Research Of Fabric Defect Detection Based On Deep Feature And Low-rank Decomposition

The effective automatic fabric defect detection plays an important role in the quality control and evaluation of textiles.Traditional artificial detection methods have the lower success rate and are easily influenced by subjective factors,resulting in the undesirable detection effect.The machine-vision-based detection methods have become a research focus.Deep learning,which has the multiple nonlinear transformation characteristics and the ability to multi-level image abstraction,can effectively characterize complex texture of fabric image.Therefore,convolutional neural networks(CNNs)is applied to fabric defect detection.In addition,low-rank decomposition model can realize the separation of fabric background and saliency defect.On the basis of CNN feature extraction and the characteristics of low-rank and sparse of human eyes,we research on the fabric defect detection methods based on deep learning and low-rank decomposition.Aiming at the complexity of the fabric texture and combing with the advantages of self-learning ability in CNN and low-rank recovery technology,a fabric defect detection algorithm based on convolutional neural network and low-rank recovery is proposed.First,the fabric image library is pre-trained with the initial network parameters based on MNIST datasets.The supervised fine-tuning is implemented to obtain the optimal network parameters,and then more accurate deep neural network model is generated.Second,the fabric images are uniformly divided into the image block with the same size,then we extract their multi-layer deep features using the trained deep network.Thereafter,all the extracted features are concentrated into a feature matrix.Third,low-rank recovery model is established and ADMM algorithm is adopted to divide the feature matrix into the low-rank matrix which indicates the background and the sparse matrix which indicates the salient defect.In the end,the iterative optimal threshold segmentation algorithm is utilized to segment the saliency maps generated by the sparse matrice tolocate the fabric defects.Based on the problems of the discontinuous defect region and the slower network training,we can improve the network structure,as well as the reconstruction and optimization of low-rank decomposition model.A fabric defect detection algorithm based on multi-scale convolutional neural network and low-rank representation is proposed.On the basis of the characteristics of CNN weight sharing,the weights of the lower level image blocks are shared to the advanced large image for training,so as to accelerate the training speed.Secondly,based on the ability of multi-scale CNN to learn the detailed features of image texture automatically,the deep features of fabric defects on different channels are extracted and fused to generate the feature matrix.A suitable low-rank representation model is constructed,and then the ALM method is adopted to optimize the solution to generate low-rank matrix and sparse matrix.Finally,the optimal threshold segmentation algorithm is utilized to segment the defect map produced by sparse matrix.Thus,the defective area of the fabric image is located accurately.