MSFE-YOLO: A Steel Surface Defect Detection Algorithm Integrating Multi-Scale Frequency Domain and Defect-Aware...

Area of Science:

• Materials Science
• Computer Vision
• Artificial Intelligence

Background:

• Surface defect detection is critical for industrial manufacturing quality control.
• Existing algorithms struggle with multi-scale features, edge details, and frequency domain information.
• Simplistic feature fusion strategies limit the performance of current defect detection models.

Purpose of the Study:

• To propose the Multi-Scale Frequency-Enhanced YOLO (MSFE-YOLO) algorithm for improved steel surface defect detection.
• To address limitations in capturing multi-scale characteristics, utilizing edge and frequency features, and feature fusion.
• To enhance the accuracy and efficiency of automated surface defect detection systems.

Main Methods:

• Developed a Multi-Scale Frequency-Enhanced Convolution (MSFC) module using depth-adaptive dilated convolutions and the Laplacian operator.
• Designed a Cross-Stage Partial with Multi-Scale Defect-Aware Attention (C2MSDA) module incorporating Sobel edge perception and multi-scale attention.
• Introduced an Adaptive Feature Fusion Enhancement (AFFE) module for adaptive aggregation of multi-level features.

Main Results:

• MSFE-YOLO achieved mAP@0.5 scores of 79.8% on NEU-DET and 66.7% on GC10-DET datasets.
• Performance improvements of 1.7% and 2.1% over the benchmark YOLOv11s model were observed.
• The algorithm maintained a high inference speed of 89.3 FPS, meeting real-time detection requirements.

Conclusions:

• The proposed MSFE-YOLO algorithm effectively detects surface defects on steel products.
• Integration of multi-scale frequency enhancement and defect-aware attention mechanisms significantly improves detection accuracy.
• The algorithm satisfies real-time industrial detection needs, offering a robust solution for quality control.