Unmanned Aerial Vehicles and Automated Photo Recognition for Eagle Carcass Detection at Wind Energy Facilities
Presentation Description: Eagle take permits for operating wind projects require life of permit monitoring for compliance. Developing efficient and accurate survey methods for eagle carcasses helps alleviate cost concerns related to monitoring requirements. We investigated use of unmanned aerial vehicles, or drones, to efficiently complete carcass searches for eagles. Phase 1 examined whether images takes from a UAV could be searched using a machine learning neural network to locate eagle carcass surrogates. Images were taken in five ground cover types that are common in wind projects in the Midwest. The better of two trained neural networks detected 93% of all decoys and 100% of all feather spots across all ground cover types with an 8.3% false positive rate. Phase 2 evaluated images taken at 120-m above ground level and how fast a 100-m radius plot could be searched. Two pre-planned UAV transect missions were flown and collected images were run through the neural network developed from Phase 1. Depending on the transect mission, 50.0% of the six feathered turkey decoys with one false positive and from 66.7 to 100% of the feather spots with 6 false positives were detected. The entire plot was photographed 1 to 2 minutes depending on flight speed. The neural network detection rates were generally higher than field biologists completing traditional scan or walking surveys. We conclude that UAV photography followed by automated carcass detection appears to be a viable, reliable, safe, and efficient method of searching for eagle carcasses in areas surrounding wind turbines. Additional field testing and training is needed at higher flight altitudes in diverse vegetation types where eagle fatalities may occur.