Photo Data Collection
Job Description
NPS UNIT: South Florida Carribean Inventory and Monitoring Network
LOCATION: Palmetto Bay, FL
Currently, the South Florida / Caribbean Inventory and Monitoring Network (SFCN) produces and processes thousands of aerial photographs per year, counting and categorizing colonial bird nests in order to measure the corresponding bird populations at six locations across Biscayne National Park. Manual processing of these photographs (which involves identifying valid nests in each photograph, categorizing the nest to a species and a stage of development, along with checking adjacent photographs to avoid nest double-counting) currently requires hundreds of person hours per year.
Advances in computer vision have produced computer programs capable of identifying and counting a wide range of objects in digital images with performs equitable to processing by humans. This project intends to automate the SFCN nest identification process by applying existing techniques in computer vision and machine learning.
Using something like the Annotation Interface for Data-driven Ecology (AIDE), an open-source annotation service with a built-in machine learning model could be used to process photos. Once the machine learning model is properly trained, AIDE or other algorithm will be able to automatically identify and “draw” rectangular boxes around those bird nests, as well as provide an interface for an user to do the same. Ideally, this project will build an algorithm that receives sets of raw aerial photographs and outputs an image with potential nest identified and categorized by nest type.
Position description:
Computer Science Researcher working on natural resource data collection. The incumbent will have the ability to use software program to analyze photographic data and determine the present of nesting birds. Preferably using machine learning (i.e. artificial intelligence) to be able to determine if a bird nest is present on a photograph and if so graphically indicate that. Then to compare the software outcome against human processed data collection that is currently being done.