Fish Disease Detection of Epizootic Ulcerative Syndrome Using Deep Learning Image Processing Technique

Abstract

Early detection of fish disease is a vital acknowledgment in Fish Cultivation since it could track later effects on the fish ponds since the disease could be contagious. Visual identification is the common method that is used by the majority of conventional cultivators in Indonesia. By applying Image Recognition technology, disease identification could be executed efficiently from process latency and the amount of batch that could be identified. Our limitation in this paper is EUS (Epizootic Ulcerative Syndrome) developed by a pathogenic fungus, Aphanomyces invadans. Common identifiers of infected fish could be identified by visual inspection acknowledging red blotches in the fish bodies. This research used Object Segmentation Inference MobileNetV2 as model system architecture and Image Processing. Using HSV Threshold base to identify which part of the fish is infected by the disease. Object Segmentation will separate the disease area from the fish's whole body. Meanwhile, the HSV Threshold setting in will identify red blotches in fish bodies. Then fish bodies that have been infected by EUS will be shown by the binary result of the HSV Threshold. To see the performance of the system we could see the indicators shown by the F1 score, as the result of the image identification from 80 augmented data we have an average score is on 84% accuracy. That means we could assist conventional fish farmers by identifying diseases faster and helping cultivators with minimum knowledge about identifying fish diseases.