NA
Master thesis
Permanent lenke
https://hdl.handle.net/11250/3137413Utgivelsesdato
2024Metadata
Vis full innførselSamlinger
Sammendrag
NA In recent years, passive acoustic monitoring (PAM) has emerged as a powerful tool for monitoring avian biodiversity. However, a major challenge has been to develop algorithms that can process large amounts of data, and at the same time, correctly identify bird sounds to species. There are several algorithms available for the identification of bird sounds but, the most prominent ones are Merlin Audio ID and BirdNet. In this study, a comparative analysis of the Merlin and BirdNet applications was carried out to evaluate their accuracy and efficiency in the identification of bird species from PAM recordings. A human-trained ear was used as a baseline to evaluate the accuracy of these algorithms. The number of false negatives i.e. bird species detected by the human ear but not by the app and false positives i.e. bird species detected by the apps but not by the human ear were estimated. Merlin correctly identified 39.8% of the bird species identified by the human-trained ear. 44.2% were false negatives and 16% were false positives. BirdNet on the other hand, correctly identified 24.6% of the bird species identified by the trained- human ear. 70.6% were false negatives and 3% were false positives. A significative difference (P < 0.05) in the number of birds detected between Merlin, BirdNet, and the human-trained ear. Finally, inconsistency in bird detection of the Merlin application, when analyzing the same recordings for a second time, was found (P= 0.006). Although both applications show promising results, they still need many improvements for optimal performance.