Multi-Level Scale Attention Fusion Network for Adhesive Spots Segmentation in Microlens Packaging

Area of Science:

• Optics and Photonics
• Machine Learning
• Quality Control

Background:

• High-power lasers require precise collimating lens packages.
• Accurate inspection of adhesive points is crucial for lens quality.
• Existing methods lack the required precision for adhesive spot detection.

Purpose of the Study:

• To develop a highly accurate and robust method for detecting adhesive spots on laser collimating lenses.
• To introduce the Multi-Level Scale Attention Fusion Network (MLSAFNet) for this task.
• To create and standardize a novel dataset for laser lens adhesive spots (LLAS).

Main Methods:

• Proposed the Multi-Level Scale Attention Fusion Network (MLSAFNet).
• Integrated a Multi-Level Attention Module (MLAM) and a Multi-Scale Channel-Guided Module (MSCGM).
• Developed the Laser Lens Adhesive Spots (LLAS) dataset using automated equipment and pixel-level standardization.

Main Results:

• MLSAFNet achieved a mean intersection over union (mIoU) of 91.15%.
• Maximum localization error was 21.83 μm, and area measurement error was 0.003 mm².
• Performance surpassed other target detection methods.

Conclusions:

• MLSAFNet offers a significant advancement in adhesive spot detection for laser lens packages.
• The developed LLAS dataset and MLSAFNet contribute to improved quality control in high-power laser manufacturing.
• The method demonstrates high accuracy and robustness, meeting the demands of precision inspection.