Fabric Defect Detection Based On Energy Residual Distribution And Gabor Feature Fusion

Fabric defect detection is a very important part of textile quality control and inspection.However,the traditional artificial defect detection method can not guarantee the detection efficiency and accuracy.Therefore,using image processing technology to realize automatic detection of fabric defects and further improve the detection accuracy and efficiency has become an important research direction at present.Among many algorithms,the time-frequency method is effective for analyzing the non-stationary signals such as fabric defects.Two-dimensional Gabor filter is an important time-frequency joint analysis method,which is very suitable for detecting embedded local anomalies in periodic textures.However,this kind of algorithm also faces many difficulties:(1)The current algorithm lacks effective means when constructing Gabor filter,and the direction and scale parameters selected by experience are difficult to apply to the changeable background texture and defect types.(2)Algorithms based on Gabor filter mainly include optimal channel method and multi-channel fusion method.Compared with the optimal channel method,the multi-channel fusion method can obtain complete image features,but it has the problems of feature redundancy and poor anti-noise ability.(3)There are many important parameters in the Gabor based algorithm,so automatic parameter optimization is needed.To solve these problems,a fabric defect detection algorithm based on energy residual distribution and Gabor feature fusion is proposed in this paper.By constructing a set of relatively complete Gabor filters,the Gabor features of the detection image and template image under each channel were extracted respectively,and the residual energy between them was calculated.The maximum mean ratio(MMR)was used to measure the significance of fabric defects in each channel relative to the background texture.Finally,the nonlinear normalization was used to calculate the energy weight under each channel.In this way,Gabor features under multiple channels are integrated.In addition,for some parameters involved in the above algorithm that affect the fusion result,we use the signal-to-noise ratio and genetic algorithm to optimize the parameters.Experiments show that the proposed algorithm has better detection and location effects for different types of defects,especially for smaller color spot defects and less obvious stains.The main research contents of this paper are as follows:(1)In view of the problem that the direction and scale parameters selected by experience are difficult to apply to the changeable background texture and defect types,this paper constructs a set of relatively complete real Gabor filters to extract multi-channel Gabor features.By increasing the number of filters,the method can adapt to different fabric backgrounds and different types of defects,and improve the universality of the algorithm.(2)In view of the difficulty in balancing the redundancy and loss of Gabor features in current channel selection methods,this paper proposes a fabric defect detection algorithm based on the fusion of energy residual distribution and Gabor features.Based on the idea of weighted fusion,the maximum mean ratio(MMR)was used to measure the significance of defects relative to the background,so as to obtain the energy residual distribution under multi-channel,and the nonlinear normalization was used to calculate the energy weight matrix under different channels,so as to reduce the redundancy of Gabor features and optimize the effect of feature fusion.(3)Aiming at the problem that some trainable parameters in the fusion algorithm affect the quality of fusion results,this paper proposes a parameter optimization algorithm based on the signal-to-noise ratio and genetic algorithm.This method is based on two criteria for good fusion results: preserving defective features as much as possible and suppressing background noise.An evaluation function based on signal-to-noise ratio is designed,and the parameters of fusion algorithm are optimized by genetic algorithm.(4)Based on the above algorithm,this paper designs and implements a multilayer fabric defect detection system.Based on fabric defect detection and algorithm parameter optimization algorithm,the system further realizes the rapid detection of fabric defect through the cooperation of software and hardware platforms.