Fabric defect detection via a spatial cloze strategy

Abstract

Deep-learning models have achieved state-of-the-art performances in a wide range of defect detection tasks. However, an inescapable criticism of one-stage fully supervised models is the lack of interpretability, which not only reduces the reliability of fabric defect detection systems but also limits the scope of their applications in production environments. To tackle the data imbalance and low interpretability of defect samples, we proposed a spatial cloze strategy for fabric defect detection, which reconstructs a local normal image and then feeds it into the detection model with the original image simultaneously. Specifically, we formulate the defect detection task as a novel image completion problem. Firstly, an end-to-end deep neural network is trained to finely restore the defect image by completing each image slice removed in sequence. Next, the progressive attention mechanism fuses the repaired normal image with the raw image, replacing the input layer of the cascade region-based convolutional neural network. Eventually, accurate instance-level defect segmentation can be obtained by comparing the repaired defect-free and the raw images. On the Tianchi dataset, the proposed method displays superior accuracy in 92% of defect classes, with a breakthrough in various categories that have hardly ever been detected. Extensive experiments on various complex fabric defect samples demonstrate that our strategy outperforms existing advanced methods.

A Lightweight Detector Based on Attention Mechanism for Fabric Defect Detection

“…As observed from the table, the present experimental results were 5.62% more accurate than those reported by the first-place winner of the Tianchi industrial competition. For validation, we compared the current experimental results with those obtained by Liu et al [23,24,32], who applied the fabric defect dataset of Tianchi. As observed, the present model displayed adequate results for individual defects, and the detection accuracy of large defects such as stain, hundred feet, and other difficult-to-detect defects such as abrasion mark were optimal compared to other studies.…”

“…Lu et al [24] proposed a fabric defect detection method based on a C-RCNN for image segmentation, but the segmentation results were inaccurate if the fabric defects were extremely small or similar.…”

Real-Time Defect Detection Model in Industrial Environment Based on Lightweight Deep Learning Network

“…The experimental datasets in this paper include PCB [18], NEU-DET [19], Tianchi fabric [20], and Tianchi ceramic tile [21]. PCB dataset: This dataset consists of 1500 PCB images, covering six types of PCB defects.…”

YOLOvT: CSPNet-based attention for a lightweight textile defect detection model

“…Zhang et al 15 proposed an attention-gated U-Net for unsupervised fabric defect detection, using attention-gated weighted residual fusion to combine decoder and encoder features, and achieving a 54.31% F1 score on the YDFID-1 dataset. Lu et al 16 performed partial defect removal and restoration on defective fabric images, generating defect-free images, and then inputting both the restored and original images into a detection network for defect detection and segmentation. Liu et al 17 introduced a multi-scale feature aggregation unit and feature fusion refinement module to effectively represent multi-scale background features, strengthening the complementary relationships between different features, and demonstrated excellent performance on their self-made dataset.…”

“…Lu et al. 16 performed partial defect removal and restoration on defective fabric images, generating defect-free images, and then inputting both the restored and original images into a detection network for defect detection and segmentation. Liu et al.…”