Research On Image Processing Recognition Method Of Surface Defects Of Strip Steel And System Design

Strip steel is widely used in industries such as bridges,construction,lathes,and machinery.Surface defects of strip steel can affect the quality of strip steel.Therefore,machine vision detection methods have developed rapidly.Among them,The recognition technology of strip steel surface defects based on image processing needs to address the issue of balancing recognition accuracy and Real-Time Property,avoiding poor Real-Time Property when the recognition accuracy is high,and also avoiding poor recognition accuracy when the Real-Time Property is good.To solve these problems,the research on surface defect recognition methods for strip steel has been carried out,with the main research content as follows.(1)An improved feature extraction algorithm of Local binary patterns were proposed to solve the problems of strip surface defects such as complex and diverse textures,different texture shapes,uneven illumination,weak contrast in local areas,and noise.In this method,three non equivalent mode texture structures in the Improved LBP(ILBP)algorithm were introduced to reduce the impact of complex textures;Secondly,the frequency histogram of LBP value was changed into a frequency histogram of gradient amplitude and direction to reduce the impact of weak contrast between defective and non defective regions;Finally,the threshold T in the local ternary patterns were introduced to reduce the impact of light and noise.Through simulation experiments,this method can improve the accuracy of defect recognition for slag inclusions,scratches,and pits.When the Gaussian noise is 50 d B and the threshold value T=5is set,the accuracy of ILLBP+SVM recognition can reach 96.44%.(2)In order to further improve the accuracy of strip surface defect recognition and reduced the impact of redundant features,an improved Salp Swarm Algorithm(SSA)feature selection algorithm(ISSA)was proposed.This method utilized the location of the salp swarm for feature selection and SVM parameter optimization.Firstly,the elite pool strategy,improved crazy operator,and parameter c2 were introduced into the leader position update formula.Secondly,the elite pool strategy and Grey Wolf follow method were introduced into the follower position update formula.The experimental results show that the recognition accuracy of ISSA feature selection algorithm is better than that of SSA,(CSSA,Crazy SSA),and(NSSA,Novel SSA)under 50 d B and 40 d B Gaussian noise environments.Compared to using SSA algorithm only to optimized the penalty factor c of SVM and the parameter g of polynomial kernel function in50 d B Gaussian noise,the recognition accuracy is improved from 96.44% to 98.43%.(3)A mixed kernel function was introduced to solve the problem of uneven distribution of strip defect samples.In this method,the radial basis Gaussian kernel function and the polynomial kernel function were weighted and mixed to solve a single kernel function or focus on finding the global optimal solution;Or focus on finding local optimal solutions.Experimental results show that the highest recognition accuracy can be achieved when the weight coefficient t=0.3.Compared with polynomial kernel functions,the recognition accuracy of hybrid kernel functions is improved by 0.22% at 50 d B and 0.8% at 40 d B,and the average recognition time of each image can reach 0.33 s,satisfying the Real-Time Property requirements to some extent.(4)Finally,a visual software testing environment is implemented,and the ILLBP+ISSA strip surface defect classification and recognition model was tested experimentally and systematically.The experimental results show that the strip surface defect recognition model can effectively achieve the classification of strip surface defects,and provides a new solution for strip surface defect recognition.