A scalable thin-film defect quantify model under imbalanced regression and classification task based on computer...

Area of Science:

• Materials Science
• Computer Science
• Optical Engineering

Background:

• Traditional optical coating damage detection relies on costly expert systems or manual inspection.
• Adapting existing methods to new film types or inspection environments is time-consuming and expensive.

Purpose of the Study:

• To develop an automated, adaptable, and cost-effective method for optical coating damage detection.
• To improve the speed and accuracy of identifying various damage types and their severity.

Main Methods:

• A deep neural network (DNN) model was developed for optical coating defect detection.
• The DNN approach bifurcates the task into damage classification and damage degree regression.
• Attention mechanisms and Embedding operations were integrated to enhance model performance.

Main Results:

• The proposed DNN model achieved 93.65% accuracy in damage type detection.
• Regression loss for damage degree was maintained within 10% across diverse datasets.
• The model demonstrated adaptability to different coating types and inspection scenarios.

Conclusions:

• Deep neural networks offer a powerful solution for industrial defect detection, significantly reducing costs and design time.
• The developed DNN tool provides an adaptable and efficient alternative to traditional expert systems.
• This approach enables the detection of novel damage types cost-effectively.