Detection and Quantification of Motile Aeromonas Septicemia (MAS) Disease in Common carp Using Deep Learning

This paper was not presented at the conference.

Abstract

This paper was not presented at the conference.

Common carp (Cyprinus carpio) is one of the freshwater fisheries commodities that has important economic value, so it is widely cultured. However, common carp farming is susceptible to diseases, such as Motile Aeromonas Septicemia (MAS), which can lead to mortality. MAS disease is visually detectable, but continuous monitoring challenges arise for farmers. This study employs the YOLOv8 and DeepSORT algorithms to automatically detect common carp and wounds caused by A. hydrophilla bacterial infection, obtaining wound area and swimming speed data. Training was conducted on a dataset with two different labels: infected fish wound areas and entire fish body, using consecutive epochs of 3000 and 1000. The training results show wound label accuracy reached 91.49% for disease concentration 10⁷ and 88.68% for disease concentration 10⁸, respectively, while the accuracy for fish label reached 96.61% for disease concentration 10⁷ and 93.44% for disease concentration 10⁸. The coefficient value of predicted wound area against actual wound area approximates 1, indicating a close match between the two variables. The obtained fish swimming speed estimate reflects the lethargic behavior of fish due to MAS. The results demonstrate that the model is effective in detecting fish, wound areas, and swimming speed.