TaxonomyNet: A Consistent and Efficient Model for Taxonomic Rank Identification in Wildlife Images

Accurate and scalable taxonomic classification is essential for biodiversity research, supporting systematic species identification across multiple hierarchical ranks. However, current image-based classification methods often fail to enforce taxonomic consistency, limiting their utility in ecological monitoring and species discovery. Additionally, field-based biodiversity studies are constrained by limited computational resources and network availability on edge devices. To address the above challenges, this paper proposes an ensemble detection model, TaxonomyNet, that integrates six independent heads corresponding to different taxonomic classifications. To improve prediction consistency across taxonomic ranks, we introduce the Weighted Agreement Loss (WAL) metric—a confidence-weighted disagreement measure designed to quantify and minimise inconsistencies between predicted outputs and a reference taxonomy. TaxonomyNet achieves high detection performance across all ranks (mAP: 90.7%–99.75%) after training on a dataset of 50 Australian animal species with taxonomic classification labels. Compared with both lightweight and large-scale foundation models, the proposed method improves hierarchical classification reliability (by up to 3.87%) and computational efficiency (reducing delay by 22 minutes across 1,500 samples) on edge devices. This work provides a practical and extensible solution for hierarchical classification in real-world biodiversity monitoring scenarios.