An Ensemble of Small Convolutional Neural Networks for Identifying Monogeneans of the Order Dactylogyridea from Digital Photographs

Abstract

Fish diseases pose a serious obstacle to the economic stability of aquaculture. Correct and timely identification of the causative agent is a necessary condition for carrying out veterinary and sanitary measures, making it possible to prevent mass fish mortality. However, many fish farms lack the laboratory infrastructure and specialists required for detailed analysis. This paper proposes an ensemble of three compact convolutional neural networks for identifying monogeneans of the order Dactylogyridea in photographs obtained through the eyepiece of a light microscope using a smartphone camera. The biological material used to train the neural networks was collected during studies conducted by the Southern Scientific Centre of the Russian Academy of Sciences (SSC RAS) in southern Russia in 2019-2026.