Aquarium Control System using Yolov5

Detecting and monitoring common infectious diseases in intensive aquaculture settings is crucial for implementing effective scientific strategies to mitigate fish mortality and economic losses. However, challenges arise due to poor quality underwater images and difficulty in identifying targets, posing significant obstacles to diseased fish detection. To address these issues, this study proposes a model for diseased fish detection in aquaculture using the YOLOv5 network. Given the extensive utilization of underwater cameras in marine ecology applications, there is a pressing need for automated tools to process the vast amount of generated data by detecting and recognizing species captured in natural environments. The classification of fish species from videos and images in such settings is particularly challenging due to noise, illumination variations, and diverse habitats. In this paper, we introduce a two-step deep learning approach for detecting and classifying temperate fishes without pre-filtering. The first step involves detecting individual fish in images using the You Only Look Once (YOLO) object detection technique, irrespective of species or sex. Subsequently, a Convolutional Neural Network (CNN) architecture, namely Fish4Knowledge, is employed for classification purposes. Both the object detection and classification models are trained and updated with data specific to temperate fishes of interest, enhancing their applicability and performance in real-world scenarios.