Warm-water divers – spatial distribution of dive activity in chick-rearing Sn...
NSA-2015_BIO-3_JOIS
1. Study of Biological noise from Coral Reef
regions off Kavaratti Island of
Lakshadweep area employing passive
acoustic technique
Kashyap Jois*1,William Fernandes2, Bishwajit chakraborty2, K. Haris2,
A. K. Saran2, Kranthi Kumar Chanda2, M.M. Mahanty3 & G. Latha3
1Sardar Vallabhbhai National Institute of Technology
Surat, Gujarat, India
2CSIR- National Institute of Oceanography
Dona Paula, Goa: 403004, INDIA.
3National Institute of Ocean Technology
Chennai: 600 100, INDIA.
BIO-3
2. • Present study area is a coral reef located near
Kavaratti island.
• The fish that were spotted during data acquisition
were,
o Indo-Pacific Sergeant (Abudefduf vaigiensis),
o Black Triggerfish (Melichthys niger) and
o Sapphire Damselfish (Chrysiptera springeri).
• Data was collected by a single hydrophone
immersed in water with a sampling frequency of
40 kHz.
Introduction
3.
4. Challenges
• Diverse nature of marine life in coral reefs.
• Various disturbances in data caused due to natural
and anthropogenic sounds .
• Accurate segmentation of data into individual sound
events.
• Identification of species by analyzing the sound
events.
•hallenges
5.
6.
7. • Coral reef have many fishes whose expected
time duration of sound event varies with species.
• Low threshold level leads to many false
detections.
• High threshold levels fails to detect low amplitude
sounds.
Constraints in Segmentation
8. Coarse segmentation
• Band pass filtered data from
40 Hz - 18 kHz .
• Coarse segmentation with
the help of cumulative sum.
• Sharp rise in the slope
indicates start time.
• Saturation of curve
indicates stop time.
9. Fine segmentation
• Signal to noise ratio (SNR) using
Taeger Kaiser parameter
calculated for each coarse
segment (blue line) .
• TK parameter is a good estimate
of sound energy
• TK parameter increases with the
increase in amplitude of fish
sound.
• Gives better estimation of
beginning of sound event.
10. Threshold
• A threshold is used to distinguishes between fish sound
and noise.
• First threshold crossing of SNR within a particular
coarse segment corresponds to start of sound event.
• Last threshold crossing of SNR within a particular
coarse segment corresponds to end of sound event.
• Various threshold values were applied on each
segment and the output waveform (i.e. with the new
calculated start and stop times) are visually inspected.
• If waveform of sound event is clipped threshold is
reduced. A single appropriate threshold is selected.
12. Power spectral density plotted
for the 4 different types of
fish. Filter cutoffs selected by
observing PSD plots for
different cutoffs keeping in
mind the expected peak
frequency.
13. Clustering
• Next we verify our species classification by means
of a clustering algorithm
• Four features are computed for each sound event.
o Zero crossing rate
o Spectral centroid
o Peak frequency
o Hurst exponent
• These features were then clustered using FCM.
Each symbol depicts a separate cluster generated
by FCM.
14.
15.
16.
17. Conclusion
• Passive acoustics can be used to locate and identify
fish.
• The Segmentation method used has succeeded in
identifying some of the fishes presented using single
hydrophone sound record.
• Further scope for improvement would be automation of
this method using newer and more powerful machine
learning techniques such as Deep Learning.
18. CSIR – National Institute of Oceanography
Dona Paula,
Goa 403004. INDIA