PPT

forestevanescentElectronics - Devices

Nov 2, 2013 (3 years and 7 months ago)

59 views

Craig Holmes

Brad Klippstein

Andrew Pottkotter

Dustin Osborn


Wind Turbines have become a great
source of alternative “green” energy



Lake Erie and North West Ohio have the
potential to be a great source of wind
energy.



North West Ohio also happens to be one
of the largest migratory bird fly
-
ways in
the country



Before wind turbines can be installed
questions need to be answered about the
effect they will have on the avian
population in the area



“As wind energy facilities becomes substantially more
numerous and as wind development continues to
grow, fatalities and thus the potential for biologically
significant impacts to local populations increases”

National Wind Coordinating Collaborative


The BBCER strives to create a way to identify what
avian species are being impacted by the wind turbines
and to allow for conclusions to be made about what
effect they are having on the species.



Bird/Bat Calls Embedded Recognition System

Bird System

Bat System

Hardware

10 bit
ATMega

328 A/D
microcontroller

AR125 Ultrasonic
Receiver

Microphones

Stand

FR125
-
III Field Recorder

Four

UEM
-
88 Mini Shotgun Microphones



MAYA
-
44



PC



Software

Bird Call Library

SPECT'R Software

Raven
-
X

SCAN'R Software


Matlab

SonoBat Software

Incoming
Unknown Bird
Call

Take First 512
Data Points
from Call

Apply
Hamming
Window to
Data Points

Derive Fourier
Transform of
Data Points

Find the
Magnitude of
the Fourier
Transform

Multiply by the
Mel Frequency
Scale

Gather Mel
Frequency
Cepstral
Coefficients

Apply Discrete
Cosine
Transfrom to
MFCC's

Place First 9
Coefficients in
MFCC Bank

Move to Next
512 Data Points
from Call

Once through all 220,500 Data
Points of
Call,
Send MFCC Bank
to Correlation Algorithm for
Matching


“Scale of pitches judged
by listeners to be equal
in distance from one
another”[1]


Converts Hertz scale to
Mel scale.



Cepstral: Fourier Transform of the log spectrum. [1]


Mel Frequency Cepstrum (MFC): Representation of
the short term power spectrum of sound, based on a
linear cosine transform of a log power spectrum on a
nonlinear Mel scale of frequency.[1]


Mel Frequency Cepstral Coefficients (MFCC):
Coefficients that collectively make up an MFC[1]

Incoming
Unknown MFCC
Bank

Calculate
Pearson's
Correlation
Coefficient with
first database
call

Calculate Next
Pearson's
Correlation
Coefficient for next
known call

Repeat calculating
correlation
coefficient for
all
database calls

Output call with
highest correlation
as
best match

If all calculated
correlation coefficients
below threshold then
unknown call


Measures the correlation between two linear
dependent variables.






Signified by r (rho) and can take on values between
-
1
to 1. Where
-
1 signifies perfect negative correlation, 0 is
no correlation, and 1 is perfect positive correlation.


The previous example of American Crow and
American Coot yield an r value of 0.4650











With our system we are able to successfully identify an
‘unknown’ bird call that has been recorded through
the UEM
-
88 mini
-
shotgun microphones and
interfaced to the PC with the MAYA
-
44 USB device.


Correlation Coefficient of .6411


Together the AR125 and FR125
-
III record the ultrasonic
bat calls from the field.


They then work with the SPECT’R, SCAN’R, and
SonoBat software to determine the species of which
the call came from.


The system has been tested and has successfully
identified species of bats in the field.


There are three software programs that we purchased for
bat call analysis


SPECT’R


SCAN’R


SonoBat
.


SPECT’R is used to perform spectral analysis, digital
tuning, and hard
-
disk recording of echolocation calls.


SCAN’R is a snapshot characterization and analysis tool
which is used to distinguish between bat calls and
unwanted noise.


SonoBat

can be used to analyze and compare high
-
resolution full
-
spectrum sonograms of echolocation calls.


Because of the Microcontroller, that we purchased,
memory limitations , we are unable to successfully
record an audio call with enough data points to be ran
through our algorithm.


In the future we recommend purchasing a
microcontroller with at least 512kB SRAM to get a
correctly sampled call plus room to store the code.


We also recommend making the system wireless so
that the system could be placed in the field and left to
record bird calls overnight or for long durations.