Understanding sow’s mothering ability by analyzing their behavioral phenotypes from overhead sensor images
Studying sow’s behavior and understanding how to improve them will lead to a better mothering ability and low piglet crushing rates which is a fundamental solution to the preweaning mortality issue.
Toggle between list and paragraph view.
- [00:00:02.490]this is Ahlam Alkiyumi and I'm
a senior majoring biological systems engineering
- [00:00:07.110]and emphasis in biomedical engineering.
Today, I will go over my research,
- [00:00:11.880]which is about understanding sow's
mothering ability by analyzing their
- [00:00:16.140]behavioral phenotypes from
overhead sensor images.
- [00:00:22.250]United State is the world's third
largest pork-producing country.
- [00:00:26.480]And one of the largest countries
that export the pork and pre-weaning
- [00:00:31.190]piglet mortality is a major
economic and welfare issue in big
- [00:00:36.650]There is a study showed that
17.8% of the piglets born
- [00:00:41.510]are lost before preweaning and half of the
piglets are lost due to crushing by.
- [00:00:46.400]Sow it's a very high percentage.
- [00:00:50.720]This issue affects piglets well
being,worker morale ,and the economic
- [00:00:55.250]returns to the us.
- [00:00:57.620]So as a way to limit the piglets
crushing by sow, swine
- [00:01:02.240]industry to develop a farrowing crate.
- [00:01:05.780]So in figure one, so the,
- [00:01:10.490]sow just has that middle space to move,
- [00:01:16.070]but unfortunately,
- [00:01:17.990]the crushing still occurs. for
our objective in this research
- [00:01:23.090]is to identify a serious of sows
postures that are potentially related to
- [00:01:27.950]their mothering ability from the
overhead digital and depth images that were
- [00:01:32.930]collected in a swine barn
over the pre-weaning period.
- [00:01:38.330]And the second one is to analyze
the time spent in each posture and
- [00:01:43.310]find the difference between
high and low mortality.
- [00:01:50.060]This flow chart shows the
method we use in this research.
- [00:01:54.800]First thing,
- [00:01:55.790]the images were collected by
using cameras that takes digital
- [00:02:00.590]and depth images,
- [00:02:02.030]and the cameras were remotely
not attached to the sow's body.
- [00:02:06.560]The time interval between
images is five seconds.
- [00:02:10.550]The second step,
- [00:02:13.340]we develop a MATLAB GUI
to display the each and,
- [00:02:17.450]following pairs of the depth and RGB images
- [00:02:20.990]So in this figure one,
- [00:02:24.530]this is what we got after
we run the program,
- [00:02:28.880]the top two images are the current frame
- [00:02:33.590]and the bottom two images
actually are short running
- [00:02:38.480]of the five following frames
to help us classify the images.
- [00:02:43.400]And here you can see that
the posture and the current
- [00:02:48.200]frame, which is standing
- [00:02:52.370]laying on belly, laying on
right side or laying on the left side.
- [00:02:56.810]So we have to classify the
images and the posture behavior
- [00:03:01.570]in these six categories,
- [00:03:07.380]labeled images were
saved as an Excel file.
- [00:03:11.580]The next step is to analyze
the total time spent in
- [00:03:16.230]each posture for each sow
and MATLAB code was developed to
- [00:03:20.640]extract the labeled depth
images from the saved
- [00:03:25.410]Excel file and calculate the
time in each behavior for
- [00:03:30.150]each Sow over the three
days after piglets birth.
- [00:03:36.240]And the last step is
to find the difference
- [00:03:41.040]between low and high mortality.
- [00:03:43.020]Sow in their time budget for each
- [00:03:47.820]posture. we use Excel data
analysis tool to make an
- [00:03:52.500]unpaired t-test here's our result and
- [00:03:56.670]discussion. more than a hundred
thousand images were labeled
- [00:04:01.620]for five low piglet
pre-weaning mortality sows,
- [00:04:05.370]and five high piglet
pre-weaning mortality sows.
- [00:04:10.650]Uh, in this figure,
- [00:04:13.110]two shows the average time
spent in each posture for five
- [00:04:17.980]low mortality and five high mortality.
- [00:04:21.870]Sow within the first 24
hours after farrowing,
- [00:04:26.250]so the red is for the high
mortality and the blue for
- [00:04:31.200]the low mortality. from this figure,
- [00:04:36.180]we can highlight some
differences in their behavior.
- [00:04:40.470]For example,
- [00:04:41.910]high mortality spent
more time on a standing
- [00:04:46.860]sitting and laying on a belly,
- [00:04:52.740]but low mortality spent more time on
- [00:04:57.420]kneeling within the
first 24 hour after piglets
- [00:05:04.800]to see the difference between high
and low mortality in the first
- [00:05:09.570]24 hours in their behaviors,
- [00:05:12.180]we made an unpaired T-test
using 0.1 alpha,
- [00:05:16.890]and the test showed they are not
significantly different in their
- [00:05:21.480]behaviors within the first 24 hours.
- [00:05:24.990]The last two columns of this table
shows the standard deviation
- [00:05:29.850]and the p-value for the first day,
- [00:05:35.160]we see that p- values
- [00:05:40.050]are higher than alpha and
maybe this because we have a
- [00:05:44.880]very small sample size, just
five Sow in each category,
- [00:05:51.720]the rest ,of this table shows
the mean maximum ,and minimum
- [00:05:57.470]of six posture behavior Within
the three days after piglet birth,
- [00:06:02.390]between low mortality
and high mortality sows,
- [00:06:06.350]we analyzed from this table,
- [00:06:10.550]low mortality spent more time
kneeling than high mortality
- [00:06:15.350]in all three days,
- [00:06:18.350]the saw kneels when she
changed from standing to lying.
- [00:06:23.270]And it's known that most of the
crushing occurs during this
- [00:06:27.980]transition from standing to lying.
- [00:06:31.010]And this is a good sign of
a good mom when she spent
- [00:06:35.870]more time kneeling.
- [00:06:40.090]based on the 10 sows, we analyzed,
- [00:06:42.180]we concluded that there are some marked
differences between low mortality and
- [00:06:47.020]high mortality sows on kneeling
sitting and standing behavior.
- [00:06:53.650]We invistigated the
potential of using CNN based,
- [00:06:58.180]deep learning on
- [00:07:01.240]Sow's posture and the model showed 98.3
- [00:07:05.500]3% accuracy on labeling sows behavior.
- [00:07:09.400]Using RGB images.
- [00:07:12.250]This model will help the analysis
of the massive data collected in the
- [00:07:17.110]future for the sow's behavior analysis.
- [00:07:23.020]I ,
- [00:07:23.470]also realized I have to add
laying other posyure into the
- [00:07:28.270]six categories.
- [00:07:32.230]We will add more sows to the research
and I'm working on publishing a
- [00:07:36.880]journal article.
- [00:07:39.460]The last thing I want to thank
UCARE program for this opportunity and
- [00:07:44.470]thank your Adil and Veronica for
helping me. Thank you for listening.
Log in to post comments