A Comparative Analysis of Deep Learning Approaches for Fish Disease Identification

Abstract

Bangladesh's fisheries and aquaculture industries play a significant role in the nation's food production, ensuring the food supply's nutritional security, growing agricultural exports, and employing 17 million people across various occupations. Farmers who farm fish face a lot of economic losses every year because of various diseases that can happen to fish. There are three common diseases of fish. They are known as black spots, red spots, and white spots. A parasite causes black spot disease. Red spot disease is also known as Epizootic ulcerative syndrome (EUS). And it is caused by a fungus. The white spot syndrome virus causes white spot disease. If a fish farmer can detect these diseases early and apply appropriate treatment, it may protect much infected fish and prevent economic loss. The manual approach of human visualization is a laborious effort for detecting and monitoring fish disease. As a result, any viable strategy that is quick, accurate, and highly automated encourages interest in this problem. Due to a lack of information and a high level of competence, there hasn't been a single piece of useful research on the fish disease. Our system provides solutions to this problem. Fish disease identification using deep learning from images is an arduous task. This study proposes a multi-classification deep learning model for identifying fish diseases (black spots, white spots, red spots, and healthy) from images. To classify and identify these diseases, we will apply five different pre-trained models (DenseNet121, MobileNetV2, ResNet101V2, ResNet152V2, and VGG16), and we have also compared their accuracy. According to the experiment data, the MobileNetV2 model performed better than the other proposed models. In comparison to other models, the model provided good detection accuracy. Keywords— Fish disease identification, Classification, Tensor Flow, Dataset, Deep learning.