SpeciMate: Improving metadata extraction from digitised biological specimens

The digitisation of natural history collections represents a critical step towards preserving and increasing accessibility to valuable scientific data. Despite their fundamental importance to taxonomy, ecology and conservation, the world's natural history collections remain underutilised due to the labour-intensive process of extracting metadata from specimen labels.

This paper describes SpeciMate, a software application that uses a human-AI collaborative approach to accelerate the extraction of metadata from digitised specimen images. The system leverages artificial intelligence web services including optical character recognition (OCR), automated translation and large language and multimodal models (LLMs) to extract structured metadata, while requiring human expertise for prompt engineering and data curation. We describe the application's architecture, functionality and workflows, which enable effective processing of various specimen types including herbarium sheets and insect slides. Our trials indicate that this tool significantly improves the efficiency of metadata extraction while maintaining high data quality. The combination of automated AI processing with human supervision and refinement represents a promising approach to accelerating the digitisation and databasing of natural history collections, thereby enabling broader access to these invaluable resources for research, education and conservation efforts.