Automated species monitoring using deep learning-based detectors

In order to protect the nature, it can be useful to investigate the distribution, abundance and behavior of certain species. This can sometimes be very difficult like when the animals you want to investigate, either hides in daylight like bats, or live beneath the surface of the sea where we cannot see them, like white-beaked dolphins.

Passive acoustic monitoring is a technique where researchers place a self-contained recording device to record all sounds around it and leave it for months- or even years. After retrieving the device, it is then possible to identify specific species by the calls or songs we know they produce. The hardware for this kind of study is consistently becoming cheaper and more capable and hence it is becoming an increasingly important tool for the scientific community. Unfortunately, the manual process of going through months- or years’ worth of data can take a very long time and the subjectivity of human observers can compromise the results. Recent developments in deep learning have shown great results in classifying sounds. Most people have experienced how well it works through translation apps on their phones. My project will develop deep learning-based detectors to be used in combination with passive acoustic monitoring. We thereby get a highly valuable tool for monitoring species like bats, white-beaked dolphins, and grey seals.