 I would try to be a bit soft and simple. So the overall idea is how to find the models in the processing to help environmental monitoring. And in this case, our application is according to ways and targeting in India. I'm going to use the answer. So what's the goal for this one to understand. To deflect the technology or the tools that save AI and computer vision and immense processing, signal processing together, so that we can monitor the growth of semi-polyweds. I'll come to that point, which is what it has, which is one of the most notorious way in the world. It looks very nice finding flowers, but practically it creates a quite health problem across the world, especially in the tropical region. So the idea, how do you use satellite imagery in order to monitor them remotely. So this is just to give you some context to why this is such a big problem and why we are interested to solve this problem by providing some process. So in this particular world, what we are trying to develop the tool say we should help to understand the policymakers or somebody responsible can help them what action plan they could take. So what this tool has and is, as I said, it's a major problem across the tropical region, and it impacts fisheries, drinking water sources, rice cultivation, navigation, recreational water bodies and lots of things. And it also leads to becoming breading ground for many diseases, especially the musculoskeletal, and that goes to diseases like chicken gonia and endangering human health. So I guess these are the some of the photographs captured all of the world with the problem starting from India, Africa, USA, Sicily, Florida area, and some parts of Europe as well, particularly in Portugal. So the way we are trying to tackle this problem, it's a multimodal problem, we're trying to take it, we're using satellite imagery, particularly Sentinel-1 and Sentinel-2 because they're free. But that comes with the problem, I'll tell you about that later. We had some drone missions, particularly who wanted to have high resolution data that we can map back in order to develop the models, develop the algorithms. And those are optical and multistakeholder. And whereas in the satellite, we took care of synthetic aperture radar or SAR, and optical and stakeholder. And finally, in the ground, we also wanted to promote optical and multisensors because we wanted to capture the water quality data. There is a direct relationship between the water quality and the growth and the growth cycle. And finally, we put them all together and to process it with data fusion. And some of the mess in the models, the ultimate aim is to create a higher regression map, interactive map, which people can use to see what is the growth in the certain targeted area. Just to give you a bit more context, this is one of the favorites that's in Southern part of India, that's in Kerala. And we had some drone missions over there. These are the some of the pictures we took. So we had a local drone company to fly over. And those are multispectral sensors on the drone payload. And these are some of the images. And it has to give a little more why this is a problem. Just to give you some idea, this is one of the drone missions we had. Kerala is a beautiful place. It has got lots of backwater. And this is very nice. These backwater channels, we used to have for fishing purposes or the transport purposes that can be over the years, 100 years of years. But what happened with invasion of the water has in those channels are typically choked. And only very large channels have been cleared and nobody has got that much of capacity to clear them out. So it is quite important to have them monitored and see what is happening and get back some of the ways to ensure the full security, particularly for the fishing purposes and the origin purposes. And this video I just wanted to show you because that shows the scale that shows the scale of the problem and why do we need digital tools to address those problems to monitor them. This is another UN organization based in India. They do quite a lot of agricultural research. Okay, part of this project, what we did, we tried to run some controlled experiment because it's not always possible to have everything from the field and we can analyze. So we just wanted to understand the growth cycle of the water has in sure what we did put the libelia migraeusensors in there and trying to see what is the growth cycle by placing water has in on this on this pond. And we have a very popular way of passing nutrients towards it. And that gives us some of the nice interesting result, which we can then compare it because we also do continuous picture, as we were doing the data. Right now, moving on to some of the results we have obtained from this project. So one of them is using the Sentinel one synthetic aperture relative of the solid data. But exactly we did. This is one of the legs, quite large legs, and we do some measurement from Sentinel one in November 2019. January and that thing, there was that way. And in that area, we have seen clear presence of water has in which was there. That was the actual picture with the mobile camera. Now, what we did measure is the backscattering of the synthetic aperture radar. And if you do the statistical analysis, you would be able to fully understand what is the distribution of those signals over the target area. When you compare them, you can actually create a chance detection map, which is clearly showing this is a process and you have a way of passing. And if you get in what happened, this is a benefit statistical modeling we put together. This is a mixture of washing dogs where you can actually create a box. You can see if the clean water is there, you can have a single nice one, so you can pick. But whereas for the presence of water has, you can already see a single double or multiple takes a mixture of question. Now, that leaves us to finally create a detection mask, which is more usable to somebody who likes to make some action of remedy elections. This is another site in Hyderabad. In this case, I'm going to show you how high regulation drone flights actually help us to understand the growth. We have actually picked up a few sites as well where we actually took the water quality data that we can see site 1345 and some of the pictures using citizen sense we created one app so that we can take images as well. So that further verification that has been quite useful. Now, in this case you can see we have taken this is from Sentinel to radar. And we have taken optical plus infrared as well. We have taken the data from various different dates and color representing if they're going to be something or not. Typically the talk who has for water has to be in stores. Bank of the lake. Welcome to where not. And what we have actually to get out with the citizen sense data to map these things. So what the interesting question I think all of us we face is the data leveling. So it's quite great, we can have all the data from the satellites for the historical data. But that data learning is painful. If you like to use any nice environments. So