SDRNet: A hybrid approach with deep convolutional networks for ship draft reading

作     者：Wang, Bangping Liu, Zhiming Shen, Yantao Wang, Siming 

作者机构：Chengdu Univ Informat Technol Sch Software Engn Chengdu 610103 Sichuan Peoples R China Aetrex Teaneck NJ 07666 USA Univ Nevada Dept Elect & Biomed Engn Reno NV 89557 USA Chengdu Neusoft Univ Sch Intelligent Sci & Engn Chengdu 610225 Peoples R China 

出 版 物：《MEASUREMENT》 (Meas J Int Meas Confed)

年 卷 期：2025年第247卷

核心收录：

学科分类：08[工学] 080401[工学-精密仪器及机械] 0804[工学-仪器科学与技术] 081102[工学-检测技术与自动化装置] 0811[工学-控制科学与工程] 

基　　金：Sichuan Science and Technology Program of China [2024YFFK0413] 

主　　题：Ship draft reading Computer vision Deep learning Keypoint detection Scene text recognition Semantic segmentation 

摘      要：Ship draft measurement is essential in the maritime freight transport industry, but traditional methods relying on human surveyors are prone to errors. With the rise of deep learning and computer vision, there is significant potential to improve both accuracy and efficiency in this process. This paper presents SDRNet, a novel approach leveraging deep convolutional networks for automated ship draft reading. SDRNet integrates a multi-task vision network, combining a keypoint detection sub-network for draft mark identification and a semantic segmentation sub-network for waterline extraction. Additionally, a scene text recognition network is employed to extract integer values from draft marks. Extensive experiments show that SDRNet achieves performance comparable to human surveyors, offering a promising solution to enhance the accuracy and reliability of ship draft measurement in the maritime industry.