A computer-vision system and methodology for the analysis of fish behavior

Abstract

Behavioral variations of small fish populations are difficult to measure quantitatively. To quantify such measurements, a low-cost computer vision system has been developed to analyze fish behavior in aquaculture tanks. With this system, 9 tanks can be observed simultaneously, enabling the study of one factor, in three triplets for further statistical analysis. The system enables the observation of the tanks at all times, with the use of a web publishing tool, while it allows the remote control of the acquisition to eliminate behavioral variations that might otherwise be caused by human presence. Evaluation of the system was achieved by measuring fish interaction (inspection and biting) on three different net conditions. Measurements were completed in three experimental sets, using stocking density as a stress factor. Results clearly demonstrate that the system successfully recorded fish behavior with minimal frame loss (<21 s in 24 h), while analysis identified every fish interaction with the net. In addition, the measured variations of fish behavior within a single day showed no statistical differences. In conclusion, an inexpensive and efficient computer vision system is presented, assisting in the monitoring and analysis of fish behavior.