Deep Learning-Based Underwater Vision System for Real-Time Fish Behavior Analysis and Health Assessment

Abstract

Monitoring of aquaculture underwater has gained more significance in the context of a sustainable fish production in order to prevent diseases and to provide the smart management of farms in the context of modern smart fisheries. Nonetheless, the traditional monitoring methods are greatly impaired by the fact that the image is compromised underwater due to light absorption, scattering, color distortion, low contrast, and motion blur reducing the accuracy of fish detection and behavioural interpretation. Moreover, the absence of automated real-time surveillance systems prevents the early detection of abnormal fish behaviour and health related problems in the aquaculture setting. To overcome these challenges, this study will introduce a deep learning-based underwater vision system to analyze underwater fish behavior and health in real-time projects in smart aquaculture. The architecture proposed combines the underwater image enhancement and intelligent fish detection in a single real-time analytical framework. Firstly, CLAHE and deep enhancement technique is used to improve contrast and visibility of underwater images to enhance the quality of underwater images under harsh conditions. After that, the optimized YOLOv8-based object detection model is utilized to locate and detect fish in real-time and with accuracy. The framework also includes behavioral analysis mechanisms to observe swimming patterns, movement abnormalities and stress related activities related to fish health status. Experimental analysis was performed with common underwater fish datasets and performance metrics such as mean Average Precision (mAP), Precision, Recall, Interaction over Union (IoU) and Frames Per Second (FPS). The suggested system had high detection rates with increased real-time processing capacity and better underwater visual observation than those that were in use. The results received indicate the usefulness of the suggested framework to obtain credible fish capture, intelligent behavioral, and automatic health measurement in dynamic underwater conditions. The proposed research will lead to the creation of AI-based smart aquaculture systems, as it will offer a scalable, real-time, and smart underwater surveillance system to the next-gen fisheries control.