Abstract

In intense fish farming, continuous identification and surveillance of prevalent infectious diseases are crucial for formulating scientific methods for fish disease avoidance, which may significantly mitigate the death of fish and financial damage. Nonetheless, subpar underwater imagery and poorly identifiable targets pose significant obstacles to detecting infected fish. This research proposes an Automated Disease Identification (ADI) system using Underwater Imaging (UI) and an Improved YOLOV10 Network to address these problems. This work introduces an innovative residual awareness unit referred to as RAU. This component is included in the main structure of the YOLOv10 model to enhance the system's attention to the intricate aspects of targets in biology during the gathering of features. Using a bilateral feature triangle (BFT) with a dynamic combination of features in the head network augments the amalgamation of contextual data from deeper layers. At the same time, localization signals from shallow levels boost the model's capacity to differentiate objects from their surroundings. Studies conducted at a fishery platform indicated that the enhanced YOLOV10 framework outperformed the baseline YOLOV10, with the mean accuracy rising from 94.36% to 99.75%, reflecting an improvement of 5.39%. The Enhanced YOLO10 system can proficiently identify unhealthy fish and is suitable for large-scale aquaculture.

Keywords: Automated disease identification, YOLOV10, R-AU, Aquaculture, underwater imaging, Fish disease.