In maritime traffic volume surveys, ship type identification for large vessels can be conducted regardless of day or night using the Automatic Identification System (AIS), which is mandatory under SOLAS. However, many small vessels operating in coastal waters are not equipped with AIS, making nighttime ship type identification labor-intensive and highly dependent on the subjective judgment of observers.
This study investigates the feasibility of ship type identification from nighttime ship images using artificial intelligence (AI) to improve the efficiency and reliability of coastal traffic surveys. Nighttime ship images were acquired using a highly sensitive camera, and suitable exposure conditions were examined to enable extraction of ship outlines under low-visibility environments. A YOLOX-based object detection model was then trained to classify ship types. Experimental results indicate that ship type identification can be achieved with an average precision exceeding 90% for most ship categories, even when using original dark nighttime images without image enhancement, demonstrating the practicality of AI-based nighttime image analysis for automated ship type identification.
