Study On Image Detection And Classification Of Steel Plate Surface Defects

As the fundamental industry of the national economy,the steel industry plays an important role in economic and social development,but the steel surface defects reduce the quality of steel products seriously.The surface defect detection is one of the key factors to improve the quality of steel.In this paper,the steel surface defects detection and classification algorithms on steel surface are studied.The main research results are as follows:(1)In order to overcome the problems of large amount of image data and low computational efficiency,an algorithm based on the gradient of the steel plate surface image for region of interest detection is provided.On the basis of the gradient image,the local gradient statistical value is calculated to judge whether the image is defective or not.The algorithm can reduce computation complexity and increase the efficiency of the system.(2)In order to overcome that partial defect images which are affected by nonuniform illumination and low contrast,this paper presents the nonuniform illumination correction and enhancement algorithm based on Retinex algorithm and guided filter.The light component is estimated by the guided filter,and the reflection component is calculated by Retinex algorithm.The reflection component is enhanced to improve the contrast and reconstruct the original gray level information.(3)Aiming at the segmentation accuracy of steel plate defect images,a segmentation algorithm is proposed based on Center-Surround Difference.A combined DoG filter with different weight is used to filter the defect images after the preprocessing.The image features of Local and Global are extracted based on the Center-Surround Difference,the two features are fused linearly.The background area is suppressed while the foreground area is restored.At last,the defect image is segmented by using adaptive threshold.(4)According to the different information between before and after the segmentation of the defect images,the 36-dimensional eigenvalues are extracted as basis of defect images classification.In order to increase the classification accuracy of defect images,the PSO algorithm is improved to optimize the BP network in two aspects:inertia weight and population diversity.Finally,the purpose of defect images'classification is completed.The experimental results demonstrate that the algorithms designed in this paper can achieve the goals of the defects' detection and classification.It is consistent with the subjective effect and objective effect.