Bird counting with acoustic sensing

Everybody likes birds, which makes them a great target for sensing. There are already a lot of platforms available for recording, detecting, identifying, and counting birds, and I decided to deploy some of them, largely for practice but also to feed-into some work we’ve been doing on target counting and estimation.

The starting point was BirdNet-Pi, an acoustic bird-identification platform running on a Raspberry Pi and making use of the BirdNET platform developed at Cornell Lab of Ornithology. The great part is that all the recognition runs locally, making use a pre-trained machine learning model.

The more traditional approach separates these two functions (collection and classification), with the former in the field and the latter offline. This is cheaper and easier to power, but more labour-intensive in terms of collection and analysis — as well as being a lot less engaging for a user (like me).