Zooplankton classification paper in JASA this month
Our second paper on using multi-view scattered sound to classify zooplankton was published in JASA this month. This paper details the developed of a laboratory scattering system for collecting multi-view data from live, untethered zooplankton, along with some processing methods we developed to highlight differences between two taxa: mysids and copepods. The paper can be found here.
Mon, 11 Aug 2008
Wed, 09 Jul 2008
Return from Acoustics 08
On Monday I returned from the Acoustics 08 meeting. In general the meeting was great. I attended several session each day, and found the content and the presentation of most talks to be superb. I presented my work in the Tuesday poster session which lasted for just under three hours and included a lot of fruitful discussion. This was my first poster session, and I was delighted with the turnout and interest others had in my work. The poster sessions were intense and productive, which could be a result of the fact that no talks were scheduled during the poster sessions, and they took place when many people were still at the meeting.
On Monday I returned from the Acoustics 08 meeting. In general the meeting was great. I attended several session each day, and found the content and the presentation of most talks to be superb. I presented my work in the Tuesday poster session which lasted for just under three hours and included a lot of fruitful discussion. This was my first poster session, and I was delighted with the turnout and interest others had in my work. The poster sessions were intense and productive, which could be a result of the fact that no talks were scheduled during the poster sessions, and they took place when many people were still at the meeting.
Wed, 25 Jun 2008
Heading to Paris for Acoustics 08
This Friday I will be flying to Paris, France to participate in the Acoustics08 conference. This is a large meeting that joins several American and European acoustics societies together.
I will be presenting a poster on some work I have done recently on understanding how bandwidth, aperture, and spatial sampling frequency affect classification accuracy in the multi-view approach. Not so surprising is the result that having more bandwidth and more views improves accuracy. More surprising perhaps is that the extent to which this is true varies significantly with the type of classification problem (eg size classification vs. shape classification).
This Friday I will be flying to Paris, France to participate in the Acoustics08 conference. This is a large meeting that joins several American and European acoustics societies together.
I will be presenting a poster on some work I have done recently on understanding how bandwidth, aperture, and spatial sampling frequency affect classification accuracy in the multi-view approach. Not so surprising is the result that having more bandwidth and more views improves accuracy. More surprising perhaps is that the extent to which this is true varies significantly with the type of classification problem (eg size classification vs. shape classification).
Wed, 04 Jun 2008
Some notes on my experimental setup
I have spent some time today checking over the equipment setup. Below, I have listed some of the settings for various parameters:
I have also spent quite a bit of time pulling strands of silk from a 100um braid. It seems that the silk can only be pulled to around 20cm before it will break, or come free from the braid. Using ASAIK a single strand yields a signal that is around 1 dB above the noise level.
I have spent some time today checking over the equipment setup. Below, I have listed some of the settings for various parameters:
- Peak voltage on 1.5" ducer during transmit: 250V
- Peak Voltage on 0.75" ducer during transmit: ~ (1/0.65)*250V
- Function generator equivalent voltage: 0.5Vpp
- Bias resistors for DAQ: 100kOhm
I have also spent quite a bit of time pulling strands of silk from a 100um braid. It seems that the silk can only be pulled to around 20cm before it will break, or come free from the braid. Using ASAIK a single strand yields a signal that is around 1 dB above the noise level.
Posted at: 08:24, Jun/04/2008
[/thesis/experiments]
permanent link
Plankton Pump
During the weekend, I tried a total different method for getting the plankton for my experiments into the FOV of the imaging system. The idea was to pump them into the volume using hydrostatic pressure.
I had a number of issues with bubbles, and the flow velocity through the volume being too high. None the less, I was able to get some reasonable data using this method. More then I had aquired with any of the other methods. So, I will be investiating this more during the next few days.
On Sunday, I aquired a data set with some adult calanus, various decopods, and some smaller copepods. The settings for the daya set were identivcal to those for the previous ones. Most of the animals were dead when they passed through the FOV, however a few were alive. I noticed interesting events between 7:45pm and 8:00pm. A lot of the data is just noise from air bubbles and water movement.
On Wednesday, I ran two more experiments, the first is with random samples from a net tow. Again, most of this data is just noise, yet there are a number of interesting events that warrent further analysis.
