A Serial Multi-Scale Feature Fusion and Enhancement Network for Amur Tiger Re-Identification.

Area of Science:

• Conservation Biology
• Computer Vision
• Wildlife Management

Background:

• Amur tigers (Panthera tigris amurensis) are critically endangered, making accurate population monitoring vital for conservation efforts.
• Re-identification (re-ID) of Amur tigers using visual data is challenging due to variations in camera perspective, background noise, motion, and body pattern deformation.

Purpose of the Study:

• To develop an accurate and efficient Amur tiger re-identification system using visible light images.
• To address the limitations of existing re-ID methods in handling complex environmental and physical variations of Amur tigers.

Main Methods:

• Proposed a serial multi-scale feature fusion and enhancement re-ID network with distinct global and local branches.
• Implemented a global inverted pyramid multi-scale feature fusion method for preserving high-level semantic features.
• Developed a local dual-domain attention feature enhancement method to improve local feature extraction and fusion.

Main Results:

• The proposed network achieved competitive results on the Amur Tiger Re-identification in the Wild (ATRW) dataset, demonstrating strong performance in mAP, Rank-1, and Rank-5 metrics.
• The model effectively fused multi-scale global and local features, enhancing the accuracy of Amur tiger re-ID.
• The system showed good transferability and simple training without requiring additional annotation or pre-training modules.

Conclusions:

• The developed Amur tiger re-ID network offers a promising solution for accurate identification in challenging conditions.
• The method contributes to improved biodiversity assessment and wildlife resource statistics for endangered Amur tigers.
• The model's efficiency and transferability make it a valuable tool for practical conservation applications.