First completely free bioacoustic analysis tool with included Auto-ID powered by deep neural net BigBAT
Neuroinformatics Project, graded 1.0 (best grade)
The automatic identification of bat species from their echolocation calls is a difficult but crucial task for monitoring bats and the ecosystems they live in. One of the main challenge is the lack of annotated data, since annotating echolocation data requires exceptional expertise and is very time consuming. Here, we experimented with different methods to incorporate unlabeled data to improve the performance of an existing model. We were able to improve performance of mixed macro-F1 by 2.73%, single accuracy by 10.69% and single F1 by 4.96% on a dataset of South African bats from the university of free state (UFS). We also proposed a method called „genus smoothing“, which was able to increase the macro-F1 by 0.47%, single test accuracy by 0.42% and single F1-score
by 4.73%. In addition, we have developed a free bioacoustics analysis tool, including automated identification.