1887
banner image
No data available.
Please log in to see this content.
You have no subscription access to this content.
No metrics data to plot.
The attempt to load metrics for this article has failed.
The attempt to plot a graph for these metrics has failed.
Automated segmentation of linear time-frequency representations of marine-mammal sounds
Rent:
Rent this article for
USD
10.1121/1.4816579
/content/asa/journal/jasa/134/3/10.1121/1.4816579
http://aip.metastore.ingenta.com/content/asa/journal/jasa/134/3/10.1121/1.4816579

Figures

Image of FIG. 1.
FIG. 1.

Block diagram of the time-frequency support detection process.

Image of FIG. 2.
FIG. 2.

Illustration of the block partitioning of the binary spectrogram.

Image of FIG. 3.
FIG. 3.

Segmentation of the spectrogram. (a) Spectrogram of a signal sample (Bay of Biscay), computed with a 512-point Hanning window, with 75% overlap. Several non-linear frequency modulations (dolphin calls) are overlapped and interlaced. (b) PSD of the noise estimated on a [10 s, × 500 Hz] time-frequency sliding neighborhood, observed CDF (blue line) and CDF ofthe fitted Chi-squared distribution (black dots). (c) Thresholded spectrogram, after step 1, with false-alarmprobability of  = 0.01. (d) Observed CDF of the number of detections in the sub-regions (black crosses), with fitted binomial distribution via Kullback–Leibler divergence (red diamonds) and fitted binomial distribution via mean (blue squares). (e) Segmented spectrogram, the binary spectrogram is divided into 7 × 7 time-frequency squares and the false-alarm probability is set to  = 0.01; components shorter than 100 ms arediscarded. Detected regions are marked in gray.

Image of FIG. 4.
FIG. 4.

Zoom on the results of the segmentation of the spectrogram of the signal for the three different methods. The noise PSD is estimated at each time-frequency bin on a [10 s × 500 Hz] sliding neighborhood. The threshold to generate the binary spectrogram is set to 10 dB. For each method, segmentation results shorter than 100 ms are discarded. (a) Original spectrogram ofa signal sample, computed with a 512-point Hanning window, with 75% overlap. Several whistles with different SNRs are seen. (b) Segmentation results for method 1 (with a 3 × 3 Gaussian window), where signals in boxes B1,B2, and B3 are missed. (c) Segmentation results for the morphological method, 3 × 3 bins structuring elements, where signals in boxes B2 and B3 are missed. (d) Segmentation results for the binomial fitting method, with the binary spectrogram sub-divided in 7 × 7 sub-matrices,  = 0.05. The signal in box B2 is missed.

Image of FIG. 5.
FIG. 5.

Zoom on the results of the segmentation of the spectrogram of the signal for the three different methods. The noise PSD is estimated at each time-frequency bin on a [10 s × 500 Hz] sliding neighborhood. The threshold togenerate the binary spectrogram is set to 3 dB. For each method, segmentation results shorter than 100 ms are discarded. (a) Original spectrogram ofa signal sample, computed with a 512-point Hanning window, with 75% overlap. Several whistles with different SNRs are seen. (b) Segmentation results for method 1 (with a 3 × 3 Gaussian window), all signals are detected. The results are corrupted by numerous false detections. (c) Segmentation results for the morphological method, 3 × 3 bins structuring element. The results are illegible, and false positives are merged together. (d)Segmentation results for the binomial fitting method, where the binary spectrogram is sub-divided in 7 × 7 sub-matrices,  = 0.05. Signals are accurately detected. There are few false detections.

Tables

Generic image for table
TABLE I.

List of the parameters used for segmentation in Figs. 4 and 5 .

Loading

Article metrics loading...

/content/asa/journal/jasa/134/3/10.1121/1.4816579
2013-09-01
2014-04-20
Loading

Full text loading...

This is a required field
Please enter a valid email address
752b84549af89a08dbdd7fdb8b9568b5 journal.articlezxybnytfddd
Scitation: Automated segmentation of linear time-frequency representations of marine-mammal sounds
http://aip.metastore.ingenta.com/content/asa/journal/jasa/134/3/10.1121/1.4816579
10.1121/1.4816579
SEARCH_EXPAND_ITEM