A steel defect detection method based on edge feature extraction via the Sobel operator.

Area of Science:

• Materials Science and Engineering
• Computer Vision and Machine Learning

Background:

• Steel surface defects like scratches and cracks compromise structural integrity and performance.
• Existing defect detection methods struggle with irregular defect shapes and sizes, leading to low accuracy.

Purpose of the Study:

• To develop an enhanced deep learning network for more accurate and efficient steel surface defect detection.
• To address limitations in current methods by integrating edge information and advanced feature reconstruction techniques.

Main Methods:

• Proposed the ECDY (EIFEM CARAFE DyHead) network, featuring a novel edge-focused feature extraction module using the Sobel operator.
• Integrated Content-Aware Reassembly Feature (CARAFE) for improved feature pyramid reconstruction.
• Utilized a dynamic unified detection head (DyHead) to adapt to various defect scales and detection tasks.

Main Results:

• The ECDY network demonstrated improved accuracy across YOLOv5, YOLOv8, and YOLOv10 versions with reduced parameters.
• In YOLOv8x, mAP@0.5 increased by 2.5% with a 12.4M parameter reduction.
• Compared to YOLOv8s, the proposed method achieved a 1.6% precision increase, 4% recall increase, and 4% mAP@0.5 increase, outperforming RT-DETR-L by 4.2% mAP@0.5 with fewer parameters.

Conclusions:

• The ECDY network significantly enhances steel defect detection accuracy and efficiency.
• The integration of edge information, CARAFE, and DyHead offers a robust solution for complex surface defect identification.
• This approach provides a competitive alternative to state-of-the-art models with improved performance and reduced computational cost.