Research on a Metal Surface Defect Detection Algorithm Based on DSL-YOLO.

Area of Science:

• Industrial Manufacturing
• Computer Vision
• Machine Learning

Background:

• Metal surface defect detection is crucial for quality control in industrial manufacturing.
• Existing methods often struggle with low accuracy, high leakage rates, and false detection rates, especially for small or occluded defects.

Purpose of the Study:

• To propose a novel and efficient model, DSL-YOLO, for accurate metal surface defect detection.
• To enhance feature extraction capabilities for challenging visual conditions like blurriness and small object detection.
• To improve multi-scale feature representation for critical image information without substantial computational overhead.

Main Methods:

• Integration of the DWRB module with C2f to create the C2f_DWRB structure for improved small and occluded target detection.
• Development of the SADown module to enhance feature extraction from blurred images and very small objects.
• Proposal of the LASPPF structure to boost multi-scale feature extraction and capture essential image details like edges and textures.

Main Results:

• The DSL-YOLO model demonstrated significant performance improvements on the GC10-DET and NEU-DET datasets.
• Achieved a mean Average Precision (mAP@0.5) increase of 4.2% on GC10-DET and 2.6% on NEU-DET.
• The model effectively addressed common challenges in metal surface defect detection, showing enhanced accuracy and efficiency.

Conclusions:

• The proposed DSL-YOLO model offers a valuable solution for industrial metal surface defect detection.
• The novel architectural components (C2f_DWRB, SADown, LASPPF) contribute to superior detection accuracy and feature extraction.
• The model provides a feasible and efficient approach for real-world industrial applications, overcoming limitations of previous methods.