yolov8_arcticbeavers

This is a yolov8 based object detection model for beaver dams and lodges from aerial imagery

This is a semi-serious side-project to detect beaver dams and lodges from aerial imagery. Beavers are expanding into Arctic regions, which can be even observed indirectly from space. With very-high resolution data from UAV or airborne missions, we can try to map dams and lodges directly.

More cool information on beaver expansion into the Arctic:

• Tape, K. D., Clark, J. A., Jones, B. M., Kantner, S., Gaglioti, B. V., Grosse, G., & Nitze, I. (2022). Expanding beaver pond distribution in Arctic Alaska, 1949 to 2019. Scientific Reports, 12(1), 7123. https://doi.org/10.1038/s41598-022-09330-6
• Jones, B. M., Tape, K. D., Clark, J. A., Nitze, I., Grosse, G., & Disbrow, J. (2020). Increase in beaver dams controls surface water and thermokarst dynamics in an Arctic tundra region, Baldwin Peninsula, northwestern Alaska. Environmental Research Letters, 15(7), 075005. https://doi.org/10.1088/1748-9326/ab80f1
• Tape, K. D., Jones, B. M., Arp, C. D., Nitze, I., & Grosse, G. (2018). Tundra be dammed: Beaver colonization of the Arctic. Global Change Biology, 24(10), 4478–4488. https://doi.org/10.1111/gcb.14332

Info

• Model file in pytorch format for ultralytics yolov8

• bounding boxes of beaver dams and lodges

• trained on aerial imagery from West and Northwest Alaska

  • More info: https://essd.copernicus.org/preprints/essd-2023-193/

• This model takes RGB aerial images in high spatial resolution, suhc as UAV or airborne imagery. It was trained on images from tundra regions in NW Alaska.

• Target objects were hand labelled with roboflow --> https://app.roboflow.com/awi-response/beaver-finder-vhr-imagery-a9hg9/

Related Code

github

https://github.com/initze/yolov8_object_detection/

Input data

RGB images

Known and potential issues

• false positives for curved shore areas

Classes

1: beaver dam 2: beaver lodge 3: building (not great)

Input data

Examples

The good ones

The bad ones