 Our next presentation is by Lisa Kellogg and Harshil Shah who are talking to AI for species identification in recreational fisheries management. Welcome Lisa and Harshil. Hello, my name is Lisa Kellogg and I'm a researcher at the Virginia Institute of Marine Science. Thank you for joining me and my colleague Harshil Shah from DX Factor to learn more about our wreckfish project. The mission of the wreckfish project is to develop a free, useful, accurate and user-friendly fishing app that encourages recreational anglers to collect and contribute catch data to improve fishery science and support sustainable fisheries management. The wreckfish team consists of several researchers from the Virginia Institute of Marine Science, a member of the VIMS Innovation Fund Working Group, and staff from DX Factor. In developing wreckfish we're using what we think of as the eBird model for engaging community scientists. The eBird effort provided services valued by birders and then asked them to contribute their data. They help birders with things like identifying their birds, finding species of interest and keeping a life list. The eBird database recently reached 1 billion bird observations contributed. Using a similar approach, we hope to engage recreational anglers in Chesapeake Bay as community scientists. Chesapeake Bay is ideal for this project because of the diversity of species found there. Right now we are attempting to model 281 fish species found in the bay. Once we've proven the concept in Chesapeake Bay, we hope to expand up and down the Atlantic coast of the U.S. and then eventually around the world. Once fully developed, the wreckfish app will include eight core features. First and foremost, the app will allow anglers to accurately identify their catch. The species name given will be the preferred common name of the American Fisheries Society, and we will also provide the confidence of the model in that species identification. The fish length will also be measured automatically, and an approximate weight will be given based on link to biomass relationships. Using information on the species identity, the length of the fish, the location of the catch, and the date, we will instantly provide anglers with information on whether their catch is legal to keep or whether they must put it back in the water. We will also provide information based on local health department fish consumption advisories on whether their catch is safe to eat. Anglers will have the option to stamp information about the species identity, the length and weight of the fish on a photograph of the fish. They will also have the information recorded in a logbook that provides them the opportunity to add additional information of interest to them. Finally, recreational anglers will be able to upload information on each fish they catch to our database with the touch of a button. To identify species, wreck fish is using a two-step process. The first step in the process is object detection. To find the fish in a photo, we are using the faster region-based convolutional neural network model. This model places a bounding box around the fish in an image and then removes the background. The second step in the process is species identification. To identify the species, we are training multiple convolutional neural networks that have different architectures to identify our species. At present, we are using the Inception ResNet version 2 and EfficientNet B5 models. These convolutional neural networks are then combined into an ensemble model that provides the final fish identification. In addition to identifying the species of the fish caught, the wreck fish app will also measure each fish. It will do this by implementing augmented reality functionality, specifically Google's AR core technology that is available for both iOS and Android. This allows measurement of the fish without an object of known size in the image. We will also make sure that the length measurement is appropriate for the species identified. In some cases, this will be total length, but in other cases, it will be fork or standard length. So far, we've been successful in modeling 60 different species of fish from Chesapeake Bay with a mean accuracy of over 95%. We've also developed a website for anglers to upload photos to help with model training. This summer, we will also be releasing a photo upload app. We anticipate that this app will increase the rate at which anglers upload photos to the wreck fish database to help with model training. Right now, we're continuing to focus on modeling additional species of fish and on developing other core app features. Recently, we've proposed adding two new functions to the wreck fish app. The first function that we would like to add is an invasive species alerting and reporting function. Once integrated, this would alert recreational anglers each time they catch an invasive species and would tell them what to do with the fish that they've caught. It would also offer them the opportunity to report their catch to the appropriate management agency with the touch of a single button. The second feature that we would like to add to the wreck fish app is one that provides infrastructure for other researchers to incorporate recreational anglers as community scientists into their own research project. What we have in mind is creating QR codes for individual projects, providing those codes to the researchers who in turn provide the code to the recreational angler. The anglers would then include the QR code in all photos they take of fish for that project. Our models would be trained to detect photos that include QR codes and send them to the appropriate researchers. The primary challenges that we're facing at present are finding enough photos of rare or rarely photograph species and in developing a viable long term funding model. With that, I would like to thank you for taking the time to learn more about our wreck fish project. I would also like to thank the groups that have provided funding and support for the wreck fish