Collaborative Research: BIO-AI: Human-in-the-loop Artificial Intelligence System for High Throughput Automated Identification of Ground-dwelling Arthropods

Insects and other arthropods make up almost 85% of all known animals on the planet, and most existing species are still unknown to science. For centuries, researchers have collected arthropod samples from habitats worldwide, many of which contain thousands of specimens. The time and expertise required to sort through these samples and make identifications are prohibitive. Consequently, hundreds of millions of specimens, representing vast amounts of undiscovered biodiversity, are unstudied and stored away in museums and research collections. There are few scientists capable of working on the backlog of specimens, and therefore, samples continue to sit in storage. This project will develop artificial intelligence-based tools to automate the imaging and identification of arthropods in samples from terrestrial habitats. Researchers will simply pour a raw sample into an imaging system, allowing the computer to image and identify the specimens, thereby greatly minimizing the time and expertise required to process samples. Successful development of this tool will have profound impacts on both ecology and biodiversity sciences, as it will allow researchers to extract more data from samples than previously possible and will unlock tremendous amounts of biodiversity data from the immense backlog of samples. Since many of these samples are decades or centuries old and come from habitats that have degraded or destroyed, identification of specimens will provide data on how biodiversity is changing through time and provide critical information about species that have gone extinct. The advanced AI algorithms developed in this project can be further extended in healthcare, robotics, manufacturing, or new material discovery applications, and used to accelerate use and development of other AI systems. The project also will provide training opportunities in entomology or computer science to pre-college, undergraduate, and graduate students. This project will facilitate the incorporation of all arthropods into studies examining soil samples, providing an efficient, user-friendly, and low-cost method to unlock the tremendous amounts of dark biodiversity data in research collections. The AI identification system will be able to identify soil arthropods from across the US and will continually learn and expand its recognizable diversity as new images and identifications are added. Additionally, the proposed AI methods go beyond applications in arthropod identification. Our developed technologies will automate image segmentation and provide a novel approach to learning with limited annotated data and fairness awareness. The developed continual learning approach provides an efficient way to adaptively update the model with new data that helps the model improve over time. Therefore, the outcomes of this project are expected to fundamentally advance the fields of biodiversity science and ecology and promote the development of broadly applicable computer vision and machine learning technologies. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.