During the weekend, I tried a total different method for getting the plankton for my experiments into the FOV of the imaging system. The idea was to pump them into the volume using hydrostatic pressure.
I had a number of issues with bubbles, and the flow velocity through the volume being too high. None the less, I was able to get some reasonable data using this method. More then I had aquired with any of the other methods. So, I will be investiating this more during the next few days.
On Sunday, I aquired a data set with some adult calanus, various decopods, and some smaller copepods. The settings for the daya set were identivcal to those for the previous ones. Most of the animals were dead when they passed through the FOV, however a few were alive. I noticed interesting events between 7:45pm and 8:00pm. A lot of the data is just noise from air bubbles and water movement.
On Wednesday, I ran two more experiments, the first is with random samples from a net tow. Again, most of this data is just noise, yet there are a number of interesting events that warrent further analysis.
Posted at: 08:24, Jun/04/2008
[/thesis/experiments]
permanent link
Data from Friday's net tow
During the last two days, I have been working late nights trying to get good scattering data from a huge group of copepods that we collected on Friday evening. After the tows on Friday, I had a whole bunch of electrical problems that crept up before I was able to collect any data. The problems seem to be related to the power amp that I have been using becuase turning the input voltage down from 0.5Vpp to 0.35Vpp helped a lot. The only thing that I can think of is that the instantaneous power output was not performing properly, and thus turning down the power help out a lot.
I tried something new for pumping the critters into the FOV. Rather than pump from above which I had been doing, I tried pumping them in from below by reversing the direction of the plankton collector that I made. This worked better than the previous method, however I was still not satisfied. I collected quite a bit of data again, yet only a small fraction of it will be useable. All of the settings were kept the same as the previous experiments.
Tonight, I spent a really long time trying to figure out a way to pump the critters exactly into the very narrow acoustic field. I was trying to setup a low flow narrow output pump, yet every configuration that I tried resulted in a very large background signal. It may be that this was caused by turbulence, however it could also be a problem with bubbles. At any rate, there did not seem to be a good way to get the guys into the field of view without disrupting the acoustic field too much. In time, this might be an option, however for now, it's back to the bottom up method. However, this time, I have been using a siphon from a large carboy of critters so that the process is completely automated, and the encounter rate is much more constant. I also upped the input power to the amp to 0.4 Vpp.
During the last two days, I have been working late nights trying to get good scattering data from a huge group of copepods that we collected on Friday evening. After the tows on Friday, I had a whole bunch of electrical problems that crept up before I was able to collect any data. The problems seem to be related to the power amp that I have been using becuase turning the input voltage down from 0.5Vpp to 0.35Vpp helped a lot. The only thing that I can think of is that the instantaneous power output was not performing properly, and thus turning down the power help out a lot.
I tried something new for pumping the critters into the FOV. Rather than pump from above which I had been doing, I tried pumping them in from below by reversing the direction of the plankton collector that I made. This worked better than the previous method, however I was still not satisfied. I collected quite a bit of data again, yet only a small fraction of it will be useable. All of the settings were kept the same as the previous experiments.
Tonight, I spent a really long time trying to figure out a way to pump the critters exactly into the very narrow acoustic field. I was trying to setup a low flow narrow output pump, yet every configuration that I tried resulted in a very large background signal. It may be that this was caused by turbulence, however it could also be a problem with bubbles. At any rate, there did not seem to be a good way to get the guys into the field of view without disrupting the acoustic field too much. In time, this might be an option, however for now, it's back to the bottom up method. However, this time, I have been using a siphon from a large carboy of critters so that the process is completely automated, and the encounter rate is much more constant. I also upped the input power to the amp to 0.4 Vpp.
Posted at: 08:24, Jun/04/2008
[/thesis/experiments]
permanent link
Preliminary data processing
During the last week, I have been working on the first stage of data processing from the experimental work that I did last year. The first step of the process involved building a software tool to explore the data ping by ping in search of good data sets.
The tool is named MultiScat and I have finished the first alpha version. I have started a new page for documenting the software here.
One of the major steps in the data processing is dealing with calibration data. Using the MultiScat software made this step quite a bit easier.
During the last week, I have been working on the first stage of data processing from the experimental work that I did last year. The first step of the process involved building a software tool to explore the data ping by ping in search of good data sets.
The tool is named MultiScat and I have finished the first alpha version. I have started a new page for documenting the software here.
