A new image dataset for the evaluation of automatic fingerlings counting

Abstract Fingerling counting in fish farming is still a problem to be solved, although there are some technological solutions used in experimental cases and controlled environments. In this paper, a fingerlings counter is evaluated using a new image database with 448 videos of the Pintado Real® fingerlings species. A total of 21,402 combinations of 6 parameters of the fingerlings counter were tested. In the training phase, 314 videos were randomly selected, which represent 70% of the database. The remaining 30%, corresponding to 134 videos, were used for testing. Focusing on the parameters that best recognize and track the fingerlings of Pintado Real®, it was obtained Pearson Coefficient of 0.9803 and a quadratic average error of 2.65 when comparing the manual and automatic counting. The results obtained six parameters sets that achieved these metrics, reaching higher performance on the fingerlings counter from a new image database. The image database used in our research is available for researchers.