Research On Lining Defect Detection System Based On Machine Vision

The lining cloth is the core of the fabric,and the quality of the lining defect detection is directly related to the quality level of the fabric.The traditional lining defect detection is to observe whether the lining has defects when manually lining the lining.This method is not only inefficient but also cannot guarantee the quality.In order to overcome the shortcomings of lining defect detection,this paper detects the defects by machine vision method and classifies the defect types by machine learning algorithm.First of all,in order to integrate with the traditional rolling machine,and not change the original machine structure,this thesis designed a machine vision hardware acquisition platform.According to the actual requirements,this thesis selects the appropriate hardware equipment and builds a visual system for the lining defects.Secondly,The characteristics of the lining's defects are tested.The existing three algorithms are improved.The defects are quickly judged based on multi-scale mean value,the defect detection based on gray level co-occurrence matrix,and the defect detection based on Gabor channel fusion.Among them,the defects are quickly judged based on the multi-scale mean value.Because of the defect area will destroy the distribution of the cloth texture,the mean value of the statistical lining is used,and the mean value statistics on different scales are analyzed,and the probability of the defect is determined by statistics.This method not only accelerates the detection efficiency but also reduces the difficulty of subsequent defect extraction.Based on the feature extraction of gray level co-occurrence matrix,the lining image is first quantized to the 16-level gray scale,and then the gray level co-occurrence matrix features in the four directions of 0°,45°,90° and 180° on the quantized image are calculated.The value,in which the gray level co-occurrence matrix calculates four design features,namely contrast CON,energy(second moment)ASM,entropy ENT,and deviation moment HOMO,the final effect diagram obtained by experiments is obvious.Based on the Gabor channel fusion defect feature extraction,the Gabor filter is particularly sensitive to texture.At the same time,Gabor's different scales and different angles have different sensitivity to different defects.This thesis improves a Gabor-based lining defect extraction based on multi-channel fusion,by using an algorithm to automatically select the appropriate parameters.Used by comparing three algorithms.Thirdly,the classification of lining defect features is proposed.This thesis compares the two existing algorithms,based on SVM support vector machine and BP neural network.Based on the experience and defect extraction experience of the predecessors,this thesis improves and designs 7 defect features for training,namely length,width,duty cycle,energy,entropy,contrast and correlation.Among them,the BP neural network is weaker than the SVM support vector machine,and the average correct rate of the six defects of the SVM support vector machine classification is 90.67%.Finally,a set of UI interface for lining defect detection is convenient for workers to test.Based on OPENCV3.0 and QT5.7.0,the platform mainly includes modules for lining defect collection,detection,classification and query.According to the research content of this thesis and the current development status of lining defect detection,the lining defect detection process proposed in this thesis can solve some problems of lining defect detection in real life,and promote automation of lining production.