 So the next presentation is by Jochen Heming of Wageningen University, talking to the development of mobile fish phenotyping device. Welcome again Jochen. Hello, my name is Jochen Heming and I work as a senior researcher, computer vision and robotics at Wageningen University in Research in the Netherlands. Today I would like to talk on one of our research projects. This project is on the development of a mobile fish phenotyping device. Wageningen University in Research has the mission to explore the potential of nature to improve quality of life. On our campus we have a group on agro-food robotics that researches and develops computer vision and agriculture robotic systems for the different sectors, such as the open field agriculture, the protected horticulture, fresh chains and food, and also livestock and last but not least marine. We have a group of experts in artificial intelligence, sensing, especially multispectral and hyperspectral sensing, on machine learning, deep learning and computer vision. The Aqua Impact project is a European funded Horizon 2020 project with a consortium of 11 companies and 13 research institutes from 9 different countries. In Aqua Impact we integrate the fields of fish breeding and nutrition to increase the competitiveness of European aquaculture. Aqua Impact developed products and services based on genomic selections on the four spaces you see here on the right. Currently measurements of trades in breeding programs of fish are labor intensive, they are subjective and they are also slow. And part of our work in the Aqua Impact project is to develop technologies for high throughput phenotyping using machine vision methods. One requirement of such a device is that it must be able to measure on living fish. This fish is out of the water, it is sedated or anesthetized, but we have to limit the time that this fish is out of the water for the well-being of the fish, for example. The prototype we are currently developing can be seen on the picture here on the left. It is a small mobile device because it should be able to get transported to different places where measurements are taken place. It consists out of a box that shields environmental light, there is artificial light included. And a conveyor belt goes through that device transporting the fish that is manually inserted here. And next to a tag reader that reads the tag by ID of the fish, the most important components are the cameras we have inserted. There is one camera here on the top that takes images from the top of the fish, from the side. And there is another camera here that faces the front part of the fish and produces images as shown here on the right-hand side. We developed this apparatus together with the company Dorset in the Netherlands. And as cameras, we use not only color cameras, but we use color cameras that also produce depth images, 3D images based on active stereo. To give you a little bit an idea how that apparatus works, I will show a very early stage laboratory testing of that device. So you see here that the fish gets inserted and gets transported. Inside the cabinet it stops to take the images and make the measurements and then the fish is transported out. Normally in normal operations it was closed of course to shield environmental light. Also there is an integrated scale in the conveyor belt that measures accurately the weight of the fish on the conveyor belt. Here are some example images and some basic image processing operations that come out of this machine. In the middle you see the distance image, the 3D image, where the color tells you something about the distance from the object to the camera. So the more yellow-orange, the closer the object is to the camera. And we can use that 3D information for volume measurements. From the set of images we can easily extract all kinds of 2D trays such as length or width, the area, shape properties, roundness and other derived values. Next to that we can also measure calibrated color features. And as already said we have the full access to the 3D information so we can also measure volume of the full fish or certain parts of the fish. The big advantage of having a 3D system is that we can make use of automatically extracted landmarks to assess difficult to measure variables. Difficult to measure by hand variables such as specific body areas of body volumes. And that gives us much more possibilities than the manual trade recording that it used at this moment. The next step in this project is that we will combine the output of the image analysis with the genetic information we have available from our database about the genetic correlations for example. And by that we hope that we can unlock the full genetic potential for the breeding programs. We will develop machine learning based real-time selection algorithms for these breeding programs such that the machine can directly classify the fish into a fish that will be used for the breeding program and a fish that will be sorted out. Our key needs for further progress on that topic is that we need a very good contact and interaction with all kinds of stakeholders such as fish breeders but also machine builders. And on the other hand we always seek to team up and to share knowledge and also facilities on the area of artificial intelligence, of deep learning, also on the fish breeding and all related topics. And we do this typically in joint research projects. What we really need is funds that subsidize these kinds of high-tech developments because only then it will be able to set the next big steps on these developments. This already concludes my short speech. Thank you for your attention. Thank you very much Joachim and your team for giving us the insights there. That reminds me very much of the value of the crossover that exists between fisheries and aquaculture and the ability for, you know, in your speech how we link expertise and link ideas and it brings me back to Amanda's suggestion of an AI marketplace. And I wonder if you mentioned you're looking for funds to develop but what opportunities are there for collaborations to see working across teams that are working on