Research On Image Dectection Method Of Zipper Selvedge Defect

The appearance inspection of zipper product is one of the key links during the production.Traditional appearance detected method of zipper product mainly adopts artificial to complete,there are the shortcomings of low efficiency and the accuracy is not high.Using the image detection instead of the artificial detection method will be a trend of detecting the zipper product appearance,and it has broad prospect and market value.Therefore,it is necessary to study with image detection method to detect the defect of zipper product,in order to achieve the purpose of improving efficiency and accuracy.Zipper selvedge defect is one of the main types of defect during the appearance defect of zipper product.However,there is no study on zipper selvedge detection alone so far,the study of using image detection method to detect zipper defect is mainly biased towards the zipper teeth zipper pull head and zipper limit code.So,this paper will segment the zipper selvedge along,and using image detection method to detect the defect.The main work in this paper is as follows:Firstly,the background and research significance of zipper defect detection is introduced,and the research status of zipper selvedge defect detection is analyzed,then introduced the research work in this paper.Secondly,the overall framework of the zipper selvedge defect detection is proposed,and the specific flow of defect detection is given.Besides,the process of the construction of the sample set is expounded.Thirdly,according to the characteristics of zipper selvedge significant texture,proposed the method of using LBP and GLCM to extract the feature.Then,discussed the problem of parameters selection,and gave the concrete steps of extracting zipper selvedge sample images feature.Fourthly,Studied on the method of identifying zipper selvedge defect by BP neural network.According to the theory of BP neural network,this paper discussed the problem of main parameters selection,and gave the training process of the network,and then expounded the specific steps of identifying zipper selvedge defect by BP neural network.Fifthly,relevant experiments were carried out,and the experimental results were analyzed.The feature extraction effect of the three models of the LBP algorithm were comparised,the result of the experiment shows that,under the condition of the same sample images,uniform pattern uses the least time to extract the feature,and it has the highest identification accuracy,thus,it is the best pattern of LBP.Then,compared the feature extraction effect of the uniform pattern and GLCM,the experimental result show that,for the speed of extracting the same sample images,uniform pattern is nearly three times of GLCM,and under the three types defect of pollution and broken and off-line,the average recognition accuracy of uniform pattern is 89.33%,and GLCM is 87.67%,uniform pattern is 1.66 % higher than GLCM.Afterwards,an analysis on the performance of the BP neural network is made,the experimental result shows that,the identification accuracy of BP neural network can reach 95.00%,and under the condition of testing 2200 samples,the average testing time of each sample is in 0.036 seconds.Thus,the BP neural network has good identification accuracy and good real-time performance.At last,researched the influence of the number of hidden layer nodes to the dectetion effect of BP neural network,it shows that,the relationship between the recognition accuracy and the number of hidden nodes is nonlinear,when the identification accuracy reaches the maximum,increasing the number of nodes,the identification accuracy is not improved,but the identification time is also increasing.Finally,a summary of the research achievements of this paper is made,and suggests for future research were given.