The Study Of Processing Algorithms Of Automatic Surface Defects Inspecting System For Casting Billet

Steel, a raw material, is crucial in a nation’s production. However, in the process ofsteel production, some steel become defective on surface as a result of the influence ofequipments and technique, which leads to huge economical losses. As a key stage in theproduction, the quality of casting billet is directly associated with the quality of other steelstrips. Thus, real-time surface defects inspecting system for casting billet is of greatimportance. Based on the research of automatic surface defects inspecting system,thispaper focus on software part of the inspecting system.The paper designs the image processing algorithms for casting billet images,including image preprocessing, fast objects detection, defects segmentation and location,image feature extraction and selection and defects recognition. On the stage of imagepreprocessing, the paper compared traditional methods such as histogram equalization andothers, use the Gaussian Curve to fix the problem caused by uneven illumination, yetobtained no ideal results. Finally, we employ the Successive Mean QuantizationTransform(SMQT)to solve the problem. On the fast objects detection stage, by employingthe dynamic threshold selection method to make gray transformation and binarization, weexclude huge mount of casting billet images without defects and locate those with defects.In the process of defects segmentation and location, we use Mathematical Morphology tofilter the binary images, and make the clustering analysis to segment and locate the defectsarea. A fast clustering algorithm is employed here, which meets the needs well. In the end,we extract the features from the ROI after segmentation. Principle ComponentAnalysis(PCA) is used to reduce the feature dimensions. Then, we introduce the SupportVector Machine to make the classification for the defects. The experiment results show thekernel-based SVM is effective.