Citation

Duque-Montoya, Diana Carolina; Isaza, Claudia (2018). Automatic Ecosystem Identification Using Psychoacoustical Features. the International ConferenceProceedings of the International Conference on Pattern Recognition and Artificial Intelligence - PRAI 2018.

Abstract

The recent changes in worldwide ecosystems require constant monitoring and actions to prevent habitat destruction and mass extinction of species. The development of Passive Acoustic Monitoring (PAM) techniques has allowed the study of ecosystem dynamics, producing no harm to them. In this study, a relation between psychoacoustical features and ecosystem degradation was explored. Psychoacoustical features are used mainly for urban soundscapes categorization and they are based on human subjective perception of sound; however, they provide a starting framework for linking soundscape to spatial structure. To map soundscape to ecosystem health, an artificial neural network (ANN) and three different types of support vector machines (SVM) were tested for automatic ecosystem type identification using the selected features. Three out of four models leaded to good results, but as future work, it is suggested to adapt psychoacoustical features to a more neutral auditory system so the analysis includes more species perception of the ecosystem soundscape.