Warp-knitted Fabric Defect Online Detecting System Based On Multiscale Geometric Analysis

Due to the features of high efficiency, high quality and diversity, Warp knitting plays an important role in the fields of clothing industry, decoration industry and industrial application. However as the rapid development of warp knitting industry, the market competition is increasingly fiercer than ever before and customers are demanding of the product quality. And that warp-knitted fabric defect detection is the key step of quality control, because the artificial detection is time-consuming and power-wasting, the warp-knitted fabric defect automatic detection is needed urgently. In order to achieve the warp-knitted fabric defect online detection, this task carried out a series of research about it.Firstly, form the wavelet transform, the one-dimensional mathematic model and twodimensional mathematic model were discussed. In according to the function expansion, the disadvantage of sparse Approximation to singular curves in high dimension function was indicated. Meanwhile this research introduced the multi-scale geometric analysis and put the ideal method contourlet transform forward. Moreover, the mathematic structure(Laplacian Pyramid and directional filter bank) of the contourlet transform was described.Secondly, the shortage of the contourlet transform is discussed and summarized while improved the traditional transform by means of the wavelet-packet transform and nonsubsampled directional filter bank. The selection of wavelet function of the transform and the decomposition level was discussed, and through the best wavelet-packet decomposition, segmented threshold de-noise and directional sub-band decomposition, the original picture was operated. In addition, the image was reconstructed by regional energy selection and the segmented image was achieved by adaptive threshold segmentation and morphological processing.Then, the image features of original image, reconstructed image and segmented image were attained by the methods, such as gray-level co-occurrence matrix. In order to simplify the number of image features and reduce the operation difficulty, the principal component analysis was applied to acquire the primary features. Meanwhile the selected features were set as the inputs of the 3-level BP neural network designed in this task. Through the training and testing of the four kinds of warp-knitted fabric defects, the network could achieved the recognition of 96%.Finally, this task designed the hardware frame of warp-knitted fabric defect online detection system and sated the operation principle of the primary hardware components and the selection requirements. According to the research above, the algorithm flow of the system was elaborated. And the software part of the system was accomplished by MATLAB GUI and the function of the algorithm was achieved.The research of warp-knitted fabric defect online detection based on non-subsampled wavelet-packet contourlet transform in this task can provide the reference for the future improved work and the related research in this area.