UAS and Machine Learning Protect Endangered North Atlantic Right Whales
Tuesday, October 06, 2020: 12:05 PM - 1:00 PM
North Atlantic Right Whales (NARWs) are one of the most endangered whale species in the world. Lethal ship strikes have been identified as a major threat to these whales. Efforts are underway to reduce ship strikes by identifying when whales enter shipping lanes, then enacting speed restrictions for nearby vessels. The Canadian Government is regularly flying manned aircraft over shipping lanes with trained marine mammal observers who visually scan for whales. This method is quite effective but incurs significant costs, risks, and resource allocation to be successful. A new program is attempting to streamline these operations using Unmanned Aircraft Systems (UAS) and machine learning software designed to automatically detect whales near shipping channels. UAS with high-resolution cameras can efficiently capture imagery of a shipping lane with 100% coverage. UAS imagery and associated telemetry are processed using machine learning algorithms to automatically detect and geo-locate whales. During summer 2018 and 2019, UAS missions were flown off the East coast of Canada. Each mission produced thousands of ultra-high resolution images, which were processed by Planck’s OpTrack whale detection software. The machine learning algorithm generated candidate whale “detections”, which included marked-up images and geo-coordinates. Trained biologists were ultimately responsible for confirming detections. The automated software was able to drastically reduce the search time for the biologists, thereby minimizing human fatigue and improving detection efficiency. This talk focuses on improvements that were made to the automated detection system in 2019. Data from 2018 was used to re-train the whale classification algorithm, which improved detection results in 2019. However, higher resolution cameras used in 2019 required optimizations to the software architecture. Interface improvements significantly reduced the timeline from data collection to operator review.
Analyst,Educator/Instructor,Engineering/Technical,Government/Policy/Regulatory,Research & Development,Student