Underwater abnormal classification system based on deep learning: A case study on aquaculture fish farm in Taiwan

Abstract

From the perspective of aquaculture systems, a grouper is bred in a high-density manner which can cause stronger contagion and a higher risk of infection. In addition to routine inspections, identifying the abnormal appearance or condition of the underwater fish in advanced via computational intelligence and taking further isolation measures will help reduce the chance of infection of other fish. This paper introduces a two-phase ImageNet pre-trained deep learning model with Convolutional Neural Network (CNN) structure which is able to classify three types of the abnormal appearance of grouper. The dataset contains 7700 underwater fish images and 11 classes, including nine common Taiwan high-economic-value fish species and the grouper with the normal and abnormal appearance. The experiment implements four ImageNet pre-trained models and validates with empirical image data. The experimental result reveals InceptionV3 pre-trained model for classifying three different types of abnormal appearance of grouper can reach average 98.94% accuracy in phase II task.