Enhanced YOLOv5 Model for Real-Time Fish Species Recognition on Jetson Orin Nano

This study develops a real-time fish recognition model using eYOLOv5, implemented on Jetson Orin Nano for an interactive display system. Tailored for fish identification, it displays results and 3D images instantly, supported by a database of 50 underwater species, addressing limited guides in marine parks. The Jetson Orin Nano ensures portability and efficiency as an embedded platform. The eYOLOv5 model, based on a CNN, incorporates adaptive activation functions and attention mechanisms, achieving 97% accuracy after training on 10,000 labeled images. It enhances recognition accuracy and user experience, showcasing deep learning potential in underwater identification. Jetson Orin Nano’s high performance, low power use, and compact size enable real-time processing without cloud reliance, reducing latency and boosting stability, portability, and flexibility. This system suits marine exhibitions and education, delivering seamless recognition.