Research On Defect Detection Method Of Paper Cup Lid And Tray Based On Image

With the progress of modern technology,industrial products have become an indispensable part of our life,such as the commonly used chopsticks and disposable tableware used in restaurants.However,when these products are initially produced,it is inevitable that there will be some defective products,such as products with cracks and other defects.Therefore,it is necessary to check out these defective products in a certain way.At present,most factories still use artificial testing to detect these defective products,which not only requires a lot of manpower,material resources and financial resources,but also relies heavily on the experience and state of the testers for the efficiency and accuracy of testing.Therefore,with the rapid development of computer vision technology today,we can use the related technology of computer vision,so that the computer can automatically detect the defects of industrial products,to achieve the purpose of detecting the defects of industrial products stably and efficiently.This thesis is devoted to studying the defect detection algorithm of paper cup lid and tray based on deep learning.To this end,this paper completes two aspects of work:(1)A defect detection algorithm for paper cup lid with label samples is designed.In this work,we can obtain samples with labels,and then directly use the convolutional neural network to design algorithms to complete the defect detection of industrial paper products.(2)Design the defect detection algorithm for paper tray without label samples.In this work,two types of data are needed,one is the source domain data with labels for auxiliary detection,and the other is the target domain data without labels that we want to detect defects,and these two types of data have similar defect characteristics.In this way,we can use the method of generating antagonistic network(GAN)to narrow the data feature distribution of this type of industrial paper products with labels(source domain)and this type of industrial paper products without labels(target domain),so as to achieve the purpose of defect detection for industrial paper products without labels.In this thesis,a complete experiment is designed to test the performance of the algorithm.First of all,we used the labeled industrial paper product defect detection algorithm to detect the defects of 60,000 labeled cup lid images,with the detection accuracy reaching 91%.Then,we used the defect detection algorithm of industrial paper products without label samples to detect the defects of 100,000 pallet images without labels.At this time,we used the lid samples with labels to detect the defects.Finally,the detection accuracy of pallet images also reached 74.5%.