Detection method and classification of steel defects based on multivariate feature analysis

Purpose To facilitate the detection and classification of complex defects that involve various points, lines, and surfaces on steel material surfaces. The study aims to propose a comprehensive detection and classification method to improve accuracy and robustness. Design/methodology/approach The approach begins by identifying regions of interest using local gray value variability and image saliency. This is followed by enhanced adaptive histogram equalization and median filtering to emphasize edges and reduce noise. Next, the Canny edge detection algorithm accurately captures edge information, while morphological processing further refines the defect areas. In the final step, a classification decision tree analyzes various features to efficiently identify and categorize defects. Findings The experimental results demonstrate that this method achieves higher accuracy and robustness in detecting complex surface defects on steel materials. It effectively classifies various defect types and improves the accuracy of detection in complex scenarios. Originality/value This study introduces an innovative integrated approach that combines preprocessing techniques, edge detection, morphological processing, and classification decision trees. This approach offers a distinct method for the precise detection and classification of complex defects on steel surfaces, achieving greater accuracy and stability than conventional techniques.