Research On Some Theoretical Problems Of Defect Detection Using Gabor Wavelet And FT Method

Increasing the accuracy and efficiency of fabric defect identification,it is of great significance to improve the quality of textile products.Aiming at the problem related to uneven illuminating and low contrast of image defect image,the method of defect image enhancement is studied based on Gabor wavelet.First,15 Gabor filters with 3 scales and 5 directions are used to filter the defect images in difference directions and scales,which can reduce the influence of uneven illuminating and low contrast on the accuracy of feature extraction.Then,the filtered image is divided into non overlapping patches with the same area.The dimensional feature vectors are extracted from the patchers.Aiming at the dimensionality reduction problem of Gabor feature vector,Isometric mapping method is employed to reduce the dimension of Gabor feature vector.The reduction of feature vector dimension contribute to eliminating redundant information,enhancing fitting ability of the classifer,and strengthening sensitivity of feature.Secondly,in the light of the problem relating to selection of the structure parameters of Isomap,Ncut standards is applied to establish the structural parameter optimization model as fitness fuction.PSO algorithm is used to search optimal parameters.A new method of parameter optimization using PSO and Ncut criterion is proposed.To enhance the efficiency of defect detection,improved frequency-tuned algorithm is introduced to detect the fabric defect substitute for Gabor filters.The frequency-tuned algorithm has the advantages of high stability in noise conditions and easy to implement.Aiming at the problem related to the size of Gauss filter in frequency-tuned algorithm,the method of discrete particle swarm optimization is applied to select parameter adaptively,and use the Ncut criterion as the fitness function.Aiming at the problem of poor effect related to Lab color space,HSV color space is introduced to replace Lab color space.However,the range of each channel in the HSV color are different,which lead to a poor salient result.A normalization model is applied to solve this problem.Then,we extract features using GLCM method.Finally,PNN is applied to detect the fabric defects.The method of fabric defect detection based on Gabor filters is introduced to compare with the method of fabric defect detection based on improved frequency-tuned algorithm.From the experimental results,we could know that the ability of detect fabric defect based on improved frequency-tuned algorithm is as well as the method of the method based on Gabor filters.Compared with themethod of fabric defect detection based on Gabor filters,the time spends on fabric defect detection based on improved frequency-tuned algorithm was shortened 70%.