Research On The Algorithms Of Surface Defects Inspecting For Casting Billet Based On Machine Vision

In the process of steel production, some steel would become defective on surfacebecause of the outdated production equipments and immature technique. These defectswill reduce the yield of the steel plate, which lead to huge economical losses. As animportant link in the steel production process, the surface quality of casting billet isdirectly associated with the surface quality of the final steel strips. Thus, if we couldtimely detect the surface defects of the casting billet, we could avoid useless processing,save energy and reduce economic losses. Based on the research of automatic surfacedefects inspection system, this paper focus on software part of the casting billet defectsdetection system based on machine vision.The paper designs the image processing algorithms for casting billet images,including image preprocessing, defects detection and location,image feature extraction andclassification. Among these three parts, image preprocessing, defects detection andlocation are the key research point.On the stage of image preprocessing, the paper compared many methods such asSegmented gray-scale transformation, direct illumination compensation, homomorphicfiltering, global histogram equalization and others. And also tried to use (Successive MeanQuantization Transform)SMQT method for image enhancement, yet obtained no idealresults and the efficiency of the algorithm is low. Finally, we employed an improvedalgorithm of histogram equalization to solve the problem.On the defects detection stage, by employing the multi-resolution image segmentationalgorithm to initial segmentation of the targets, and then detect the abnormal regions tofind the suspicious casting billet images. We use this more efficient algorithm to fastprocess all the casting billet images.In the process of defects segmentation and location, we use Morphological filtering tofilter Some pseudo defects of the binary images, and connect and analyze defects to recover the disconnected defects.In the process of feature extraction and classification, we extract the features from thedefect images after segmentation. Principle Component Analysis(PCA) is used to reducethe feature dimensions. Then, we introduce the Support Vector Machine method (SVM) toclarify defects.The algorithm used in this paper integrally considered efficiency and timeliness tobasically reach the requirements of online slab defect detection.