After ecological quick scans, research into bats often needs to take place. At the moment this is a time-consuming activity and also very expensive. In addition, there are too few ecologists available for all research to meet the demand. This is not surprising: it is night work, which not everyone likes.

This was the reason for Aquila Ecology to start with an AI model to identify bats. In collaboration with the NDFF, we have created an AI model that can identify bat individuals by species based on sound with 91% accuracy. These sounds must be recorded with special recording equipment that can record high sound frequencies. This can then be determined by the developed AI model. But how does developing such an AI model actually work technically?

First of all, we need a large dataset of bat sounds whose identification has been established with certainty. This is easy with some bat species, for example there are many recordings of the common pipistrelle bat. However, there are also species that make sounds that are difficult to distinguish from other species. These sounds are often recorded before or after animals are captured and identified by appearance. There are far fewer sounds available.

Machine learning actually needs the widest possible variety of sounds. In this way, the computer learns to distinguish what exactly is typical about the call of a particular bat species, and what happens to be a variation or background noise. To artificially increase the amount of data, we use a trick called data augmentation. We mix the sounds on which the AI ​​is trained with different background sounds to add variation. We had to collect these background sounds first, because they have to be recorded at high frequency.

All data is then 'viewed' repeatedly by an algorithm. This algorithm always adjusts itself slightly based on the data. It searches for patterns in the data in a complex way. Every time the algorithm 'looks' at the data, it gets a little better at recognizing bats. The Aquila Ecologie software indicates this while training the model:

The next step is to make this technique useful for analyzing large data sets of, for example, several weeks. To achieve this, the software must not only be able to distinguish bat species, but also know when there is no bat in the recording. We do this by recording a lot of ambient noise and teaching the software to distinguish between them.