Research on an Improved YOLOv8 Detection Method for Surface Defects of Optical Components

Area of Science:

• Optics and Materials Science
• Computer Vision and Artificial Intelligence

Background:

• Optical components are vital in aerospace, microelectronics, and precision measurement.
• Surface defects degrade system performance, requiring advanced detection methods.
• Current defect detection methods face challenges in resolution, precision, and efficiency.

Purpose of the Study:

• To develop a high-resolution, high-precision, and efficient optical surface defect detection algorithm.
• To improve the performance of object recognition models for optical component defect analysis.
• To address the limitations of existing methods in complex detection scenarios.

Main Methods:

• An improved YOLOv8 object recognition algorithm was proposed.
• The BRA attention mechanism was integrated into the YOLOv8 backbone for enhanced multi-scale feature processing.
• The Context-GuideFPN module replaced the original feature fusion module for adaptive multi-scale feature integration.
• The algorithm was trained and evaluated on a high-quality microscopic dark-field image dataset.

Main Results:

• The enhanced BACG-YOLOv8 demonstrated excellent performance in optical component defect detection.
• The optimized network accurately extracted defect details, including refined edge features.
• The algorithm effectively suppressed noise interference, reducing detection errors.
• Improved defect extraction accuracy was achieved compared to baseline models.

Conclusions:

• The BACG-YOLOv8 algorithm offers a significant advancement in optical surface defect detection.
• The integration of BRA attention and Context-GuideFPN enhances adaptability and accuracy in complex scenarios.
• This approach provides a robust solution for ensuring the quality and performance of optical components.