Improving Acoustic Bats Survey Accuracy Using
Wavelet Analysis
An interdisciplinary endeavor
This web page records the ongoing efforts of an interdisciplinary team
of Truman State University faculty and students to improve the
accuracy of current acoustic survey methods using open source software
and cutting edge mathematics.
This is an organizational space for the group of people currently
working on the project. As results become available, they will be
linked through this page as well as other.
Description
Conservation biologists have begun to use acoustic techniques to
survey populations of bats. One reason for using acoustic techniques
to survey is that such surveys do not require a field biologist to
catpure and handle a bat as the only means of acquiring data. Reasons
for surveying bat populations include the need to determine whether or
not threatened or endangered species of bats are present in a survey
area. For this and other reasons, it is important for field
biologists to have techniques that allow them to identify a bat
species using only acoustic information.
A common tool for acquiring such data is the inexpensive Anabat
system by Titley Electronics (Australia). Our project aims to develop
numerical techniques for taking data acquired by the Anabat system and
identifying the species of those bats so recorded. We will build on
and try to improve the current techniques employed by the research
teams of Dr. Lynn
Robbins (Southwest Missouri State University) and Dr. Eric Britzke
(Clemson University).
The faculty and students of Truman State University that have
contributed to this work are:
Faculty
Undergraduates
Dr. Matt Beaky (Physics),
Dr. Scott Burt (Biology) and,
Dr. Jason Miller (Mathematics)
Gregory Knese (Mathematics, 01-02; IR),
Chris Bay (Mathematics, 02-present; SRS 2002, SH),
Lindsay Palmer (Mathematics, 03-present; SH),
Kathleen Field (Mathematics, 03-present; SH, Cap),
Clarke Cooper (Computer Science, 04-present; SH) and,
Raymond Feilner (Mathematics, 04-present; SH)
Dr. Dean DeCock (Statistics) deserves special thanks for assisting
with Katie Gustafson's work on discriminant function analysis.
Abbreviations: 'SRS' means research supported by a Truman Summer
Undergraduate Research Stipend, 'SH' means worked for scholarship
hours; 'IR' means independent research (no pay); 'Cap' means used for
Capstone requirement.
In all the PARAMS-[species]-cut.xls files, insert a row between those chirps that end a call sequence and begin a call sequence. This will indicate which calls in a seqnece can be used to create a three chirp datum. Also add to these xls files a 'time between chirp' calculation cell to each row that gives the time between the first & second, and the second & third chirps in a three-chirp datum. (Katie, in progress) (We'll need to compare these excel files with visualizations of the chirps to see that the inferred breaks occur at the right places.)
Cut and paste data in xls files created by Katie to create three-chirp rows. (Lindsey)
Modify the call analysis matlab m-files so that they identify a chirp and uniformly resample it from its starting point (time, frequency) until its end. Use this modified m-file to calculate the parameters used in the Britzke DFA model that we are recreating. (Clarke Cooper, in progress)
Get the list of files that Katie uses in her DFA and
return to her the full set of ten paramters from cleaned data
sets using filter-0, filter-7, and manual cutting. Give her
that data in Exccel format so she can easily get it into
SPSS. (Done by Jason)
Clean anabat generated data files so we can be sure that
the results of some of our algorithms are the result of our
algorithms and not funky data. (in
progress)
Incorporate 'inter-chirp' delay data into Katie's DFA.
(in progress)
Acquire other parameters used by Britzke in his DFA model
so we can better compare our work with his. (Done by Jason, see Data link below.)
See if R can do what SPSS is doing. It's always better to
be working with open source solutions than proprietary
solutions.
Begin acquiring 'full' datasets from Anabat systems using
Beaky's method fo bypassing the Anabat's ZCA module.
Ask the makers of anabat if they would be willing to let
some students develop Anabat software for Linux and make it
available under the GPL license. It would be good to ask
Chris Corben about thie, too. The economic benefits to Titley
Electronics and conservation biologists are clear (the
software would be more accessible). Clarke Cooper expressed
interest in this.
With pre-ZCA anabat dataset that contains complete
acoustic information, begin to investigate the possibility of
approaching the species characterization from a scale-space
theoretic direction using the deep structure of those data
sets.
Katie: Add 'time between chirps' into latest DFA model and
gauge improvement in differentiation. (in
progress)
Katie: document SPSS work so that a student can some and
pick up where you leave off after you graduate in May
Katie: write a short paper (in LaTeX) that summarizes the
work you did this year, including results and non-results.
You can expand on the short paper you wrote at the end of the
Fall semester.
Chris: continue working on curvature code; if we get
reliable data from the code we can try to add that parameters
into the DFA and measure increase in accuracy (if any) (in progress)
Chris: continue work on using matlab to visualize call
sequences from anabat. (in
progress)
Chris: write a short paper (in LaTeX) that summarizes the
work you did this year, including results and non-results
Raymond: understand the algorithms linguists have used to
compare words; explain how the algorithms work in writing with
a view toward applying it to 'pre-ZCA' data from anabat system
(which we will later acquire with Matt Beaky's help). (in progress)
Clarke: write Matlab/Octave code to extract other anabat
chipr parameters that are in Britzke's DFA. This needs to be
done ASAP so katie can incorporate this into her DFA model and
we can gauge the improvements or non-improvements that
result. (in progress)
Clarke: repeat Greg Knese's work with a Daubechies-2 based
analysis (this time using Matlab's wavelet toolbox) of anabat
data and write Matlab/Octave code that will automatically
count the numbers of different types of 'spikes' that occur in
the trend data. (Summer project)
Here is the [Data] that you
will need to convert. Remember that the naming convention that we
will use is this: file ends in # if it is an Analook fil, file ends in
X if it is an export from Analook, file ends in Y if it is converted
to a list of ordered pairs, and file ends in Z if it has
zero-frequency data points inserted into the dataset. The format we
choose for our 'Y' files should be the format that is easiest to
import into Mathematica and (at the same time, if possible) Matlab.
It's possible that we create a different set of files for Matlab.
(See the documentation for Mathematica's Export command to see what I
mean.)
http://users.lanminds.com/corben/hrmncs.htm#Harmonics:
Here is information on the harmonics in bat echolocation calls.
Contact this guy to get references concerning the use of fourier
analysis in bat identification.
When I find some helpful web pages for using
mathematica, I will list them here. Personally, I find
Mathematica's Help Browser sufficient for everything I do.
If you find good pages, tell me, and I'll list them here.