project. And most importantly, I would like to thank the recreational anglers without whom this project would not be possible. Thank you very much Lisa. It seems like recreational fisheries are offering us a rich environment to develop AI systems. Alongside the work that you're doing, we're just hearing a lot in Europe about the Dutch angling app, Maine Fistmart, which was only launched in 2020 as already have over 100,000 images added and is getting about 500 records a day coming in. So these are the kind of groups we'd like to link up. Peter Bieland from the Sports Fisheries Netherlands would be a good link with your team and hopefully we can use these teams to help push forward. And there's a whole conversation there about how to get the size and the linking with QR codes or other. But my question to you is, do you see the funds available for this kind of development from wealthier recreational fisheries joining with the funds from commercial fisheries where they have common aims? And what other mutually beneficial opportunities do you think we can leverage by bringing in the recreational fishers to understand the overall stock health? Thank you. Thank you for the question. I really think that there's a huge potential within the recreational angler community. The experiences that we've had with recreational anglers, they're incredibly enthusiastic and they're helping us through every phase of this. They're serving as beta testers. They're making sure that things are easy to use before we put them out at a larger scale. So I really think there's a lot of desire within the recreational community to actually contribute and give back. And so I think the potential there is huge. And I think, you know, with the rec fish app that we're developing, we're trying to make it as adaptable and scalable as possible. And I really don't think it would take much to adapt it to some small vessel fishery applications in places where there's not much existing data. Two questions coming your way. Two questions coming your way one from another Dutchman up in the north there of the screen, and then Ellenbrook but also Matt Matt would you like to go first? Thanks, Kim. Thanks Lisa for a really great presentation. I was just interested if you had any plans to gamify the image store. I know there's been some success with, you know, kind of the top crumps of collecting data. And that also relates to where are the images stored. Does the, does the contributing person using the interface still have access to their pictures. Do you have a record for them as well with the log and they still can see the pictures or, you know, what's your long term plans to incentivize it? Well, and we're still we're still working on that we're quite early in the project. And so we certainly have considered gamification. We haven't gotten there yet, to be honest. You know, we've thought about doing some things like fish quizzes and let people help us with the identification by giving us what they think the fish ID is to sort of streamline and make our QAQC more efficient and sort of give us some free labor on that front. And I'm sorry, what was the second part of your question? It was really about gamification. I was wondering if you could try and involve sort of, you know, recreational compliance or win something for, you know, being a good recreational angle. You know, to me, it was really about gamification. And we've spoken a lot over the past few days about incentivizing the reasons why stakeholders will use these technologies. You know, there's, there's a lot of cases where we've seen the technologies is being developed and it's excellent. The actual use case on the ground is for me a really critical question to look at before that, in my opinion, before you even start developing the technology. Yes, and we're, you know, we're working on that and we're we're literally testing it out with the recreational anglers to see what, what attracts them to using it and paying a lot of attention to that and working with the anglers. To be honest, the group that we're working with right now is so eager that we really don't have to incentivize them. They just want to contribute and be a part of it because for years they've been frustrated that they couldn't provide more data to manage the fish that they target and they get sort of frustrated by some of the management actions that are at a sort of broader scale than what they're actually seeing in their local fishing spots and so offering them the opportunity to actually record data that's going to be considered a valid data at the end of the day. I think is something that's actually very attractive to a lot of recreational anglers in the US. It's amazing what a young angler will do to get one of those hats from one of the fishing companies that and your turn. My question was basically the part that you forgot about Matt's first question so long term storage and how can other people use data from other anglers if they either as a scientist want to develop a new model, or use the data for a reason how do they get access to this then fantastic big data set. Well we were still working on creating that data set so the back end we're right now it's just it's just Azure storage in the cloud and we haven't built the back end of the app yet because we don't have a fully functional app yet but one thing that we are working on is with with this upload app anglers have the choice to upload their photos either for wreckfish use only for research only or to make it no rights reserved and so those the no rights reserved photos we can make publicly available to anyone who wants them. For research only we're requiring a data sharing agreement that you agree not to share those publicly and only use them for research or they can choose to just have them held privately by wreckfish and no one else ever sees them. Thank you very much Lisa and these are the kind of standards that you know people will start to settle on what what other terms and so on. Thank you for joining in the background there.