Multimodal Fusion of Visual and Morphometric Features for Avian Bone Classification

This proof-of-concept study combines images and osteometric measurements to classify avian skeletal remains. Its pipeline uses BiRefNet and SAM2 for segmentation, a pretrained EfficientNet_V2_S backbone for visual features, and feature-level fusion with standardized morphometric data. On a dataset of more than 10,000 images, the reported test accuracy is 86% for skeletal-element identification. Family-level taxonomy is harder, with 51% top-1 and 75% top-3 accuracy. The abstract does not provide class counts, split methodology, modality ablations, or cross-collection generalization results.