what we do, either you stick together, create polygons with fairing that we actually spend hours. There's another way to do which we think is quite useful what we did. We take random samples from the target area, and that was there we have grown using machine learning model something called random forest, and which we have verified with the data that actually is giving quite nice results and that can partially solve the data learning problem. Once we have that, then we try to segmentation using different techniques. Multiple techniques we tried, and also we combined modification of some of these techniques, and find the previous segmentation man. We are the second team and you can see the green area is the water has impact, whereas the blue is water and other visitations are like this kind of stuff. So having seen that those kind of algorithms that is working fine, but we had a solution, but that's not the end of the story, but you really like to see what are the challenges in these kind of approaches in the model. And that challenge is not only for aquatic vegetation monitoring, it's clear for everybody, every problem we try to solve using this monitoring model systems. And it's the ability of the data. We have seen European space agencies setting a data is free, but they are low resolution. One pixel represents 20 square meter, 20 meter by 20 meter, not large enough because if you're trying to see small water bodies or anything smaller than that will miss it. You can't probably get very high resolution data, either you have to pay to planet or picture or maybe Airbus Locked Martin. But the other way, what do you think we could actually use what we are trying to achieve for this one and future work by using the drone data which is very high resolution code five centimeter resolution, and creating the model, which is called super resolution, having the low resolution satellite data and the high resolution drone data, create the model, so that the low resolution drone data from 20 meter by 20 meter is free code that surely bring that to the resolution of one meter is going to be substantially useful for multiple different detections. So that's that's the kind of the goal from the competition side, I mean, my background is complex and so I look from that side. But that's not the only problem, because that comes with the registration problem. When you're flying drone, you're passing the satellites, they're not exactly same time. They're not exactly in the same location either. So how do you shop that problem. That's also a challenge. Finally, something we're working on for transfer learning, because the optical data, often problematic because rain cloud, any environmental issues, or sometimes satellite passes in the night, so what do you do. And we can probably use in that case, Sentinel one data, which is, which can penetrate through cloud, which is fine in the night, but the site data is very much noisy. So use the site data to fill the gaps within the optical or multiple strata multiple strata data is one of the key questions it is a challenge with us, but it is something doable, which we're going to work on. The problem is small object detection and the simplification from the remote data, or the satellite data is a problem. So that is something we're trying to work on. And finally, this is the Asia because not all the high ground with data, we want to transfer in our large server that takes a lot of communication problems. And then finally, can I do processing at the is, and can I do all this intelligent AI putting back that in the device that can actually process it and send the data that is minimal and that is required to send. So these are the some of the challenges we're trying to work on. And hopefully this is a generation you're pressing for working in this domain. And with that, I'm going to stop and have to take any question. Let's find out. Any questions for the floors. Yeah, I want to bring you out. Why are you in our factory. Sorry, could you please take it. I want to ask why people's factories is not the same to go on like, I know India has defense in which I'm trying to do this kind of work using one ceiling. So I believe in the factory, you know the variations in the defense in the internal. Yeah, that's that's absolutely correct. So we take care of the weather data, the rain and the temperature as well. So we started looking at data from African countries, so that it has got the right variation. And what we are more keen on understanding this, these tools and algorithms is transferable for various geographical locations. But yes, that is quite important to include the seasonal information within that. Thank you. And the second question for Steve, have you considered the jets and now it's not actually for your age processing. Yes. Indeed say I work on multiple different is processing device. One is Jackson Nano. Yes, but also we have created something a heterogeneous system, which has got this and nano, we have one carrier board where we put very small I don't know how many of you are familiar with that so there are three or four types of computer architecture so the process architectures. One is the CPU most computer use and say Raspberry Pi uses GPU will run lots of large machine learning models. They get some nano from Nvidia, they have got that. And the another is called field programmer to get ready. Basically you can design your processor, and that is very much energy efficient. You can do stream processing, but all of these different types of computer architecture are good for certain purposes. So what we're trying to see large algorithms like CNN, how do you partition them into different types of architecture to get based off all to give you a very fast response time, but the energy consumption is minimal. Okay, there's another question for Anna Smith. Deepak, might the new data hub offer an opportunity to address challenges around your data availability. What else might be needed so this is about the information architecture for observation images. I mean, in terms of the infrastructure, or in terms of the how the data should be curated. I'm guessing both. Yeah. The infrastructure and they're asking lots of advancements has happened from European space agencies and the data has been quite incredible, but still what we felt is understanding that data processing that to such an understanding everybody can understand what the domain expert that gap is still exist. That is what we're trying to do here. So after processing what we're trying to have interactive say Google Maps type thing, where you can just put your location. And that would try to show this is a another semi transparent layer of what it has in different growth states. So, taking from raw data into a usable visualization is quite important. In terms of the other data what we have curated. And one of the things we are trying to do haven't done that properly yet is the metadata information, because all the grown flags we have collected the proper metadata information is so important, even if you want to do any processing with it. So those are the things often missing.