A computer vision approach for detection and quantification of feed particles in marine fish farms

In the realm of marine fish farming, there is increased focus on employing numerical models and tools to optimize production. A model describing the distribution of pelleted fish feed in time and space within a sea cage, a process which is essential for proper fish growth and welfare, has been established, but proper data for model validation have been scarce. A device based on computer vision which is able to accurately quantify the feed density within a specified volume of the sea cage as a function of time was thus developed. This paper describes the physical design of the device, as well as the application and combination of well-established algorithms to reliably detect and quantify feed pellets. Results from tests using realistic feed densities showed that the device was capable of detecting and quantifying with an error of 1.3 %.