A Computer Vision Method for Finding Mislabelled Specimens Within Natural History Collections

Area of Science:

• Biodiversity research
• Evolutionary biology
• Museum informatics

Background:

• Natural history collections are vital for biodiversity and evolution research.
• Specimen mislabelling poses challenges for collection management and research integrity.
• Existing verification methods like genetic analysis can be resource-intensive and damage specimens.

Purpose of the Study:

• To develop and apply a computer vision pipeline for automated taxonomic verification of digitised specimens.
• To identify mislabelled specimens within large natural history collections.
• To enhance the efficiency and accuracy of natural history collection management.

Main Methods:

• A computer vision pipeline was developed and applied to the digitised British and Irish Lepidoptera collection at the Natural History Museum (NHM).
• The pipeline identified specimens with potential mislabelled species status.
• Specimens flagged consistently were visually examined by taxonomic experts and some were selected for genetic verification.

Main Results:

• The computer vision pipeline flagged 99,350 out of 350,208 specimens (28.37%) as potentially mislabelled.
• Expert examination of 210 consistently flagged specimens revealed 145 (69%) were indeed mislabelled.
• A combination of computer vision and genetic analysis improved identification accuracy.

Conclusions:

• Computer vision offers an innovative, non-destructive method for taxonomic verification in large natural history collections.
• Automated identification of mislabelled specimens significantly enhances collection data quality.
• The synergy between computer vision and genetic analysis improves management and preserves collections for future research.