juil. 2017 - Présent
台灣 台北市
[Projects]
╴TFT-LCD defect detection
╴Scratch detecion
╴Algorithms system Development & Maintain
[Detail]
╴TFT-LCD defect detection
Achieve the accuracy 85-95%+, 5~30 categories (depend on different projects)
improve the accuracy by 8%(compared with the previous)
╴Transfer Learning (transfer the model that trained from domain A to domain B (same categories as domain A) without label data of domain B)
Increase the accuracy from 13%(before transfer) to 82%(after transfer) (11 categories).
╴Scratch detection by only the golden image
╴The techniques have been studied
Inbalance data Handle
Domain Adaptation
Semi/Self supervised method
Data Augmentation
Noise data detection
New class detection
Abnormal detection