Enhanced transformer method coupled with transfer learning for surface defect segmentation of myopia control...

Area of Science:

• Ophthalmology
• Materials Science
• Computer Vision

Background:

• Myopia affects many schoolchildren globally, necessitating visual correction.
• Nanostructured myopia control spectacle lenses (NMCSLs) are used to slow myopia progression.
• Effective defect detection is crucial for NMCSL manufacturing quality.

Purpose of the Study:

• To develop an efficient and accurate method for segmenting surface defects in NMCSLs.
• To improve the quality control process in the precision manufacturing of myopia control lenses.

Main Methods:

• An enhanced transformer method combined with transfer learning (E2Trans) was developed.
• Two auxiliary decoders were incorporated to refine training loss and improve segmentation.
• A dedicated lens defect dataset was created for model validation.

Main Results:

• The E2Trans model demonstrated high segmentation accuracy and speed for five defect types: notches, black spots, bubbles, fibers, and scratches.
• The proposed method significantly improves defect detection capabilities for NMCSLs.
• A real-time lens defect detection system was successfully developed.

Conclusions:

• The E2Trans method offers a robust solution for high-accuracy defect segmentation in NMCSL manufacturing.
• This advancement contributes to improved quality control and potentially better myopia management through precise lens production.
• The developed system enables real-time monitoring and defect identification in lens production lines.