| Summary: | © 2018 IEEE. A spectrogram-based technique is presented for detection of time-and-bandwidth limited broadband signals in underwater audio. The approach employs an iterative 1-dimensional variant of a 2-dimensional multi-scale blob-detection technique commonly used in image processing. In contrast to the referenced 2-dimensional technique subject to an inherent bias for circular features, the iterative 1-dimensional approach enables detection of features of arbitrary bandwidth and duration. The iterative nature (of processing successive frames) makes it an attractive choice for in-situ streaming-mode applications. The algorithm automatically chooses values for several parameters based on the input spectrogram’s frequency bounds and hence is capable of being readily employed for a variety of applications. The technique’s applications include detection of broadband signals of interest, such as Omura’s whale (Balaenoptera omurai) calls, underwater earthquakes or explosions. With long-term spectral averages, the technique may be used in identifying long-lasting sounds contributing to ambient noise, such as fish choruses or the sounds of wind and rain. A theoretical analysis of the proposed technique’s performance is provided and its detection capability is demonstrated using representative examples.
|
|---|