Research On Classification Algorithm Of Printing Roller Surface Defects Based On Deep Learning

Printing industry is the basic industry of national economy,which plays an important role in the process of cultural development and inheritance,and plays an important role in social development,information dissemination and cultural transmission.With the rapid development of the market,fine and meticulous printing products are more and more popular.As the core part of printing equipment,the quality of roller directly affects the quality of printing products.At present,the detection of roller quality mostly depends on manual detection,which is inefficient and easy to fatigue.In order to solve the above problems,this paper proposes a method of machine vision and deep learning to detect and classify the defects of printing roller,which not only improves the detection efficiency,but also lays a foundation for the intelligent development of printing industry in the future.In this paper,the surface defects of the gravure roller are studied.The defect image is preprocessed and detected,and the defect classification algorithm is studied.On the basis of optimizing feature extraction network,a roller defect classification network based on improved RESNET is proposed to complete the overall construction of defect detection and classification system.The specific work is as follows:(1)This paper proposes an improved algorithm of hog eigenvalue,uses SVM to classify it,and adds the method of double line interpolation,which makes hog feature more obvious for roller defects and improves the classification effect of SVM.(2)A defect classification network based on improved RESNET is proposed for defect image classification.The network is improved on RESNET network model.Using the advantage of residual network structure to retain data information,an improved thin RESNET feature extraction network is proposed,which greatly optimizes RESNET network and saves image processing time.The experimental platform of roller defect detection and classification is constructed,and the appropriate hardware equipment and configuration are selected to verify that the above improved RESNET network reduces the error rate of roller defect classification in gravure printing,improves the recognition rate and reduces the image processing time,which proves that the algorithm is practical,suitable for industry application,and saves enterprise cost;the improved network ensures the accuracy of classification network On the basis of this,it can greatly reduce the amount of parameters and calculation of the network model.There are 25 pictures,8 tables and 64 references.