Mapping Waste Piles and Plastic Pollution using UAV Imagery and Machine Learning
This is a modal window.
The media could not be loaded, either because the server or network failed or because the format is not supported.
Formal Metadata
| Title |
| |
| Title of Series | ||
| Number of Parts | 41 | |
| Author | ||
| License | CC Attribution 3.0 Unported: You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal purpose as long as the work is attributed to the author in the manner specified by the author or licensor. | |
| Identifiers | 10.5446/58218 (DOI) | |
| Publisher | ||
| Release Date | ||
| Language |
Content Metadata
| Subject Area | ||
| Genre | ||
| Abstract |
|
State of the Map US 202220 / 41
1
4
7
8
14
18
21
22
24
27
28
30
37
39
00:00
CAN busLevel (video gaming)MappingMachine learningOpen setInternetworkingMultiplication signBitForestGradient descentSoftware testingCross-validation (statistics)SmoothingComputer filePoint (geometry)IdentifiabilityType theoryMessage passingNumberAlgorithmDependent and independent variablesTerm (mathematics)Group actionSource codeComputer programmingNeuroinformatikEndliche ModelltheorieMereologyExterior algebraLink (knot theory)BuildingWave packetStructural loadProcess (computing)Uniform resource locatorPoint cloudFrequencyTouchscreenHypermediaArithmetic meanMeasurementIntegrated development environmentExistencePhysical systemDampingSurfaceVirtual machineData managementDifferent (Kate Ryan album)Neighbourhood (graph theory)Medical imagingUniqueness quantificationStrategy gameScaling (geometry)Wage labourInformationSampling (statistics)PixelComputer animation
02:18
Uniform resource locatorLevel (video gaming)Computer animation
02:32
Integrated development environmentVideo trackingObservational studyMatrix (mathematics)Uniqueness quantificationWage labourData managementInformationoutputPattern languageTerm (mathematics)Local ringGroup actionResultantComputer programmingError messageSource codeFunction (mathematics)MereologyMachine learningPoint (geometry)Different (Kate Ryan album)SurfaceExistenceHypermediaMultiplication signSingle-precision floating-point formatMappingArithmetic meanPoint cloudProcess (computing)Wave packetPhysical systemMeasurementIntegrated development environmentUniform resource locatorStrategy gameLevel (video gaming)Virtual machineData managementMedical imagingNeighbourhood (graph theory)Endliche ModelltheorieInformationSampling (statistics)Wage labourForestPixelComputer animation
Transcript: English(auto-generated)
00:04
Thank you very much. My name is Patrick Amunde. I'm here representing about mapping of waste piles and plastic pollution using UAV and machine learning. So a little bit about me. I come from the Republic of Marau. This is one of the most beautiful countries in
00:24
South East Africa, as you can see on the map. And a few years ago, particularly in 2017, I was at the U.S. Embassy in Malawi. And there was this guy who was talking about maps. So he talked about someone by the name of Dr. John Snow. And he talked about a
00:49
time in the 1800s when London was affected by an outbreak of cholera. And his talk took my heart and said, ah, this is amazing. But what really took my attention was how
01:06
this guy used maps to solve the problem of cholera. So my talk is about the same subject. So after hearing about the story of Dr. John Snow and looking back to the
01:20
community that I come from, as you can see on the map, you can see buildings, loads, rivers. But it's also a community that has problems which are not there on the map. For example, if you go along the streets, you see that there are piles of waste actually in the rivers. You see plastics and so many other things. So we,
01:50
myself and my friends, we say, look, can we maybe just map these waste piles? So we're on the internet to look for previous work that we've done by other
02:02
people. And we found an article that was published by Mr. Sean Rink, which is showing open data on mapping plastic pollution. And with that, we were fascinated. We said, let's do this in this community. And we mobilized the few friends going around collecting coordinates for locations where waste is
02:26
being disposed. And with that, we were able to generate a map of our community with waste piles, as you can see there. So now, as I said previously,
02:40
every problem can be mapped. So the problem of waste, we were able to map it. But then, at that time, people say maps are not the real solution that we want. Maps are just a means to the end. It's not the end. So they say, let's do a cleanup campaign. So we mobilized resources, volunteers, and we
03:03
did a cleanup exercise in the community. So we were excited. At that time, there was a movement called hashtag trash challenge. So we cleaned one of the places, and we sent it on social media. People were, oh, this is nice. But after a few days, people wanted to dispose of waste at the
03:25
same location. We were disappointed. Our hearts were broken. And it happened. In 2020, the president of the Republic of Malawi, Dr. Aziz Jafran, announced that every single flight in Malawi would be waste
03:43
cleanup campaigns. So looking at my experience doing cleanup campaigns, I was like, let me write in the national newspaper. So I sent an article to the newspapers in Malawi, that you can see on the screen, saying that cleanups suggest scratching the surface. So this is one of the
04:07
activities that my friends mobilized other volunteers to support the national cleanup campaign, even though I published the paper. So what we observed was that the problem that we were trying to solve, it's
04:25
