Research On Detection And Classification Of Strip Surface Defects

As one of the most commonly used raw materials in modern society,strip steel is widely used in many fields such as aerospace,machinery manufacturing,etc.Its surface quality problems not only affect the service life of steel plate,but also may bring serious safety risks.Detection of strip surface defects and classification of different types of defects has become an essential process in the production process of many iron and steel enterprises.With the rapid development of machine vision technology and image processing technology,this paper designs a set of high resolution and high precision strip surface defect detection and classification system according to the specific demand of a domestic steel production line.Firstly,the image acquisition method based on linear CCD industrial camera is determined,and the hardware equipment such as camera,lens and light source is selected.Then,an image screening algorithm based on regional gradient projection is proposed for the massive surface images collected.Compared with the traditional gradient method,the accuracy of screening is greatly improved,up to 99.05%.Part in image processing,first finished to uneven illumination algorithm based on gray level correction,and then compared the five filter algorithm performance,decided to adopt the pseudo median bilateral filter to remove the image noise,then chooses the homomorphic filter algorithm for image enhancement,and finally in the edge detection after image segmentation to get the high definition defect target image.In the defect classification part,Hu invariant moment feature,gray scale feature and texture feature of the defect are extracted and the principal component analysis method is adopted to carry out feature dimension reduction.A multi-classifier based on support vector machine is designed,and a variety of parameter optimization algorithms such as K-fold cross validation,particle swarm optimization,genetic algorithm and gray Wolf algorithm are compared.The grey Wolf algorithm based on clustering initialization population and adaptive convergence factor and proportional weight was proposed to optimize the vector machine parameters,and the final classification accuracy reached 97.61%,meeting the production needs of steel mills.