of narrow fabric defects (such as narrow yarn defects), resulting in decreased detection accuracy. To address this challenge, an improved object detection method is proposed in this paper, achieved by embedding a Convolutional Block Attention Module (CBAM) into an advanced Faster R-CNN network, thereby enhancing the ability to capture the characteristics of narrow defects. Furthermore, the impact of different CBAM embedding positions was investigated in order to optimize the maximum contribution of CBAM to fabric defect detection. Experimental results, based on a 6317-sample fabric defect dataset, demonstrate that the proposed method achieved a maximum improvement of 2.6% in mean Average Precision (mAP), and a maximum improvement of 5.97% in AP for narrow yarn defects. The findings of this research are promising in offering a possible solution for efficient defect detection within textile manufacturing.