Research On Novelty Detection And Defect Classification In Industrial Vision

In the field of industrial vision,more and more algorithms are used to classify sample images obtained in actual production to facilitate detection of defective products.Although these algorithms greatly reduce the manpower and material resources,improve the efficiency and accuracy,most of the algorithms only detect and classify the defect categories that have been obtained,and do not consider the unknown defects that may occur in the production.This leads to the ability of these algorithms to detect and classify them in the face of unknown defects in actual industrial production.Therefore,the actual industrial production may produce all the defective products and testing and classification,has become an urgent problem to be solved.The research topic of this paper is based on solving the problems arising from the actual industrial production process,considering the processing time of the algorithm and the requirement of resource consumption,and testing and sorting a large number of image data obtained in industrial production and has good The detectionclassification ability.In the practical application has achieved good results.The paper focuses on the need to solve the practical problems,mainly carried out the following work.On the basis of one class of support vector machines,the distance of theneighborhood is used to measure the difference of the picture in order to shorten the processing time of the picture.While using the difference between heterogeneous images to determine the category they belong to the realization of the unknown defective image classification goals.The unknown defect detection classification system developed in this paper mainly includes feature extraction and dimension reduction,a class of support vector machine training and classification,unknown defect detection and classification module,each module maintains relative independence and can work together.The system takes less time to deal with the picture,and to facilitate follow-up developers according to the actual needs of the choice of work modules and its expansion and improvement.This article developed by the actual industrial project sample image detection,while maintaining a high processing accuracy at the same time,greatly improving the processing efficiency.