摘要
船舶六面水尺数值的准确读取是水尺计重过程中基础且极为关键的环节。为了排除传统人工测看方法中人为因素的干扰,提高水尺计重结果的准确性和客观性,本文基于Faster R-CNN的深度学习算法来识别确定船舶水尺字符,利用边框矫正算法精调水尺字符位置,利用RG双通道像素差分法和深度学习算法来识别确认吃水线,并以此确认六面水尺读数进而联合船舶其他测量参数得出最终的船舶载货量,开发出了LeonZX-IDSS智能水尺测定系统,该系统的测定读数精度可达0.001 m。结果表明,基于深度学习算法的智能水尺测定系统可以有效识别水尺影像并自动分析,给出水尺读数,与行业内依照现有标准而采用的人工读数相比,测定系统的智能读数准确率可达97%。
In the process of draft surveying,the accurate reading of water draft value on six sides of a ship is a basic and crucial link.In order to eliminate the interference of human factors in the traditional methods of artificial observation and improve the accuracy and objectivity of the results of draft survey,based on Faster R-CNN algorithm to identify and determine draft characters of ships,using border correction algorithm to accurately adjust the position of recognized characters,using RG double-channel pixel difference method and deep learning algorithm to identify and confirm the waterline,using the six-sided draft reading to obtain the final cargo weight of the ship combined with other measuring parameters,the Intelligent Draft Survey System(LeonZX-IDSS)is developed whose precision of measuring reading is 0.001 m in this paper.The results show that the LeonZX-IDSS based on deep learning algorithm can effectively identify the draft image and analyze the draft reading automatically.The intelligent reading accuracy of the system can reach about 97%compared with the manual reading used in the industry according to the existing standards.
作者
朱学海
罗陨飞
ZHU Xuehai;LUO Yunfei(Leon Intelligence&Information(Beijing)Technology Co.,Ltd.,Beijing,100102,China)
出处
《检验检疫学刊》
2020年第3期102-105,111,共5页
Journal of Inspection and Quarantine
关键词
水尺计重
深度学习
水尺字符
吃水线
读数对比
Draft Survey
Deep Learning
Draft Characters
Waterline
Reading Contrast
