Trail Camera Data Classification with AI (primary focus on Fishers)

Description

An AI accelerated workflow begins with detection of the camera trigger with the use of MegaDetector, then species classification is initiated with either an already trained classifier (SpeciesNet) or a classifier trained on NJFW data (Mega Efficient Wildlife Classifier). Once classification is complete, human verification then takes place in a database (Timelapse). According to the creators of MegaDetector, it can detect humans, vehicles, animals, or empty images at a rate of ~9.5 images per second using a GPU such as the NVIDIA A4000 (or similar). The Mega Efficient Wildlife Classifier and SpeciesNet will perform at similar rates on the NVIDIA A4000. At these rates, detection of wildlife and classification of footage can be completed within days and human verification can be completed in a few weeks to months compared to years without AI classification. AI accelerated workflow will significantly decrease the time and effort required, thus resulting in a more achievable goal. The goal of implementing an AI accelerated workflow is achievable with current staff that have knowledge and experience with the programs and workflow

Project Justification

The data collection aimed at understanding Fishers through a grant-funded 5-year project ('22-'27) involving 150 trail cameras will support new decision making for management of Fishers and will also generate lots of new insight about multiple species. The AI applications for image processing needed for the Fisher-focused data can also render multi- species info (possibly including human activity insights), and it can be deployed on previous image collections that fell short on human resources for image processing (notably the bobcat study set).

Attachments

Activity