Research On Surface Defect Recognition Of Steel Plate Based On Machine Vision

In recent years,although the iron and steel industry of China has focused on resolving excess production capacity and eliminating backward production capacity,the output of strip steel products has generally shown an upward trend.the iron and steel industry of China not only leads the world in quantity,but also has made great progress in many aspects such as variety quality,equipment level,technology economy,etc.However,the domestic steel surface defect detection technology has lagged behind,and most SMEs still remain in the labor force.At the visual inspection stage,it is imperative to carry out research on the surface defect inspection system of the steel plate.In this context,this paper aims at the steel strip production line of steel enterprises,and studies the image preprocessing technology,feature extraction and selection techniques,and defect identification algorithms in the surface defect detection system of steel plates.(1)Selection and application of image pretreatment technology for surface defects of steel plates.The image preprocessing technology has done the following three aspects: ○1 Using the peak signal-to-noise ratio to compare the effects of several common filtering algorithms on the surface defects of the steel plate;○2 Using the histogram equalization to image Enhancement enhances the image clarity;○3 Analyze several classic edge detection operators,and select the Sobel operator as the segmentation operator of the image based on the operating efficiency.(2)Extraction and selection of surface defects on steel plates.The image itself is a set of high-dimensional data,so it is necessary to grasp the main information and eliminate redundant information.Firstly,the 70-dimensional features including geometric features,gray features,projection features and texture features were extracted from the preprocessed images.Fisher’s criterion was then used to select the contribution to class separability from the threshold of 0.15.The largest 42-dimensional feature.(3)A high-precision,high-stability defect recognition algorithm was proposed.Due to the limitation of the error rate of the weak classifier,the traditional AdaBoost algorithm can not combine enough weak classifiers to accurately identify the surface defects of steel plates.To solve this problem,based on Ada Boost.M1,an AdaBoost.BK algorithm for iterative filtering of "appropriate" weak classifiers was proposed.(4)Modeling and verification of surface defect recognition algorithms for steel plates.The image preprocessing,feature extraction and selection were performed on the surface defect data set of six steel surface defects and 1800 images.The BP neural network,the traditional AdaBoost algorithm and the proposed Ada Boost.BK algorithm were used to characterize the features.The defect tag was learned and tested.Experimental verification shows that compared with other traditional methods,AdaBoost.BK algorithm achieves the highest classification accuracy of 85.89%,while the stability is also significantly improved.