not just a global problem for my community. This is a global problem. In fact, particularly for plastics, their leakage into the environment is actually linked to the existence of systems for management of waste. And the various international guidelines, for example, the United
04:43
Nations Environmental Program, they have published guidelines that emphasize on the need for monitoring waste and plastics in the environment, because they say if you can't measure it, you cannot know that you're making a difference. And looking at the way we
05:05
did our exercise of collecting coordinates of locations where waste patterns are in our community, the process that we did was labor-intensive and also difficult rock scale, because we only collected coordinates in locations that were accessible to us. As we saw
05:22
previously, most of the waste is actually in the river. So working in the river to collect coordinates, that's something that we did not do. So for example here, the whole river is full of waste. So we realized that drone technology is a promising technology. And if we bring in drone
05:43
technology, this can actually strengthen the efforts of monitoring the impacts of the policies and strategies that we put in place, particularly to combat the problem of waste management. So this was a quick question. How best can we integrate the UAV technology,
06:04
machine learning, particularly for locating these waste patterns in the community? So we had to, because they say the only way you can see better is by standing on the shoulders of giants. So we had to look at what others have done. And we found that there
06:22
were some papers published in peer reviewed journals, particularly in Portugal and other countries. So we developed our methods based on what others have done. So we started with like a small experiment. We had some plastics on the ground
06:43
and we were flying a drone at different heights. So for example, you can see like at the top there, we flew our drone at 50 feet, like we could see our plastic. But when we increased the height, the plastic was not even visible. And we saw that
07:04
because flying a drone in a location where there are buildings, it is not able to observe that there are plastics when you fly it lower. So it was a big challenge for us. So we collected some drone images
07:23
in a neighbourhood in Malawi, as you can see there, and clearly when you just zoom into the image, waste patterns are actually visible, as you can see there in the middle of the picture. So the question is, should we do this manually or should we automate the process?
07:43
So we employed a object-based image classification approach. So we had to group pixels which are seen here together into segments, as you can see there. And we had to create some training samples that can be used as examples to build our machine learning model. So we
08:03
are, based on the literature that people have done, one of the approaches machine learning and coding that was very successful was the use of land forests. So we had to create like a five-fold cross validation, which was repeated five times, and
08:23
we developed our model which performed far much more better. As you can see there, from the data which we had, we had to receive some critical testing, and out of the 20 points that we had for waste patterns, it was able to identify 19 waste
08:43
patterns, and it only missed one. Of course there was also some confusion where things which are not waste patterns were classified as waste patterns. And in terms of the activation, those are the numbers. The performance is just very good. So this is the output.
09:07
So you can see the same community. You have like, in late, the waste parts that are there in the river. So we also excited to see like, since like the map that I showed you previously, one cannot see plastics
09:23
because people are much more interested in mapping plastic pollution. So we try to experiment to see like if we map plastics, because it was difficult for us to find the error given the obstacle. So we say, okay, let's just map some plastics across the ground and see if we are able to collectively identify them using machine learning.
09:45
And yes, indeed, we were able to detect some of these plastics, as you can see in there, and this is the performance in source of all, but not good as the previous one.
10:00
So what's the take home message? The use of drone technology for monitoring the environment is very promising. It creates unique opportunities for the environment management, particularly given that we can be able to monitor the interventions that we put in place to see if they are really achieving the results that we want.
10:25
So for us, our next step is to move beyond waiting with our computers and maps, but rather to also integrate this information in existing urban environmental management programs. So we are considering engaging stakeholders, policymakers,
10:45
local communities, and beyond that, we also have a belief that as an individual you can go fast, but as a group you can go far. So we want to make such data to be open so that we should be able to
11:04
get solutions from different places. They say that great ideas return from unexpected sources. So we want to make this thing to be much more open as much as possible. And the last point is the process of developing training data was
11:24
labor intensive. It took a lot of time for us to develop this data. So we are exploring opportunities, for example, cloud sourcing, for example, the way that a lot of people come together to create a lot of data within a short period of time. So that's one of the things that we want to integrate
11:43
into the future. So these are the people who have been helping me along the way. Thank you very much.
Recommendations
Series of 2 media
Series of 2 media