SpeciMate: Improving metadata extraction from digitised biological specimens

Area of Science:

• Biodiversity Informatics
• Digital Curation
• Natural History Collections

Background:

• Digitization of natural history collections is crucial for data preservation and accessibility.
• Underutilization of collections stems from labor-intensive metadata extraction from specimen labels.
• Metadata is vital for taxonomy, ecology, and conservation research.

Purpose of the Study:

• To introduce SpeciMate, a software application designed to expedite metadata extraction from digitized natural history specimens.
• To present a human-AI collaborative framework for efficient and accurate data processing.
• To detail the architecture, functionality, and workflows of the SpeciMate application.

Main Methods:

• SpeciMate employs artificial intelligence (AI) web services, including optical character recognition (OCR), automated translation, and large language/multimodal models (LLMs).
• A human-AI collaborative approach is utilized, with AI for extraction and human expertise for prompt engineering and data curation.
• The system is designed to process diverse specimen types, such as herbarium sheets and insect slides.

Main Results:

• SpeciMate significantly enhances the efficiency of metadata extraction from specimen images.
• The application maintains high data quality throughout the metadata extraction process.
• Trials demonstrate the tool's effectiveness across various specimen formats.

Conclusions:

• The human-AI collaborative approach in SpeciMate accelerates the digitization and databasing of natural history collections.
• This approach improves accessibility to valuable scientific data for research, education, and conservation.
• SpeciMate offers a promising solution for overcoming bottlenecks in natural history collection management.