Automatic Fabric Defect Detection Using Cascaded Mixed Feature Pyramid with Guided Localization

Abstract

Generic object detection algorithms for natural images have been proven to have excellent performance. In this paper, fabric defect detection on optical image datasets is systematically studied. In contrast to generic datasets, defect images are multi-scale, noise-filled, and blurred. Back-light intensity would also be sensitive for visual perception. Large-scale fabric defect datasets are collected, selected, and employed to fulfill the requirements of detection in industrial practice in order to address these imbalanced issues. An improved two-stage defect detector is constructed for achieving better generalization. Stacked feature pyramid networks are set up to aggregate cross-scale defect patterns on interpolating mixed depth-wise block in stage one. By sharing feature maps, center-ness and shape branches merges cascaded modules with deformable convolution to filter and refine the proposed guided anchors. After balanced sampling, the proposals are down-sampled by position-sensitive pooling for region of interest, in order to characterize interactions among fabric defect images in stage two. The experiments show that the end-to-end architecture improves the occluded defect performance of region-based object detectors as compared with the current detectors.

A Robust Fabric Defect Detection Method Based on Improved RefineDet

“…The raw images used for recent research were mainly from public datasets or collected by textile factories and laboratories. Some typical public defect detection datasets are TILDA dataset ( ), DAGM2007 dataset ( ), and Hong Kong patterned texture database ( ); and some self-built datasets are DHU-FD-500 [ 7 ], DHU-FD-1000 [ 7 ], lattice [ 8 ], FDBF dataset [ 19 ], etc. Image preprocessing.…”

“…[17] applied some data augmentation methods and soft NMS to further improve the performance of the model, and [18] combined the high-level features of the ROI pooling layer output with the low-level features obtained by the Histogram of Oriented Gradient (HOG) in the original Faster RCNN. Wu et al [19] designed a Composite Interpolating Feature Pyramid Network (CI-FPN) as the neck structure and introduced a guided anchor mechanism and position-sensitive RoI-Align in head structure. Liu et al [20] introduced SSD to the defect detection for the first time and added the third-level feature conv3_3 to the feature pyramid to achieve the detection of small objects.…”

“…These feature pyramid structures are roughly divided into the following categories [ 24 , 25 ]: (1) SSD-style, (2) FPN-style, (3) STDN-style, (4) M2Det-style, (5) PAN-style. Some recent studies have made improvements on the neck structure and achieved a good detection result [ 19 , 26 ]. Flexible selection of the model.…”

“…PAN-style. Some recent studies have made improvements on the neck structure and achieved a good detection result [19,26].…”

“…Some typical public defect detection datasets are TILDA dataset (https://lmb.informatik.uni-freiburg.de/resources/datasets/ tilda.en.html), DAGM2007 dataset (https://hci.iwr.uni-heidelberg.de/content/weakly-supervisedlearning-industrial-optical-inspection), and Hong Kong patterned texture database (https://ytngan. wordpress.com/codes/); and some self-built datasets are DHU-FD-500 [7], DHU-FD-1000 [7], lattice [8], FDBF dataset [19], etc.…”