nose-to-id: a deep learning framework for dog identification using nose-print biometrics

Accurate identification of individual dogs plays a crucial role in various applications including pet recovery, veterinary management, and animal welfare. this study proposes a fully automated dog identification framework based on unique nose-print biometric patterns, leveraging deep learning techniques to overcome limitations of traditional identification methods. the proposed approach processes user-submitted videos by selecting the most frontal frame via head pose estimation, detects the nose region using a fine-tuned yolov8 model, and extracts discriminative embeddings through a multi-resolution resnest-based convolutional network enhanced with advanced augmentation strategies. the resulting embeddings are fused to produce robust identity descriptors capable of distinguishing between thousands of individual dogs. experimental results demonstrate the system’s efficacy under real-world conditions, emphasizing its potential for practical deployment in pet identification and management systems.