Research On Defect Detection And Classification Of Continuous Casting Steel Plate Based On Image Features

In the process of the production of the industrial continuous casting steel slabs,the evaluation and classification of the quality of the continuous casting billets still mainly rely on the artificial experience.Based on the pretreatment,segmentation and classification identification of the defects in the cold-acid-etched image of the continuous casting blanks,provides the preparation for the subsequent development of the defect rating system of the continuous casting blanks.In the pretreatment,the median filter and the wiener filter can effectively remove the noise in the image,but the multiple filter may result in a large amount of loss of the image information.In order to solve this problem,a fast weighted median-wiener filter under a switch structure is proposed in order to reduce the loss of the information in the image filtering process,and the effectiveness of the algorithm is verified by comparing with the result of the median and wiener filter.The particle swarm optimization(PSO)algorithm is introduced into the search process of the threshold vector of the 2D Otsu algorithm.The inertial weight update formula and variation strategy are improved.The 2D Otsu algorithm based on the improved PSO algorithm is constructed.Both the performance of the improved PSO algorithm and the ability to segment the defect image are verified,and the optimal parameters are found to deal with the defect image of continuous casting blanks.A method of gray scale standardization is designed for the problem of the non-uniform gray scale of the sample set given by the steel mill.The features of the center segregation defects are selected by the recursive feature elimination method based on support vector machine(SVM-RFE)to complete the feature reduction.Compared a variety of classifiers,selects the support vector machine based on decision direct acyclic graph(DDAG-SVM),introduces the fuzzy function and improves the original class-based separability measure formula based on the Euclidean distance,and constructs the DDAG-SVM structure to explain the poor classification effect of the other classifiers.The experimental results show that the model has a significant effect on the classification of the central segregation defects of the C class,and the classification of A/ B class has good results.It has certain application value.