Abstract:
Accurate, non-invasive individual identification is critical for endangered Amur tiger (Panthera tigris altaica) conservation. However, it remains challenging due to pose variability in stripe patterns and inconsistent imaging conditions. Inspired by human fingerprint recognition, we propose TigerNet, a hybrid framework that integrates biological feature (stripe geometry and orientation) with deep similarity networks to identify tiger individual. TigerNet adopts a three-stage pipeline: ) (1)adaptive feature enhancement, which transforms tiger stripes into a fingerprint-like representation to preserve stripe-specific information and reduce background interference; (2) hierarchical representation learning, leveraging a modified EfficientNetV2Small to extract discriminative feature embeddings; and (3) discriminative similarity verification, using a hybrid loss function (CosineLoss + Euclidean distance regularization) to optimize stripe angles and orientations against pose variations. The effectiveness of TigerNet is verified through experiments conducted on 12,625 tiger images captured from Siberian Tiger Park, Harbin, China. TigerNet achieves 94.96%/96.45% accuracy on left/right body images, out-performing pure CNN baseline method. Ablation studies confirm that CosineLoss enhances feature discriminability (reducing inter-class overlap by enforcing angular separation) and stripe feature enhancement cuts recognition time by 37%. TigerNet demonstrates strong cross-species generalization, achieving 90.2% accuracy on a public wild zebra dataset while maintaining robust recognition under pose variations. It consistently outperforms the MiewID method, improving accuracy by 2.5–5.7% on tiger data and 5.7% on zebra recognition. This work provides a robust, non-invasive tool for wildlife monitoring that balances accuracy, interpretability, and scalability, with potential generalization to other endangered species with unique morphological patterns.
Shi, C., Fan, H., Roberts, N.J. et al. TigerNet: individual identification of Stripe patterns in Amur Tigers like human fingerprint. Mamm Res 71, 16 (2026). https://doi.org/10.1007/s13364-025-00836-8
