There is tremendous interest in learning how whale populations are recovering following the end of intense commercial whaling, yet their large ranges and mobility make this a significant challenge.
Coastal areas sometimes concentrate populations at key times of the year and are logistically easier to survey. Aerial surveys have proven an effective way to locate and study movements but can be costly and are restricted by aircraft range.
A collaboration of ocean ecologists, including HiDef Aerial Surveying Ltd from the UK, Stony Brook University in New York, and BioConsult SH from Germany have created a semi-automated process for whale detection from very high-resolution satellite images using deep learning artificial intelligence. Trial images were successfully used from Hawaii and Argentina in areas known to host whales.
Using down-scaled HiDef digital aerial images and satellite images, the trained algorithms were used to classify whether a tile image was likely to contain a whale. The best model correctly classified 100% of tiles with whales, and 94% of tiles containing only water. While the relatively poor resolution of commercially available satellite images continues to make whale identification a challenging problem, our approach provides the means to efficiently eliminate areas without whales and, in doing so, greatly accelerates ocean surveys for large animals.
Dr Grant Humphries, one of the lead authors of the paper said, “the application of this algorithm to satellite imagery will have vast implications towards how we survey large marine mammals in the open ocean”. HiDef and BioConsult SH will be launching a demonstration project of the technology through 2020.
Further reading: Borowicz A, Le H, Humphries G, Nehls G, Hoschle C, Kosarev V, et al. (2019) Aerial-trained deep learning networks for surveying cetaceans from satellite imagery. PLoS ONE 14(10): e0212532. https://tinyurl.com/spacewhale2019