Not for profit, Open source, Market Shaping Phase, Open to new members is building the largest crowd sourced fish image collection to teach machines how to identify fish species using ai.

The Problem

Currently, there isn't a highly accurate Fish Recognition artificial intelligence technology that is open source for fish identification that focuses on immediate fish identification worldwide. Incorrect fish identification due to human error or sampler bias can have negative effects for many different stake holders. Fisheries scientists need effective management in conservation efforts and this includes correct fish identification from dockside validation efforts, fisheries dependent monitoring, and creel surveys. Efficient and effective training and continuing education tools are important as well to ensure quality control. Academia based research teams and labs spend hours in the field collecting data and then twice as many hours in the lab identifying specimens. Additionally, there is a financial burden on paying for the technician’s time and travel for the hours in the field and the lab as well as sampler bias and human error in correct identification. Training new employees on proper fish identification also takes up time and resources. Correct, quick species identification can be a challenging task for anglers, especially when in unfamiliar waters. This can lead to severe consequences with the law for the angler, and most times unknowingly. Potential for species mortality increases with increased time it takes to correctly identify landed fish. Over time this can accumulate to have negative impacts on a fishery.

Our Proposal

The solution will be to create an open source model that can be used to create a successful artificial intelligence fish identification program. This technology has the potential to be impactful in many different arenas. Fisheries scientist will have the ability to implement the open source fish identification model into their already existing platforms on smart devices, this could then be used in place or supplement to dockside validation efforts, fisheries dependent monitoring, and creel surveys. has the ability to reduce human error, sampler bias and to minimize resources to conduct fisheries management efforts. Fisheries agencies may also implement the model as a training source for biologist and enforcement officers. The open source model from can also support research efforts conducted by academia lead research teams if it is implemented into a platform of their choice, or they can build a platform around the model. The model can be used in the field to identify fish species as they are collected, saving hours of lab time in identifying specimens. Once back in the lab the images can be reviewed by a technician on a smart device or computer. The images collected from the field can also be used as a source for training and continuing education within the lab. Commercial use of can include apps for recreational anglers that help reduce misidentification, and to novel approaches to solve conservational woes.

We Assume that...

Constraints to Overcome

Artificial intelligence that can identify fish species has been attempted before. Between not using enough images to train the model to lack of technology available at the time, the program failed. We plan to avoid both of these issues. We have partnered with the FishAngler App to have access the hundreds of thousands of fish images on their platform to train the model. We are building a web based portal to create a citizen science opportunity for individuals and all organizations or groups. Users of the portal will be able to contribute to the training and testing of the model by uploading their fish pictures. Between images collected from the portal and FishAngler we will be able to ensure a wide variety of fish images to train our model, allowing it to be as robust as possible. Our team is committed to using the best technology available to build the model. We will be testing several different methods and using the best fit that allows for the most accuracy.

Current Work

Creating a database that has the identifying attributes of each sport fish species. Following the completion of the database the team will work on processing images to use for the training sets. This will consist of properly identifying the fish species(s) in the photo and tagging all of the appropriate identifying attributes to each fish. We will also have a team working on building the website portal for citizen science. Here web-based users will be able to submit their images to be used for the training of the model.

Current Needs

We will need interns to help with processing the images for the training batch (identifying fish species in images and their identifying attributes). We will need software developers, database developers, web developers, AI experts, AI developers, and marine biologist.