Enhanced detection of Argulus and epizootic ulcerative syndrome in fish aquaculture through an improved deep learning model.

Abstract

Objective: Fish disease in aquaculture is a major risk to food safety. The identification of infected fish and disease categories present in fish farms remains difficult to determine at an early stage. Detecting infected fish in time is an essential step in preventing the spread of disease. The aim of this work was to detect fish infected with epizootic ulcerative syndrome and fish lice Argulus spp.

Methods: An improved YOLO (You Only Look Once) version 5 (YOLOV5) model was developed. In the context of transfer learning, our improved model used a pretrained model on binary images. The improved model was deployed and integrated into a Raspberry Pi board.

Results: The experimental results showed that it is more effective than a simple YOLOV5 model.

Conclusions: Using the evaluation metrics of precision, recall, mAP50 (mean average precision at an intersection over union threshold of 0.50), and mAP50-95 (average of the mAP values calculated for intersection over union thresholds ranging from 0.50 to 0.95 in steps of 0.05), our new model achieved accuracy rates of 0.944, 0.969, 0.989, and 0.954, respectively.