Research On Image Processing And Classification Of Strip Surface Defects

Hot strip is widely used in automobile manufacturing,shipbuilding industry,chemical engineering and other industries,and the product quality is directly determined by its surface quality.Due to hot-rolled equipment,technological process and other reasons,crazing,inclusions,patches,pitted surface,rolled-in scale,scratches and other defects often appear on the strip surface.These defects not only affect the product appearance,but also easily lead to parts cracking and corrosion,which greatly reduces the strength,service life of strip,thus it is very important to study the surface defects of strip and identify the types of defects.In this paper,based on machine vision,the image preprocessing,image feature extraction,feature data processing and image classification algorithm are studied.For preprocessing process of defect image,we mainly use evaluation parameters to select a variety of preprocessing algorithms: The Peak Signal to Noise Ratio parameter is used to select image filtering algorithm;the Average Gradient parameters are used to select image enhancement algorithm;the Mean Pixel Accuracy and Mean Intersection over Union parameter is used to select image segmentation algorithm.The preprocessing algorithm flow which is suitable for the surface defect image of hot rolling strip is determined,and the accuracy of feature extraction data of defect image is improved.Aiming at the problem of extracting features directly in image domain or frequency domain in traditional methods,the two-dimensional discrete wavelet transform is proposed firstly to decompose the surface defect image into different sub-bands.Then,through the Information Entropy parameter,it is determined that the low-frequency subband of the image contains the most image classification information,Therefore,texture feature,moment feature and projection feature are extracted based on low frequency subband of defect image.At the same time,standardized processing and importance analysis of feature data are carried out to realize the weighted fusion of moment feature and texture feature,which provides a reliable feature data base for the subsequent classification and recognition of the defect image.By using support vector machine classifier and comparing the classification ability of texture feature,moment feature and projection feature,the texture feature on the low frequency sub-band is selected as the classification feature of the strip surface defect image.According to the shortcomings of the single feature single classifier algorithm,which contains single defect information and has limitations on the classification of strip surface defects,the weighted voting classification algorithm based on multi feature fusion and multi classifier integration is used to solve this problem.The classification accuracy of the two algorithms is 96.11% and 97.22% respectively,which effectively realizes the surface defect classification of hot rolled strip.Finally,the algorithm and theory involved in the subject are verified by experiments,which further proves the effectiveness of the algorithm in this paper for image classification of surface defects of hot strip.