Vessel Visual Surveillance Based on YOLOv8 Architecture: Daytime and Night Comparison

This paper is conducted by researchers focusing on the field of vessel visual surveillance using deep learning techniques. They aim to develop an effective surveillance system to enhance safety and security in the maritime environment. This study investigates the performance of the YOLOv8 architecture in detecting vessels during daytime and night. YOLOv8 is a powerful and efficient object detection algorithm that has shown promising results in various applications. The researchers compare the performance of the model trained on a daytime dataset when applied to night images, to assess its generalization ability under different lighting conditions. This paper focuses on vessel surveillance in maritime environments, including ports, shipping lanes, and other water areas. The ability to accurately detect and track vessels under varying lighting conditions is crucial for efficient traffic management and incident prevention. This study is conducted to address the challenges of round-the-clock vessel surveillance, both during the day and at night. Varying lighting conditions can significantly affect the performance of object detection algorithms, making it essential to evaluate and optimize models for 24/7 operation. Reliable visual vessel surveillance is critical for various applications, including maritime traffic management, collision prevention, illegal activity detection, and emergency response. By developing robust surveillance systems based on deep learning technologies like YOLOv8, researchers aim to enhance the safety, security, and efficiency of maritime operations. The YOLOv8 model is trained using an extensive dataset of daytime vessel images. The trained model is then evaluated on night images to assess its detection performance under different lighting conditions. The results show that while the model achieves high accuracy on daytime data, its performance deteriorates on night images, with some cases of false positives and false negatives. To optimize the model, the researchers suggest data augmentation techniques such as coloring the background to mimic night conditions or using a combined dataset of daytime and night images. This study emphasizes the need for further research and the development of more diverse datasets to improve vessel surveillance performance under various lighting conditions.