Automated detection of broadband clicks of freshwater fish using spectro-temporal features

Navinda Kottege, Raja Jurdak, Frederieke Kroon, Dean Jones

Research output: Contribution to journalArticle

8 Citations (Scopus)


Large scale networks of embedded wireless sensor nodes can passively capture sound for species detection. However, the acoustic recordings result in large amounts of data requiring in-network classification for such systems to be feasible. The current state of the art in the area of in-network bioacoustics classification targets narrowband or long-duration signals, which render it unsuitable for detecting species that emit impulsive broadband signals. In this study, impulsive broadband signals were classified using a small set of spectral and temporal features to aid in their automatic detection and classification. A prototype system is presented along with an experimental evaluation of automated classification methods. The sound used was recorded from a freshwater invasive fish in Australia, the spotted tilapia (Tilapia mariae). Results show a high degree of accuracy after evaluating the proposed detection and classification method for T. mariae sounds and comparing its performance against the state of the art. Moreover, performance slightly improves when the original signal was down-sampled from 44.1 to 16 kHz. This indicates that the proposed method is well-suited for detection and classification on embedded devices, which can be deployed to implement a large scale wireless sensor network for automated species detection.

Original languageEnglish
Pages (from-to)2502-2511
Number of pages10
JournalJournal of the Acoustical Society of America
Issue number5
Publication statusPublished - 1 May 2015
Externally publishedYes

Fingerprint Dive into the research topics of 'Automated detection of broadband clicks of freshwater fish using spectro-temporal features'. Together they form a unique fingerprint.

Cite this