Researchers at JCU have developed a tool which could streamline the process of breeding fish.
The tool utilises AI technology to scan fish and identify which have the ideal feature for breeding.
The Mobile Fish Landmark Detection network (MFLD-net) uses a conveyer belt, industrial camera and an artificial intelligence algorithm to detect the ideal characteristics of barramundi in order to improve selective breeding practices.
Associate Professor of Engineering at JCU Mostafa Azghadi says they are aiming to eventually implement the tool right across the industry.
“The idea is to gather as much data as possible on an industrial-scale on the characteristics of the fish from an image using computer vision and machine learning,” he says.
“The data captured may help aquaculture farmers with decision making – for example, which animals should be put together for selective breeding, or grading animals to sell to a particular market. This process can now be automated and scaled, saving a great deal of time and money.”
The algorithms, developed by PhD candidate Alzayat Saleh, firstly locates key data points on the body of a fish before using the relationships between landmarks to predict the weight or other characteristics of the animal.
“In order to train the algorithms, you need to begin by manually placing the landmarks on images from thousands of fish as a starting point,” Associate Prof Azghadi says.
“In our case we have done this for over 2500 fish images.”
