Research on Small Sample Dynamic Human Ear Recognition Based on Deep Learning.

Area of Science:

• Computer Science
• Biometrics
• Artificial Intelligence

Background:

• Dynamic human ear recognition research is hindered by a lack of comprehensive datasets.
• Existing datasets often fail to capture the variability of real-world conditions.

Purpose of the Study:

• To introduce the Changchun University dynamic human ear (CCU-DE) database, a novel small sample dataset.
• To evaluate the effectiveness of deep learning models for dynamic human ear recognition using this new database.

Main Methods:

• Developed the CCU-DE database, incorporating diverse conditions like pose changes, illumination variations, and occlusion.
• Designed a dynamic human ear recognition system utilizing deep learning models (YOLOv3-v5, Faster R-CNN, SSD) pre-trained with transfer learning.
• Conducted simulation studies on the CCU-DE database to assess model performance under various challenges.

Main Results:

• The CCU-DE database effectively supports dynamic human ear recognition tasks.
• Multiple deep learning models demonstrated high test accuracy when evaluated on the CCU-DE database.
• The database's design enhances the applicability of dynamic ear recognition in complex, real-life scenarios.

Conclusions:

• The CCU-DE database is a valuable resource for advancing dynamic human ear recognition research.
• Deep learning models, particularly when pre-trained, show significant promise for accurate ear recognition using the CCU-DE dataset.
• The developed database facilitates more robust and practical dynamic human ear recognition systems.