An automated procedure has been developed for detecting and localizing frequency-modulated bowhead whale sounds in the presence of seismic airgun surveys. The procedure was applied to four years of data, collected from over 30 directional autonomous recording packages deployed over a 280 km span of continental shelf in the Alaskan Beaufort Sea. The procedure has six sequential stages that begin by extracting 25-element feature vectors from spectrograms of potential call candidates. Two cascaded neural networks then classify some feature vectors as bowhead calls, and the procedure then matches calls between recorders to triangulate locations. To train the networks, manual analysts flagged 219 471 bowhead call examples from 2008 and 2009. Manual analyses were also used to identify 1.17 million transient signals that were not whale calls. The network output thresholds were adjusted to reject 20% of whale calls in the training data. Validation runs using 2007 and 2010 data found that the procedure missed 30%–40% of manually detected calls. Furthermore, 20%–40% of the sounds flagged as calls are not present in the manual analyses; however, these extra detections incorporate legitimate whale calls overlooked by human analysts. Both manual and automated methods produce similar spatial and temporal call distributions.
This work has been supported by Shell Oil Exploration and Production Company (SEPCO). Kristin Otte and Sara Tennant, among many others, helped derive the manual training sets used in this effort. Bill McLennan developed the acoustic data format used by the automated processor, and devised a nonlinear calibration scheme used in the 2007 data set. Steve Eddins of Mathworks provided background on connected component labeling algorithms. Professor Garrison Cottrell at UCSD provided advice on the training and implementation of neural networks, and Dr. Sergio Belongie of UCSD provided advice on image processing methods. Melania Guerra and Delphine Mathias assisted with field deployments, and they, along with Diana Ponce, also assisted with data analysis and fact-checking. Julien Delarue of JASCO provided references on walrus sounds.
A. DASAR description
B. Deployment geometry and timelines
A. Bowhead whale call diversity and previous automation research
B. Seismic survey signals and other interfering transients
IV. AUTOMATED PROCEDURE
B. Event detection
C. Interval filtering
D. Image processing and feature extraction
1. Spectrogram equalization and double-thresholding
2. Morphological processing and connected component labeling
3. Binary and SNR-weighted ridge feature extraction
4. Ridge segment splicing and linking
5. Contour/ridge linking and feature vector assembly
E. Feature filtering: neural network processing
1. Creation of training data sets
2. Architecture and training protocol
F. Cross-DASAR call matching
G. Bearing estimation and localization
A. Neural network training datasets, bulk processing runs, and threshold test runs
B. Example of bulk processing using default network thresholds
C. Comparing manual and automated performance before localization stage
D. Example of whale call spatial distributions from manual and automated processing
A. Relative importance of various stages during bulk processing
B. Sensitivity of neural network performance to training set size and feature selection
C. Interpreting excess call fraction
Data & Media loading...
Article metrics loading...
Full text loading...