similar questions to offer advantages instead of battling alone on certain issues. And I wonder how I already see that you're linking the genetic teams with the engineering teams with, you know, in controlled environments and the aquaculture but how would you see, for example, the Barton and teams linking in with maybe an international cooperation marketplace where we try to set out key themes where people working on different areas could go to share ideas, share codes, share opportunities. For funding and so on. Yeah, thank you. Thank you for the question. Yeah, this is, that is the, the topic I would say, indeed, as a little bit explained and also my colleague Anzalo explained yesterday so we are not coming from from the marine sector so we come from the plant but we are coming and then the interesting thing was to see that the same hardware if it comes to cameras but also the same software that we use to classify for example, a flower or a plant can be used in a very similar way to different products like the fish. Of course, there are also new challenges there but so that is, that's very nice because the same kind of technologies can be applied to very many different domains. And indeed, and that also comes very clear when listening to all the other presentations during the past two days, that there's a lot of things going on in the world and it does not make so much sense to reinvent the wheel so really we have to team up and that is also what we always try to do we always seek collaboration and then of course there are different mechanisms to do this and the European Union they have these European projects that we can, where we can acquire and that is also already an international cooperation but but yeah, there are also possibilities beyond that and yeah also sometimes it's a sort of bilateral agreement server several countries have to do joint projects but but it goes much more beyond that and we in Wageningen we are always open to step into these kind of collaborations. Max, do you have any questions for you? Yeah, sorry not so much collaboration but really sort of technical thought process. I've worked in high-respectful multispectral sensing in agriculture on drones and also underwater transects of reefs and aquatic ecology and gathering 3D spatial data as well. Obviously, can be incorporated with LiDAR if you're looking at crop production. What lessons have you learned from combining 3D data and hyperspectral data with fish and how have you handled the 3D data as well? Are you producing vector data sets, meshes and then giving the algorithms measurements from those to look at the reflectance value? How are you handling that data pipeline? Yeah, I mean, first of all, it's a massive amount of data so we're really talking about this magical big data you have to deal and and certainly if you want to approach that in real time that then it gets a big challenge. I think we are still quite at the beginning when it comes to having combined data of 3D and hyperspectral and 2D and so on. So that is there, the research is really at and just had started and there are so much more possibilities still. But we also see the last potential of course because 3D is something when you go more into detail that you cannot assess easily and objectively with the human eye. So and also for sure the hyperspectral I mean this project which I just showed here that that does not make use of any spectral information but there are other projects that are doing this and there's a lot of things to win actually to use this information. But then the issue or the challenge is to get this data processed in the available time. And maybe it's a little bit comparable to what in the previous presentation there's an enormous amount of video data collected and you only need a very small fraction of that to really extract the information. If you record the full hyperspectral data set then you have a lot of information that is not useful, and then you need clever and smart tools to extract only the useful information. That's great. Now I've actually got another question for you from Anton Ellenbrook. He's from the information and knowledge management section of the issue. Yes comments from your data collection perspective. So what is, are there any steps where they are going on for calibration of and for the validation of instruments to measure fish or plants or whatever. That's in front of your camera and do you also see that as a role where international organizations like FAO can have a sort of a stake or help to define the standards and classifications for optical and other measurements on life and in eight objects. That's a good question I would say. I think if it comes to project that like these documented fisheries and so on where you have really sort of legal tasks, or where you try to use these automatic systems for for doing like sort of certified measurements. I think it will be essential also to specify and to put some norms out or to develop them and to specify them when it comes to calibration and these kind of things. Yes, sure. That's not something we are looking to at the moment, but I can imagine that certainly in the fishery. There can be a very important task. And if you if you use that information not only to sort of to make the profit for the, let's say the fisherman better because they can sort their fish or something like this but you really have to supervise and you have to be sure that you collected the right information, then, then the question of giving out norms and regulations will be a very important one. Yes. Thank you very much. I just want to add one point that the idea of norms and standards is not necessarily only for compliance. It's the ability for these data sets to then become very shareable between groups so we can try to agree. And this is one of the things about an agency like FAO. It's not really an agency it's an amalgamation of all the country's governments and getting them together to decide on the standards that they will use often allows great sharing. Let's see where we're going with this but let's let's move on. Thank you very much for the presentation.