Manufacturing, Visual Testing
Multiscale Defect Detection
on Hot-Rolled Steel Strip Surfaces Based on Edge Enhancement and Bidirectional Feature Fusion
ABSTRACT
Hot-rolled steel strip is widely used in industrial manufacturing. However, complex surface textures and large variations in defect scales make multiscale defect detection challenging for existing machine vision methods. To address these issues, this paper proposes a multiscale surface defect detection method for hot-rolled steel strips based on edge enhancement and bidirectional feature fusion. First, an edge information (EI) module is designed to reduce background texture interference and enhance defect boundary features. Second, the C3K2-DB module is developed by integrating FasterBlock and the efficient multiscale attention (EMA) mechanism to improve the representation capability of defect category features. Finally, a RepBiPAN bidirectional feature fusion network is constructed to strengthen the interaction of multiscale features and improve defect detection performance. Experimental results on the NEU-DET dataset demonstrate that the proposed method achieves a precision of 83.9%, a recall of 84.4%, and an mAP@0.5 of 85.6% with a lightweight model structure. The proposed method effectively improves multiscale surface defect detection performance for hot-rolled steel strips and provides a feasible solution for industrial inspection applications.
KEYWORDS: hot-rolled steel strip, surface defects, multiscale, edge enhancement, bidirectional feature fusion
