CNN-based fish iris identification.

As in many other areas, digitization is also on the rise in intensive aquaculture. A vision for the future is continuous monitoring and the recognition of each individual fish in the system. Previous work has shown that Atlantic salmon can be recognized using lateral and iris images. For salmon iris identification a traditional texture feature-based approach was used. Results indicated a high distinctiveness but a low stability of the salmon iris. In this work we employ a CNN-based fish iris identification approach and reassess the previous results. One question is whether a CNN-based approach performs better in terms of long-term stability. Furthermore, a second database for European seabass iris images is used in the experiments. This makes it possible to check the applicability of iris identification in another fish species and whether the statements regarding distinctiveness and stability are also confirmed here. Results show that the CNN-based approach performs worse compared to the texture feature-based approach. Same as for the salmon iris a high distinctiveness of the seabass iris but a low stability can be reported.