One of the major steps in the data processing is dealing with calibration data. Using the MultiScat software made this step quite a bit easier.
Posted at: 08:24, Jun/04/2008
[/thesis/research]
permanent link
Information about the Optical Setup
The system uses two Sony CCTV cameras that happen to have very high sensitivity in the Red through Near-IR. The model number is SPT-M320. For the zooplankton, a 50mm Nikon lens with f/11 aperature was used for the top view camera, and a 77mm Tamron zoom lens with f/11 aperature was used for the side view camera.
For the fish so far, a Fujion 25mm lens with f/5.6 aperature was used for the top camera, and the 50mm Nikon was used for the bottom camera again with f/5.6 aperature.
The physical distance from the top lens to the center of the FOV is around 95cm, and the physical distance from the side view camera to the center of the FOV was around 145cm.
The system uses two Sony CCTV cameras that happen to have very high sensitivity in the Red through Near-IR. The model number is SPT-M320. For the zooplankton, a 50mm Nikon lens with f/11 aperature was used for the top view camera, and a 77mm Tamron zoom lens with f/11 aperature was used for the side view camera.
For the fish so far, a Fujion 25mm lens with f/5.6 aperature was used for the top camera, and the 50mm Nikon was used for the bottom camera again with f/5.6 aperature.
The physical distance from the top lens to the center of the FOV is around 95cm, and the physical distance from the side view camera to the center of the FOV was around 145cm.
Posted at: 08:24, Jun/04/2008
[/thesis/experiments/optics]
permanent link
Crustacean Zooplankton and Me
I ran another set of experiments today. The input power to the amplifier was increased to 0.5 Vpp. This time, I was careful to use water that had been pumped directly from the FOV as the transport medium from the critters. This will ensure that I am not getting a spurious signal from the temperature differential but only from the scatterer. I got some nice data tonight. Theree was an exceptionally good set around 9:28pm where a single large copepod was wandering through the field of view for several ping series in a row.
Most importantly, I move the input source of water to the tank to be better alinged with the FOV of the system. I think that this made a significant difference in the number of animals that were coming through the FOV. The way I did this was to place a shinny object in the center of the input pump. I then moved it around unit it was centered in the FOV of one of the cameras. Once the flow was running, I then moved the input pump around until the flow seemed to be centered on the FOV of the other camera.
Tonight, I also tried to tether some fish with no success. This will require some thinking prior to the fish experiments next month. The main problem is simply that the fish can wriggle free from loops around thier body as well as tethers that are glued on. The trick might be to sew the tether into the body of the fish while it is knocked out. At this point, I have do clue as to how to go about putting a fish to sleep for surgury.
I ran another set of experiments today. The input power to the amplifier was increased to 0.5 Vpp. This time, I was careful to use water that had been pumped directly from the FOV as the transport medium from the critters. This will ensure that I am not getting a spurious signal from the temperature differential but only from the scatterer. I got some nice data tonight. Theree was an exceptionally good set around 9:28pm where a single large copepod was wandering through the field of view for several ping series in a row.
Most importantly, I move the input source of water to the tank to be better alinged with the FOV of the system. I think that this made a significant difference in the number of animals that were coming through the FOV. The way I did this was to place a shinny object in the center of the input pump. I then moved it around unit it was centered in the FOV of one of the cameras. Once the flow was running, I then moved the input pump around until the flow seemed to be centered on the FOV of the other camera.
Tonight, I also tried to tether some fish with no success. This will require some thinking prior to the fish experiments next month. The main problem is simply that the fish can wriggle free from loops around thier body as well as tethers that are glued on. The trick might be to sew the tether into the body of the fish while it is knocked out. At this point, I have do clue as to how to go about putting a fish to sleep for surgury.
Posted at: 08:24, Jun/04/2008
[/thesis/experiments]
permanent link
A new transducer and a smaller thread
Today I installed a new transducer in the experimental setup. The ducer was 2 times larger in diameter than the previous one that I had been using for the sound source. This increase in diameter should give an increase of a factor of 4 in pressure as the beam width is halved. This results in an overall increase of 16 times in intensity. Hopefully, this will be enough to give me a reasonable SNR for the smaller scatterers.
