Automated feeders and computer vision to revolutionize fish farming in Africa
CMU-Africa professor Jesse Thornburg is using automatic feeders and computer vision to make fish farming in Rwanda more economical.
In fish farming, the highest cost isn’t the fish—it’s the food. From buying the feed itself, to the labor needed to throw it out by hand multiple times a day, up to 70 percent of the cost associated with fish farming can be attributed to feeding. But Jesse Thornburg, assistant teaching professor at CMU-Africa, is looking to change that by automating the feeding process.
“We saw that the farmers’ big need was in tracking and improving the feeding of their fish,” Thornburg said. “Automatic feeders were key for improving their metrics.”
Through the use of solar-powered automatic feeders and computer vision, Thornburg and his team at the Grid Automation for Development Lab are working with industry partners, including Lakeside Fish Farm, to make tilapia farming in Rwanda more economical. Other feeding methods like demand feeders, which release food when triggered by a fish, can lead to excessive or wasted feed. As a result, many in the global industry have adopted automatic feeders, which are shown to increase feed accuracy and reduce waste. According to Thornburg, automatic feeders can also improve feed conversion ratios, or the measure of how the volume of feed converts to fish weight gain.
Despite the growing use of automatic feeders in aquaculture, by-hand feeding is still the standard in Rwanda and nearly all of Africa. Yet, Thornburg’s latest review paper, published in the Journal of the World Aquaculture Society, finds that automatic feeding produces better outcomes, including greater weight gain, when compared to traditional by-hand feeding.
Source: College of Engineering
An automated feeder setup at a tilapia pond in Southern Rwanda.
To better track how much and how fast fish eat, Thornburg is using computer vision, a field of AI that allows computers to analyze and interpret data from images. He is specifically focusing on tilapia because they are the most-consumed fish in Rwanda and sub-Saharan Africa. Since tilapia eat floating feed, the team’s cameras monitor the water surface and capture images that are analyzed locally by microcontrollers (low-power edge devices). This system also facilitates image uploads to a remote server network, or cloud, where they are processed by the team’s software to enhance feeding algorithms.
After installing their first feeder in February on a tilapia pond at Lakeside in Southern Rwanda, Thornburg’s team is now in the process of collecting more image data to characterize fish behavior and optimize feed timing. But improving their feeding algorithm is not the only purpose of the camera system—it can also be used for surveillance to help farmers deter fish theft and offer reassurance that their fish are being fed enough. Thornburg is hopeful that the surveillance system will help support farmers regardless of whether an auto-feeder is installed.
“Not many farmers buy into auto-feeding immediately,” Thornburg said. “They’re hesitant to take that out of human hands and give it to a robot, an automated feeder, or to a company that is quite new.”
Through industry partnerships, Thornburg is developing a monitoring app that alerts farmers of when their fish are being fed or if any unusual activity, like theft, occurs at their ponds. In the future, Thornburg plans to develop a version of the app that sends alerts as text messages, allowing farmers without smartphones to still receive updates about their fish.
Source: College of Engineering
Thornburg’s team demonstrates how the automatic feeder works.
The data Thornburg and his team collect will also be used to help guide what goes into the feed pellets. Not only will this improve the nutrition of the fish, but it will benefit the people who rely on fish as a source of protein.
“Many Rwandans live near water and would like to buy fresh fish, but it was too expensive. We saw that with automatic feeding and optimization, we can drive the cost down below 40 cents per pound of tilapia,” Thornburg said. “We foresee this being even bigger than just characterizing how much and how quickly fish are eating.”
While Thornburg’s team is currently working in Rwanda, they have plans to expand their automated feeding system to other parts of Africa, including Egypt, Kenya, Morocco, Nigeria, and Ghana. To commercialize the solution and scale across fish farms in these different countries, they have formed a company called Aquabotics Limited and raised funding from Rwanda Green Fund.
The initial research work is funded by Rwanda’s National Council for Science and Technology. The research team includes Emenike Goodluck (MS ECE ’22), Emmanuel Annor (MS ECE ’22), Samiratu Ntohsi (MS EAI ’25), Emmanuel Adjei (MS EAI ’25), and Isaiah Essien, a student at the African Leadership University studying software engineering.