Automatic Fabric Defect Detection And Classification Based On Machine Vision

The defect of the fabric is the main factor for the quality of the fabric products. Thetraditional fabric defect detection is mainly done manually, but manual inspection has somedrawbacks. It consumes lots of manpower and damages human eyesight. In addition,subjective aspiration, emotion, and long-term work which cause human fatigue and boredominfluence the efficiency of the manufacturing process. In order to perform the task of humaninspector and overcome their weakness, it is necessary to propose a rapid, efficient, reliable,and real-time automatic detection system, to replace the traditional fabric defect detection.Recently, machine vision has been more widely used in the field of industrial surfaceinspection, then, automatic fabric defect detection system for the research and development ofmachine vision has great significance. This subject is mainly to research the algorithm offabric defect detection and classification. The work is shown as follows:(1) According to the fabric defect detection method, this thesis proposed the algorithmthat is based on the improved Gabor filter fabric defect detection. For the fabric defect texture,the distortion of fabric feature is mainly prominent in the directional characteristics. Defecttexture is different from the normal texture in which it mainly has different directionalcharacteristics, so it is suit to use multi-scale and multi-directional Gabor filter. This thesisused four scales, six directions Gabor filter for fabric defect detection algorithm, then, usedthe cost evaluation function to select the most optimal filtering for output, and finally doimage binarization of the optimal filter output, and the threshold for defect image binarizationis from the normal sample obtained. The improved Gabor filter can highlight the fabric defectof the scale of the change preferably. The algorithm can accurately locate the defects positionfor7types of defects mentioned in this thesis and the detection effect is better.(2) To solve the problem of automated fabric defect detection and improve the quality offabric in the production, this thesis presented an approach which was based on multi-scalewavelet transform and Gaussian mixture model. Firstly, the sample images were tackled bythe multi-scale wavelet decomposition algorithm, then, to reconstruct the new image withwavelet thresholding denoising via the produced wavelet coefficients. Secondly, the obtainednew images were segmented by applying the Gaussian mixture model that based the EMalgorithm. Various fabric samples were used in the evaluation and the experimental resultsshowed the designed algorithm could precisely locate the position of defect and segment thedefect. The algorithm is better than the improved Gabor filter fabric defect detection.(3) Due to the characteristics of single often cannot effectively describe the fabric defectfeatures for the fabric defect classification, an algorithm based on local binary pattern and Tamura texture features which combined for fabric defect classification was proposed. Themain task of the algorithm was to extract feature vectors fabric; local binary pattern describedthe texture feature changing from the local, while Tamura texture features method describedthe texture feature changing from the global. A good description of defect can be got bycombining the two methods. After extracting feature vector, used the conjugate gradient BPalgorithm to handle the feature vectors. The convergence of conjugate gradient BP algorithmwas better in the training speed and training accuracy. The experiments showed that theproposed algorithm for defect classification had higher accuracy.(4) This thesis built the fabric defect automatic detection system which was based onmachine vision. It expounds the system hardware composition, the hardware platform designand build, and its working principle.