[8a-N205-1~9] CS.2 Code-sharing session of 3.2&4.4

11:00 AM - 11:15 AM JST(2:00 AM - 2:15 AM UTC)English Presentation

[8a-N205-7] A Cascaded Classification Framework for Small Datasets with Data Augmentation Support

〇(M2)yen yu liu¹, Jia-Han Li¹ (1.Dept. of ESOE, Natl. Taiwan Univ.)

Keywords:

sea turtle identification,deep learning,data augmentation

This study presents a cascaded classification framework designed for sea turtle facial recognition using small datasets. The system applies three YOLO-based classifiers to sequentially identify species, count facial scales, and analyze scale features. Data augmentation methods such as Gaussian noise and CutMix significantly improved performance, especially for fine-grained classification. The framework achieved over 98% accuracy on key tasks and effectively reduced manual verification workload, offering a practical solution for marine conservation efforts.