Fabric Defect Detection Algorithm Based on Convolution Neural Network and Low-Rank Representation

Abstract

To accurately detect the fabric defects in the textile quality control process, this paper proposed a novel detection method based on convolution neural network(CNN) and low-rank representation(LRR). First, the characteristics of multiple nonlinear transformations and multi-level abstraction ability of images in deep learning are used to characterize the multi-layer features of fabric images using CNN, and then the extracted features are concentrated into a feature matrix. Second, low-rank representation model is adopted to divide the feature matrix into low-rank and sparse matrices, which indicate the background and salient object defects, respectively. Finally, the iterative optimal threshold segmentation algorithm is used to segment the saliency maps generated by the sparse matrix to locate the fabric defect region. Experimental results show that the features extracted by CNN are more suitable for characterizing fabric texture than traditional methods, such as HOG, LBP, and other hand-crafted feature extraction method, and the detection results outperforms the state-of-the-art.