Technology Research On Surface Defects Detection And Recognition Of Strip

As an important product of steel industry, strip steel becomes the necessary raw materials for the defense and equipment, automotive manufacturing, aerospace fields.The quality of the strip surface has a great influence on the performance of the final product, and therefore efforts to improve the surface properties of the strip to improve the strip surface quality is of great significance.In this paper,considering the steel surface inspection of cold hot-rolled steel in certain large domestic steel group and all the disadvantages of human visual inspection method,designed on-line detection system of strip surface quality. This paper carried on a thorough research on image processing and recognition technology of strip surface defection,came up with the overall scheme and put them into practice at the same time.It proves that this detection system has successfully implemented the real-time detection and accurate identification and classification of strip surface defects.As a result,we get effective control of the quality of the strip surface.The paper mainly did some researches showed as follows:(1) Under the image preprocessing stage,focused on the common uniform illumination algorithms and determined the best solutions to uneven illumination according to experimental contrast,laid an important foundation for the subsequent images defects detection.For site dust disturbance noise is introduced, considering the filtering effect of denoising method and combining experimental analysis.As a method of filtering process,the final choice of bilateral filtering having edge security features,and experiments show that this method has better image noise resistance, and the mean square error is smaller too.(2)On the basis of the defect segmentation research,the image edge detection algorithm and image segmentation algorithm is analyzed, tried a different algorithm for edge detection and segmentation techniques, determined using Kirsch edge detection algorithm.And for the acquired image defect characteristics,proposed an efficient defect segmentation algorithm,which can accurately achieve the target defect region with isolated background.(3)Considering the insufficient merging or secondary merging issue,by studying several existing ROI merger rules,put forward the optimized ROI merger rules which proved a good solution to this problem.(4)According to class separability criterion and affine invariant guidelines of feature vectors, extracted a variety of classification features, including gray feature, geometry features, and projection characteristics.(5)SVM-based classifiers has a good classification ability and rapid convergence advantages etc.For this system,designed SVM classifier based defect recognition.The classification experiments show that this classifier has considerable value with high accuracy and short time-consuming.