Confidence Based Ship Detection Using YOLOv8

Abstract

Maritime monitoring and ship detection play a crucial role in ensuring coastal security, managing maritime traffic, and supporting environmental surveillance. This paper presents an efficient and lightweight approach to ship detection using YOLOv8, a state-of-the-art object detection model, applied to both satellite and aerial imagery. The proposed system is designed for real-time inference and enhanced user interaction by integrating a dynamic confidence threshold slider in the web interface. This feature allows users to fine-tune detection sensitivity on-the-fly based on image conditions and detection accuracy preferences. The model is trained on a diverse dataset containing annotated images in YOLO format, ensuring robustness across different imaging conditions and ship sizes. The implementation emphasizes accessibility, allowing deployment on devices without GPU support, and includes a clean, responsive web interface built with Streamlit for intuitive interaction. Experimental results demonstrate reliable detection capabilities with acceptable precision-recall balance for practical applications. The integration of a confidence slider and user-focused design elements marks a significant improvement in usability for non-technical stakeholders, making the system suitable for both research and field deployments.