Intelligent evaluation for lens optical performance based on machine vision.

Area of Science:

• Optics
• Computer Vision
• Artificial Intelligence

Background:

• Current visual inspection of collimating lens light-spot images is inefficient and prone to human error and fatigue.
• Machine vision and deep learning offer potential solutions for enhancing the accuracy and efficiency of optical performance evaluation.

Purpose of the Study:

• To develop and validate a deep learning-based model for the automated optical performance evaluation of collimating lenses.
• To improve the efficiency and accuracy of quality inspection in collimating lens production.

Main Methods:

• A dual-branch structure deep learning model was designed for light-spot image evaluation.
• A dataset of 9000 labeled lens light-spot images was created for training and testing.
• A weighted multi-model voting strategy was employed to enhance classification performance.

Main Results:

• The proposed deep learning model accurately classifies collimating lens optical performance.
• The integration of a weighted multi-model voting strategy further boosted model performance.
• The final classification accuracy reached an impressive 98.89%.

Conclusions:

• The developed deep learning model provides an efficient and accurate solution for collimating lens optical performance evaluation.
• The model, integrated into application software, is suitable for real-time quality inspection in manufacturing.
• This intelligent detection system significantly enhances the quality control process for collimating lenses.