DiversityScanner: Robotic handling of small invertebrates with machine learning methods

Area of Science:

• Ecology and Evolutionary Biology
• Bioinformatics and Computational Biology
• Robotics and Automation

Background:

• Invertebrate biodiversity is vast but poorly understood due to challenges in specimen processing.
• Traditional manual sorting is labor-intensive, and existing molecular methods lack precise abundance data.

Purpose of the Study:

• To develop an automated system for preparing invertebrate specimens from bulk samples for DNA barcoding.
• To integrate imaging and machine learning for automated taxonomic assignment and biomass estimation.

Main Methods:

• A novel sorting robot detects, images, measures, and plates individual specimens from bulk samples.
• Convolutional neural networks (CNNs) were trained on specimen images for taxonomic classification.
• Specimen length and volume were estimated from images for biomass assessment.

Main Results:

• The robot successfully prepares specimens for barcoding, automating a key bottleneck in biodiversity research.
• CNNs achieved an average assignment precision of 91.4% for 14 common insect taxa.
• Automated image analysis enables taxon-specific subsampling and biomass estimation.

Conclusions:

• The DiversityScanner robot, combined with machine learning and DNA barcoding, offers a scalable solution for invertebrate diversity assessment.
• This integrated approach has the potential to significantly advance the study and monitoring of invertebrate diversity at an unprecedented scale.