Strip steel surface defect classification method based on combination of statistical features and image features

Abstract

A strip steel surface defect classification method based on combination of statistical features and image features comprises steps: collecting a set of marked training samples; S2, extracting corner points of each defect sample image and description of each corner point; describing the corner points of all defect sample images in a K-dimensional space, to perform unsupervised learning clustering;combining M-dimensional image eigenvectors and N-dimensional statistical eigenvectors of defect samples to form (M+N)-dimensional eigenvector of the defect samples; using an adaptive boosting tree training method to perform supervised learning training on the (M+N)-dimensional eigenvector, to train a self-learning classifier B, outputting classification results of defects. The method separates different defects and improves classification accuracy of strip steel surface defects. The self-learning classifier is used to classify detected defects automatically and accurately when surface defectsare detected on-line in real time. Classification rules are obtained by supervised learning instead of manual input.

Description

technical field [0001] The invention belongs to the field of steel strip surface defect detection systems in the metallurgical industry, in particular to the field of surface defect classification, and specifically relates to a strip steel surface defect classification method based on the combination of statistical features and image features. Background technique [0002] Strip surface defects are an important factor affecting the surface quality of cold-rolled strip, directly affecting the appearance and performance of the final product. The surface inspection system uses camera sensors to scan the surface of the steel strip to obtain two-dimensional images of the surface of the steel strip, and uses machine vision technology to detect and classify surface defects. At present, the accuracy of defect detection based on various advanced algorithms has reached 98% %, which basically meets the production requirements, but the correct rate of defect classification is always uns...

AI Technical Summary

Problems solved by technology

[0003] The technical problem to be solved by the present invention is to provide a strip surface defect classification method based on the combination of statistical features and image features for the above-mentioned deficiencies in the existing strip surface defect classification, so as to improve the accuracy of the strip surface quality detection system. The correct rate of classification of steel surface defects; when surface defects are detected online in real time, the self-learning classifier is used to automatically and accurately classify the detected defects, and the classification rules do not need to rely on manual input

Images