Real-Time Ship Detection and Text Recognition Using YOLO-OCR for Smart Port Applications
Aminu Yahaya *
School of Computer Science and Technology, Zhejiang University of Science and Technology, Hangzhou, China.
Rui-Cai Jia
School of Computer Science and Technology, Zhejiang University of Science and Technology, Hangzhou, China.
Xingli Gan
School of Computer Science and Technology, Zhejiang University of Science and Technology, Hangzhou, China.
Chong Shen
School of Instrument and Electronics, North University of China, Taiyuan, China.
De-lin Zhao
CHN Energy Group Huanghua Port Co. Ltd, Huanghua, China.
*Author to whom correspondence should be addressed.
Abstract
This study presents a new real-time computer vision architecture designed for maritime environments that combines YOLO for object detection and PaddleOCR for text recognition. The YOLO algorithm was tuned to identify ships and text regions using a dataset of over 600 annotated photos. Two output layers with good detection accuracy (mAP 0.90, F1-score 0.89) were obtained by removing the smallest detection scale (P3) in order to speed up inference and lower computational complexity. making it appropriate for marine applications with bandwidth constraints. To enhance OCR ‘robustness in low-quality or variable lighting conditions, detected text regions undergo a lightweight preprocessing pipeline consisting of grayscale conversion, contrast enhancement, and noise reduction. The proposed framework enables automated and continuous ship monitoring, thereby supporting compliance verification, port logistics, and security operations in real seaport environments. Furthermore, the architecture demonstrates scalability toward large scale, real-time maritime surveillance systems.
Keywords: Port automation, object detection, text recognition, maritime surveillance, YOLOv5, PaddleOCR