Study On Image Processing Method Of Low-contrast And Small Defect Detection For Strip Surface

The iron and steel industry is an important basic industry of national economy,it is the supporting industries to realize the industrialization.Strip steel as one of the main product of iron and steel industry,and it has become one of the important raw materials industries such as cars,ships and aerospace.Demand of surface quality of strip steel for subsequent processing is higher and higher as the continuous development of science and technology.Surface quality is an important indicator of strip steel.So the strip surface quality detection is of great importance.In this paper,some key technologies for the detection of the often appeared strip surface defect in the rolling process were researched.The overall scheme of the defect detection system was put forward.And emphatically analyzed the image denoising method,defect segmentation method and feature extraction method of the detection system.This main research content was as follows:1.According to the noise characteristics of the defect image,the effect of traditional denoising method and nonlocal average denoising method denoising were compared.The experimental results shown that the nonlocal average denoising method worked effectively and at the same time can better retain the image texture and edge and other important details.2.Aiming at the problem that the filtering parameter “h” was hard to choose in nonlocal average algorithm.PCA noise estimation method was proposed to improve the algorithm,the value of filtering parameter “h” adaptively various according to the image,thus better denoising effect could be achieved.3.According to the characteristics of the strip surface defects with uneven lighting,target smaller,low contrast,the regional segmentation method based on the Mask well light processing was put forward.Experimental results shown this method can effectively solve the problem of uneven illumination and get very good segmentation effect.4.When the extract of texture features in the strip surface image of multiple information fusion feature vector.The graylevel co-occurrence matrix(GLCM)and Tamura texture feature as a common combination was put forward to extract image texture feature.Then the improvement was validated by the classification accuracy of BP network.