Groundtruth

How to make a groundtruth

For the Bm-Bcall-Detector

Description

Before you can run the automated detector on a deployment, you need a groundtruth. The groundtruth is a section of a dataset from which B calls have been manually picked and logged. By comparing the calls you find to the calls the detector finds, we can ensure that the detector is performing optimally on individual datasets.

The month in which to make a groundtruth, as well as the number of calls the groundtruth should contain, can vary based on the region your data is from.

For data in SOCAL, groundtruths are generally made for the following months:

• October (400 calls)
• December (200 calls)
• June (200 calls)

💡 Tip: October represents the peak calling season and so we log more calls in order to capture a better sample of the data. June and December represent the start and end of the calling season, respectively. However, if you needed to make a groundtruth in a different month you could still do that (e.g., a November groundtruth should have around 300 calls).

You should make one groundtruth for every deployment you want to detect B calls in. For example, if a deployment recorded from August to January, the groundtruth for that deployment should be made in October.

Requirements

• MATLAB 2013b (or later)
• Triton

Make a groundtruth

1. Open MATLAB and start Triton

2. Open the LTSA for the dataset you want to make a groundtruth for

3. Scan through the data during the appropriate month for a section of calls to log

   • Optimal window settings for the LTSA: 1 hour plot length, 0-300 Hz, Brightness 20 dB, Contrast 250%
   • Optimal window settings for the spectrogram: 60 s plot length, 0-200 Hz, FFT 2000, Overlap 90%, Brightness 20 dB, Contrast 250%
   • Look for a sequence of strong, continuous calls
   • 💡 Tip: If you log a sequence of calls contained within one xwav file, things will be easier later on

4. Once you have located a sequence of calls to log, begin logging the start times of each indididual call

5. Continue logging call start times until you have logged the appropriate number of calls for the month

6. Close the log in Triton

7. Open the xls log you just created

8. Make a note of the time of the first and last calls you logged and save this information somewhere

9. Convert the column containing the start time of your calls into number format

   

10. Save your log

Next steps

If you've already generated a kernel, you are now ready to determine the threshold