The ducer was too large to fit in the custom mounts that I made for the array. To fix this, I wrapped a 3/4" piece of pipe around the UHF connector for the ducer. I wrapper the UHF connector in tape until the 3/4" pipe snuggly fit over the connector. I then coated the end of the pipe in 5 minute epoxy and slid it onto the connector. This whole assembly then slid into the existing mount. To make sure things stay snug, I put another coat of epoxy on the mount, binding it to the back of the ducer.
Once I had aligned the ducer, I did a comparison test between the previous ducer and the new one. The target I used was a vertical line of unknown reflectivity. In theory, I should see a factor of 2 increase in received voltage. This is basically what I saw, so at least the new ducer was aligned as well as the previous one was, which was very well AFAIK. The factor of 2 rather than 4 comes from the fact that the vertical line is much longer than the beam width, so in the vertical dimension, there is no increase in energy. For both beams, the line intercepts all of the acoustic energy. In the horizontal, this is not the case, as the line is around 100um thick. So the reduced beam does deposit more energy in the horizontal axis. If I had a target that was confined in both dimensions, then the increase should be by a factor of 4.
While tested the new ducer, it suddenly dawned on me to take the braided silk that I was using for a target, and extract one of the braids to use as a tether. One of the braids is easily less than 50% of the diameter of total braid, probably more like 20-40%. The full braid has a diameter of roughly 100um. This puts the individual braid around 30um. Since these lengths are around 10 times smaller than the wavelength of sound, the scattering intensity scales with length to the fourth power. Therefore, reducing the length by a factor of 3, would in theory reduce the scattering intensity by a factor of 81, almost 20 dB! I tested this idea with what I thought was a single strand. The reduction was substantial, but not enough to put the reflection below the noise floor. However, I found out afterwards that I could actually pull out even smaller stands, so this may in fact do the trick. I should also mention that I tried a similar idea out with some 75um monofilament, where I tried to stretch it to reduce the thickness. This did not work at all, even with heating or cooling the monofilament.
Today I installed a new transducer in the experimental setup. The ducer was 2 times larger in diameter than the previous one that I had been using for the sound source. This increase in diameter should give an increase of a factor of 4 in pressure as the beam width is halved. This results in an overall increase of 16 times in intensity. Hopefully, this will be enough to give me a reasonable SNR for the smaller scatterers.
The ducer was too large to fit in the custom mounts that I made for the array. To fix this, I wrapped a 3/4" piece of pipe around the UHF connector for the ducer. I wrapper the UHF connector in tape until the 3/4" pipe snuggly fit over the connector. I then coated the end of the pipe in 5 minute epoxy and slid it onto the connector. This whole assembly then slid into the existing mount. To make sure things stay snug, I put another coat of epoxy on the mount, binding it to the back of the ducer.
Once I had aligned the ducer, I did a comparison test between the previous ducer and the new one. The target I used was a vertical line of unknown reflectivity. In theory, I should see a factor of 2 increase in received voltage. This is basically what I saw, so at least the new ducer was aligned as well as the previous one was, which was very well AFAIK. The factor of 2 rather than 4 comes from the fact that the vertical line is much longer than the beam width, so in the vertical dimension, there is no increase in energy. For both beams, the line intercepts all of the acoustic energy. In the horizontal, this is not the case, as the line is around 100um thick. So the reduced beam does deposit more energy in the horizontal axis. If I had a target that was confined in both dimensions, then the increase should be by a factor of 4.
While tested the new ducer, it suddenly dawned on me to take the braided silk that I was using for a target, and extract one of the braids to use as a tether. One of the braids is easily less than 50% of the diameter of total braid, probably more like 20-40%. The full braid has a diameter of roughly 100um. This puts the individual braid around 30um. Since these lengths are around 10 times smaller than the wavelength of sound, the scattering intensity scales with length to the fourth power. Therefore, reducing the length by a factor of 3, would in theory reduce the scattering intensity by a factor of 81, almost 20 dB! I tested this idea with what I thought was a single strand. The reduction was substantial, but not enough to put the reflection below the noise floor. However, I found out afterwards that I could actually pull out even smaller stands, so this may in fact do the trick. I should also mention that I tried a similar idea out with some 75um monofilament, where I tried to stretch it to reduce the thickness. This did not work at all, even with heating or cooling the monofilament.
Posted at: 08:24, Jun/04/2008
[/thesis/experiments]
permanent link
