A²S²C-Det: Dual-Path Adaptive Aggregation with Spatial-Semantic Compensation for Strip Steel Surface Defect Detection

Area of Science:

• Materials Science
• Computer Vision
• Artificial Intelligence

Background:

• Accurate steel surface defect detection is vital for manufacturing quality and reliability.
• Deep learning methods face challenges with background texture, spatial detail loss, and feature imbalance.

Purpose of the Study:

• To propose A²S²C-Det, a novel detector enhancing feature representation for steel surface defect detection.
• To address background interference, spatial detail loss, and semantic imbalance in defect recognition.

Main Methods:

• Introduced a Semantic Refinement Bottleneck (SRB) to suppress background noise and capture defect shapes.
• Developed a Dual-Path Adaptive Aggregation (DPAA) module for cross-level feature fusion.
• Implemented a Spatial-Semantic Gated Compensation (SSGC) module to recover lost spatial details.

Main Results:

• A²S²C-Det achieved high performance on three benchmark datasets.
• Demonstrated superior mean Average Precision (mAP) and mAP50 scores compared to state-of-the-art methods.
• Achieved mAP50 scores of 82.0%, 73.0%, and 91.2% and mAP scores of 47.1%, 36.5%, and 60.6%.

Conclusions:

• The proposed A²S²C-Det effectively enhances feature representation for steel surface defect detection.
• The integration of SRB, DPAA, and SSGC modules significantly improves detection accuracy.
• A²S²C-Det offers a promising solution for robust and accurate defect identification in industrial applications.