Multi -factor Analysis of Visual Inspection Technologies for Strip -Steel Surface Defects
Abstract
Strip steel surface defect detection serves as a vital link for intelligent quality management in the steel industry. Surface defects can impair the mechanical properties of strip‑steel products and bring potential economic risks during actual manufacturing, making effective defect inspection highly essential for industrial production. This paper conducts a comprehensive review on strip steel surface defect detection technologies. It sorts out the fundamental principles of four major technical categories, namely traditional detection, conventional machine‑learning‑based detection, deep‑learning‑driven detection, and generative data augmentation. The strengths and limitations of different representative methods are systematically compared. Combined with existing research based on public benchmark datasets such as NEU Surface Defect Detection Dataset (NEU‑DET), this paper analyzes practical bottlenecks of existing approaches, including insufficient annotated defect samples, imbalanced sample categories, complex interference from industrial production environments, difficulties in multi‑scale defect identification, and challenging model deployment requirements. Corresponding improvement suggestions are put forward from dataset construction, model robustness enhancement, multi‑scale feature fusion, the fusion of Convolutional Neural Networks (CNN) and Transformer, as well as lightweight deployment. This review provides valuable references for subsequent research on intelligent strip steel quality inspection.