Bryton Shang started Aquabyte with the idea of using machine learning and computer vision to make fish farming even more efficient. Barely three years later, Aquabyte has grown from a prototype in a bathtub in San Francisco into an established software company that makes work easier for fish farms along the Norwegian coast, every single day.
Automatic lice counting has been launched
Aquabyte has launched automatic lice counting software based on artificial intelligence and machine learning, and soon a module will be available to measure the weight of fish in the cage and provide an estimate of weight distribution and biomass.
For Bryton Shang and Aquabyte, things have happened quickly, but Shang is used to that. Aquabyte is his fourth company since graduating from Princeton University in 2012.
After having developed algorithms to streamline stock trading on Wall Street, and to diagnose cancer in tissue samples, he began looking at other ways of using machine learning technology. As part of his team, he had the support of NEA, one of the largest venture capital companies in Silicon Valley.
From prototype to reality
- After a while, I started to look at how I could apply this technology in the aquaculture industry. I had an idea about developing a system that could use computer vision to measure the weight of the fish. I made a simple prototype in my bathtub at home in San Francisco and bought some robot fish on Amazon to see if I could calculate the size and weight of the fish with the help of a camera and computer algorithms. I used two lenses in order to be able to measure the distance to the fish, make a 3D model and then estimate the weight.
The prototype was successful, and the next step was to make the system work in real life: Under far rougher conditions, in a cage with tens of thousands of fish.
- I considered various fish farms in California and British Columbia but had heard a lot about Norway and Norwegian aquaculture, and about Aqua Nor in Trondheim, so I ended up booking a trip to Norway.
Convinced about Norway
- I was amazed at how big the aquaculture industry was here, and how well developed it was. I was also convinced that Aquabyte had to establish itself in Norway in order to become a successful company.
Bryton Shang brought this enthusiasm back to his investors in the United States. They had barely heard of the Norwegian aquaculture industry and were sceptical, but he still ended up raising $ 3.5 million in new investments, not only from American investment companies and universities, but also from Norwegian Alliance Venture.
This meant that Aquabyte could start up in Norway. The first two employees were in place by the end of 2017, and in the course of 2019 the number has grown to 40, divided between Silicon Valley and Bergen.
American innovation culture and Norwegian aquaculture expertise
- It has been important for us that this should not be a Silicon Valley company with a fully developed technology that we just brought into the Norwegian market. It was necessary to develop the products and the company here, as a Norwegian entity in its own right. We wanted to tap into the professional knowledge and listen to the needs of the Norwegian fish farming industry.
Shang sees this combination as Aquabyte’s greatest strength.
- We want to combine the best of the innovation culture from Silicon Valley with the foremost expertise from Norwegian fish farming. That is why we are represented in both places, not in the form of a head office and a branch office, but as two equally important working environments that are in continuous dialogue with one another. It is the combination of these two environments; technology and aquaculture, that is so powerful. Interestingly enough, our Bergen office is now larger than our San Francisco office, and the development of technology in San Francisco looks to Norway. The requirements here determine what software needs to be built, not the other way around.
Another important goal has been to concentrate on being a pure software company.
- This is a conscious choice. Some of our competitors produce physical hardware in addition to developing software and algorithms, all on limited resources. We have dedicated all our resources to building the best software and algorithms and getting them tested. We then prefer to enter into partnerships regarding hardware. This is how we have been able to grow as quickly as we have done.
An example of such a partnership is the one Aquabyte has entered into with Imenco Aquaculture for camera equipment.
Norway is good at developing underwater technology, so we leave it to Imenco to develop good camera solutions, while we build the best software. Unlike some of our competitors, we have opted for a standard camera, as opposed to a hyperspectral or multispectral camera. It uses traditional RGB technology, just like a regular camera, or your phone, for that matter.
One camera for everything
This presents a number of additional challenges in terms of algorithms.
- It also makes it more difficult to analyse the image. You need even more advanced algorithms, but the advantage is that it has more areas of use.
This is our vision: You do not need a separate camera for lice, one for biomass and one for feeding. All software can run on the same camera, and we develop all our products and algorithms with that in mind. This allows us to make more services available without having to upgrade the equipment package. For example, customers who purchase our lice counting system now will be able to upgrade it easily to run weight measurement and biomass estimation. It is a software upgrade that uses the same type of hardware.
The quality of the algorithms is the key to success:
- Being able to interpret images taken underwater is quite complicated. You have to deal with different light conditions, particles and turbidity in the water, as well as the behaviour of the fish - there are a number of different factors to consider. Building a straightforward algorithm to estimate weight or count salmon lice is not the hardest part. The difficult thing is to get the algorithms to work in all these different environments, and it requires a series of different algorithms: For example, some to identify which images are good enough to use, some to identify individual fish based on dot patterns, and some to identify lice. And they need to function in difficult and changing conditions.
Practice makes perfect
The most important job is to train the algorithms.
- At the beginning, the algorithm does not know what a fish weighs or what salmon lice look like. We have a team of experienced professionals who train the algorithms so that they become increasingly precise.