Research On Strip Surface Defect Detection System And Classification Algorithm Based On FPGA

In the production process of cold rolled strip,different kinds and degrees of defects are easy to occur,such as scratches,oxidation,spots and so on.If these defects can not be detected and solved in time,it will have a bad impact on the quality of steel plate.It is found that a considerable number of steel mills still judge the defects by the naked eyes of the workers in the production line,which consumes a lot of human resources,and there is a significant gap between them and the automatic production inspection widely used in foreign countries.According to the requirement of Chongqing Panzhihua steel plant for unmanned detection of cold rolled strip production line,this paper designs a visual detection system for surface defects of cold rolled strip based on FPGA.The main contents are summarized as follows:(1)A hardware system for image acquisition,processing and transmission of strip defects based on FPGA is built.The peripheral circuit of Xilinx series XC7A35 t chip is designed;the interface of OV5640 camera module is designed;the Ethernet module based on KSZ9031 RNX chip is designed;The mathematical models of different edge feature extraction algorithms are compared,and the effect of each algorithm in strip surface defect detection is analyzed with MATLAB.Finally,Sobel operator is selected;the pipeline architecture is used to filter,grayscale and Sobel the defect image.(2)The KNN classification algorithm model with improved K value selection strategy is designed.The mathematical model of PCA maximum variance value and eigenvalue is deduced theoretically,and the dimension of strip defect image feature is reduced,the key feature vector of defect image is extracted,and the data structure is simplified;the classification model is preliminarily verified by NEU-CLS strip defect image data set.(3)The experimental platform is built,and two groups of comparative experiments are designed,which are: the comparison of classification accuracy of traditional KNN algorithm and improved KNN algorithm for gray defect image;the comparison of classification accuracy of improved KNN algorithm for gray defect image and defect image after feature edge extraction.The former verifies the classification effect of the improved algorithm,and the latter tests the classification effect of the FPGA system after the defect feature edge extraction.