Hi! I'm trying to estimate abundance and survival in freshwater fish populations (trout and salmon) using AI for individual recognition from crowdsourced data: https://www.usgs.gov/centers/eesc/science/enabling-ai-citizen-science-fish-biology
We are seeing good performance of our individual ID models so far - at least for species with distinct spot and parr mark patterns (> 90% accuracy over seasons and years), so we are cautiously optimistic this approach will be useful for population dynamics. And we hope to move this forward with a new collaboration this year involving Trout Unlimited (social network of anglers and conservationists) and Wild Me (imagery platform and individual ID pipeline).
Are others working along similar lines? Any ideas or questions are welcomed. You can reach me here or at nhitt@usgs.gov. Thanks!
Hitt_ESIP2020.pdf5 May 2023 6:15pm
That sounds like a really neat project! Do fish get re-caught often enough that individual ID is useful? Is sample bias (more data from popular spots) an issue?
Phil Mitchell