NSF Postdoctoral Fellowship in Biology FY 2016

Kuhn William R, Bloomfield NJ

Investigators

Abstract

William R. Kuhn Proposal Number: 1611642 This action funds an NSF Postdoctoral Research Fellowship in Biology for FY 2016, Research Using Biological Collections. The fellowship supports a research and training plan for the Fellow to take transformative approaches to grand challenges in biology that employ biological collections in highly innovative ways. The title of the research plan for this fellowship to William R. Kuhn is "Leveraging face-detection methods to identify insects from field photos, automatically." The host institution for this fellowship is the University of Tennessee (Knoxville), and the sponsoring scientist is Dr. Mongi Abidi. This work aims to automate the identification of species, thereby helping to alleviate the 'taxonomic impediment,' i.e., an urgent need for more taxonomy from fewer taxonomists. The impetus for doing this research is that although understanding Earth's species is one of the grand challenges of the twenty-first century, training and funding for the field of taxonomy (identifying and describing species) has declined markedly over the past several decades. The Fellow is integrating existing computer vision and machine-learning methods to build an automatic system for identifying species from photographs of them in their habitat. The Fellow has three main research objectives: (1) develop and train a model to locate the subjects in images by modifying an existing method for detecting human faces in photographs; (2) characterize the images based on the appearance of the subjects' body parts by adapting an algorithm for describing the key features of an image; and (3) predict species identity by comparing features from unknown images to those of known species, utilizing a robust machine-learning framework. The Fellow is developing a system to identify dragonfly and damselfly (Odonata) species, but the underlying code will allow researchers to train systems for other organisms. The Fellow is utilizing citizen scientist data on OdonataCentral (odonatacentral.org), a digital collection of species records and imagery of Odonata from the Western Hemisphere. Images of the 600+ species included in this digital collection are being used to train the most taxonomically-broad automatic identification system ever created, and the ability of this software to accept field-based images makes it extremely versatile. The Fellow is receiving advanced training in image analysis, computer vision, and machine learning, and becoming proficient in software design and programming. The Fellow will be able use these skills to solve problems in biology in the future, when he is running his own lab. The Fellow is also developing his skills as a mentor and scientific communicator. He is mentoring two undergraduates and one graduate student in computer science, encouraging them to focus on solving biological problems. The odonate identification system is being integrated into OdonataCentral as well the mobile app that it powers, Dragonfly ID. This will allows both researchers and citizen scientists to make rapid identifications in the field, for applications such as assessing biodiversity and monitoring water quality. Ultimately this research will benefit taxonomists studying other organisms, since the Fellow is releasing his source code, allowing others to train and implement their own automatic identification systems.