MCH-YOLOv12: Research on Surface Defect Detection Algorithm for Aluminum Profiles Based on Improved YOLOv12

Area of Science:

• Materials Science
• Computer Vision
• Artificial Intelligence

Background:

• Surface defect detection in aluminum profiles is crucial for quality control in industrial production.
• Existing methods face challenges with imbalanced defect categories, small-scale defects, and irregular shapes, limiting accuracy and robustness.

Purpose of the Study:

• To develop an improved defect detection algorithm (MCH-YOLOv12) for aluminum profiles.
• To enhance feature extraction and capture fine-grained defect characteristics.
• To improve adaptability to various defect sizes and shapes for real-time inspection.

Main Methods:

• Modified YOLOv12 architecture incorporating MultiScaleGhost convolution in the Backbone for enhanced feature representation.
• Introduced Spatial-Channel Collaborative Gated Linear Unit (SCCGLU) in the Neck to capture directional and edge-specific defect features.
• Implemented a Hybrid Head combining anchor-based and anchor-free detection mechanisms.

Main Results:

• MCH-YOLOv12 demonstrated improved accuracy and reduced category imbalance on an aluminum profile defect dataset.
• The enhanced algorithm showed better detection of small-scale and irregularly shaped defects.
• Achieved lower computational complexity (parameters and FLOPs) compared to baseline models.

Conclusions:

• MCH-YOLOv12 offers a robust and accurate solution for surface defect detection in aluminum profiles.
• The algorithm's improvements make it suitable for real-time industrial inspection applications.
• The study highlights the effectiveness of MultiScaleGhost and SCCGLU modules in defect detection.