基于Faster R-CNN的电子海图和雷达图像的数据融合  

Data fusion of electronic navigational chart and radar images based on Faster R-CNN

作　　者：张大恒[1,2] 张英俊 张闯[1] ZHANG Daheng;ZHANG Yingjun;ZHANG Chuang(Navigation College,Dalian Maritime University,Dalian 116026,China;School of Navigation and Naval Architecture,Dalian Ocean University,Dalian 116023,China)

机构地区：[1]大连海事大学航海学院,辽宁大连116026 [2]大连海洋大学航海与船舶工程学院,辽宁大连116023

出　　处：《系统工程与电子技术》2020年第6期1267-1273,共7页Systems Engineering and Electronics

基　　金：国家自然科学基金(51679025)资助课题。

摘　　要：船载导航雷达和电子海图(electronic navigational chart,ENC)是船舶重要的导航仪器,雷达图像和ENC图像的融合能够给出更加丰富的航行和避碰信息。为此,提出了一种基于深度学习理论的提取雷达图像中鲁棒特征的数据融合算法,实现了ENC和雷达图像较高层次的数据融合。首先,利用深度学习算法对雷达图像进行目标检测,识别船舶雷达的特征目标。其次,对检测到的特征区域执行图像处理,并确定用于ENC和船舶雷达图像配准的参考点。最后,根据参考点进行仿射变换,实现融合算法。利用连续时间段的狭窄水域中的真实船舶雷达图像数据对融合算法进行验证,结果显示船舶雷达图像和ENC的海岸线边缘信息匹配良好且满足实时性要求。该算法与简单的像素级图像融合算法相比鲁棒性更强,实现了ENC与雷达图像的特征级融合。Marine narigation radar and electronic navigational chart(ENC)are important navigation instruments.The fusion of radar images and electronic navigational chart data can give more abundant navigation and collision avoidance information.Therefore,this paper proposes a data fusion algorithm based on deep learning to extract robust features from radar images.At first the ability of deep learning is exploited to perform target detection for the identification of marine radar targets.Then,the image processing is performed on the identified targets to determine the reference points for consistent data fusion of ENC and marine radar information.Finally,an affine transformation-fusion algorithm is built to merge the marine radar and electronic chart data according to the determined reference points.The proposed fusion algorithn is verified through simulations using ENC data and marine radar images from real ships in narrow waters in a continuous period.The results show a suitable edge matching performance of the shoreline and real-time applicability.The algorithm is more robust than the simple pixel-level image fusion,and it realizes the feature level fusion of ENC and radar images.

关 键 词：深度学习 船舶雷达图像 电子海图 数据融合 

分 类 号：TN957.51[电子电信—信号与信息处理]