Answers are grounded in this paper and cite their sources.
Precise and non-invasive cattle identification is essential for modern livestock management, supporting traceability, health assessment, and data-informed decision-making. Traditional methods like ear tags and RFID are prone to loss and damage, whereas single-modality biometric systems, such as facial or muzzle identification, struggle with occlusions, varying lighting, and unpredictable farm environments. To address these constraints, this study introduces CattleBioNet, an innovative multi-modal biometric identification framework that combines complementary cattle facial and body features through an Adaptive Feature Fusion (AFF) mechanism within a dual-backbone architecture that integrates Faster R-CNN and YOLOv8-ResNet18 networks. The Adaptive Feature Fusion (AFF) module dynamically integrates multi-scale features, assigning spatially varying weights to highlight the most informative modalities. The fused features are processed by a Faster R-CNN detection head for accurate localization and identification of cattle faces and bodies. Experimental results demonstrate that CattleBioNet achieves state-of-the-art performance, achieving a mean Average Precision (mAP@0.5) of 98.6%, which establishes a new benchmark for cattle identification by effectively leveraging complementary biometric modalities through adaptive fusion.
No takes yet. Share an insight, caveat, or question.
Kofa et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: