Scientists from a trio of universities have combined satellite imagery with deep learning to detect elephants from space. The goal is to help protect these endangered species from poachers or habitat destruction. Their study was published in the journal Ecology and Conservation. Interesting Engineering reports: The team’s method proved comparable to human detection accuracy and could help solve a number of existing challenges, such as cross-border limits, and cloud coverage, among others. The team used Maxar WorldView-3 satellite imagery, which is capable of collecting more than one million acres (5,000 km2) imagery in one go in just a few minutes. This allows for fast repeat imaging when necessary, and minimizes the risk of double counting as it’s so rapid.
Then the team leveraged deep learning to process the vast amount of imagery it collected from Maxar’s WorldView-3 satellite. In a matter of hours, the team collected its relevant data. This process usually takes months when sorting out by hand. On top of speed, the deep learning algorithms also provided consistent results less prone to error, as well as false negatives and false positives. In order to develop this method, the team created a customized training dataset of over 1,000 elephants, and then fed it into a Convolutional Neural Network (CNN). After trials, the team concluded that its CNN can detect elephants in satellite imagery with as high an accuracy as human detection capabilities.