Printing Defect Detection Algorithm And System Based On GPU Parallel

Printing defect detection is one of the important links in the production and quality management of printing industry.At present most of the domestic manufacturers to adopt the automatic detection of defect detection system for printed matter printing defects,but because of the foreign system is expensive,in the domestic use range is not wide,defect detection system for printed matter and domestic development of many trapped in defect detection accuracy,speed and cost of the system has not been put into practical use.In recent years,the powerful parallel computing ability of GPU provides a new idea and solution for large scale data computing tasks.In this paper,based on the GPU parallel technology and machine vision,the defect detection algorithm for printed matter based on GPU is implemented.In this paper,the hardware structure of NVIDIA GPU,CUDA programming model and memory model are described in detail.This paper analyzes the defect detection algorithm based on CPU,and summarizes the existing problems.This paper introduces the parallel image processing technology based on CUDA,and analyzes the feasibility of the parallel algorithm of printing defect detection.In order to solve the problem of serial detection algorithm for printed matter defect,a parallel detection algorithm for printed matter is proposed,which is suitable for CUDA parallel architecture.This algorithm is based on the structure principle of four fork tree and divide and conquer idea,In the first four sub regional unit divides the image into several sub regions,sub regional defects were detected,and whether there is a defect detection sub region connection between regions with time,region merging,similar to the four sub regional units are merged into one image until now.When the defect is searched in the sub region,the number of iterations is reduced by using odd and even row alternate search.In terms of the time complexity of the algorithm,the image processing of a size of m* n,in this paper,the parallel detection algorithm for printed matter defect is O(n~2/d) which is much lower than the O(n~2)of the defect clustering algorithm based on the sequence and the length of the code,which d is the number of blocks in the Y direction of the thread grid.And the use of the performance of CUDA program developed by NVIDIA Parallel Nsight analysis tool to analyze the performance of the implementation of the CUDA program,put forward to change thread layout mode,the use of shared memory and kernel fusion of three kinds of optimization scheme.According to the characteristics of the object of the subject,the paper designs the defect detection platform of simulated printing,and designs and realizes the image acquisition scheme,the synchronization scheme and the software system.In this paper,four kinds of printing images with different degrees of defects are used to verify the proposed algorithm and optimization strategy,and compared with the serial detection algorithm.The experimental results show that the proposed algorithm can improve the detection speed by 4~6 times,while taking into account the accuracy.