Fabric Defect Detection in Textile Manufacturing: A Survey of the State of the Art

Table 3

Dictionary learning algorithms for fabric defect detection.

Author	Proposed method	Dataset	Evaluation
Li et al. [52]	Low-rank representation (LRR)	(1) TILDA fabric images dataset; (2) dataset from the research associate of industrial automation research laboratory	Precision and recall
Li et al. [53]	Low-rank representation	500 fabric images from the textile kind C1 of the TILDA database	(a) Sensitivity and specificity; (b) false alarm rate (FAR), missing rate (MR)
Gao et al. [54]	Gabor filter and tensor low-rank recovery	Dataset from the research associate of industrial automation research laboratory	Receiver operating characteristic curve (ROC)
Shi et al. [39, 47]	Low-rank decomposition with gradient information	Dataset from the research associate of industrial automation research laboratory	TPR, FPR, PPV, NPV
Liu et al. [55–57]	Multi-scale convolutional neural network and low-rank decomposition model	(1) TILDA fabric images dataset; (2) dataset from the research associate of industrial automation research laboratory	Means and standard deviations of average precisions, recalls, F-measure, and mean absolute error (MAE)
Mo et al. [58]	Weighted double-low-rank decomposition method (WDLRD) to treat the matrix singular values differently by assigning different weights	Database is from the research associate of industrial automation research laboratory, HKBU	Visual defect locating results, the metrics of false alarm, recall, precision, accuracy, and F-measure
Li et al. [59]	Low-rank decomposition of multichannel feature matrices	(1) TILDA fabric images dataset; (2) dataset from the research associate of industrial automation research laboratory	ROC curves and precision-recall (PR) curves
Yang et al. [60]	Sparse and dense mixed low-rank decomposition	Real-world samples of 512∗512 with 256-gray levels	Saliency map (qualitative)
Wang et al. [61]	A randomized low-rank and sparse matrix decomposition model named GoDec	Fabric image dataset collected by Dr. Henry Y. T. Ngan [62]	Precision, recall, and F-measure