| 直线特征辅助的靠岸舰船检测 |
| |
| 引用本文： | 余东行,郭海涛,赵传,李道纪,张保明.直线特征辅助的靠岸舰船检测[J].测绘科学技术学报,2019,36(3):275-280,286. |
| |
| 作者姓名： | 余东行 郭海涛 赵传 李道纪 张保明 |
| |
| 作者单位： | 信息工程大学,河南 郑州,450001 |
| |
| 摘 要： | 针对靠岸舰船难以检测的问题,提出了一种直线特征辅助的靠岸舰船检测方法。首先利用高精度的卷积神经网络目标检测算法YOLOv3对影像进行粗检测,获取可能存在舰船目标的区域作为兴趣区域;然后提取影像的直线特征,将直线的方向作为确定舰船方向的辅助信息;最后利用具有一定角度的滑动窗口遍历兴趣区域获取候选目标,并对侯选目标进行二次分类和识别得到最终检测结果。利用不同港口的遥感影像进行实验的结果表明,提出方法能够有效检测港口内多种方向和并列停靠的舰船目标。
|
| 关 键 词： | 靠岸舰船检测 直线检测 YOLOv3算法 卷积神经网络 遥感影像 |
Detection of Ship Docked in Harbor Assisted with Line Feature |
| |
| YU Donghang,GUO Haitao,ZHAO Chuan,LI Daoji,ZHANG Baoming.Detection of Ship Docked in Harbor Assisted with Line Feature[J].Journal of Zhengzhou Institute of Surveying and Mapping,2019,36(3):275-280,286. |
| |
| Authors: | YU Donghang GUO Haitao ZHAO Chuan LI Daoji ZHANG Baoming |
| |
| Institution: | (Information Engineering University,Zhengzhou 450001,China) |
| |
| Abstract: | Aiming at the problem that the docked ship is difficult to be detected, a method of inshore ship detection assisted with line information is proposed. Firstly, YOLOv3 which is an object detection algorithm with high precision using convolutional neural network is adopted to search the image to obtain the regions of interest where ships may exist. And then the straight lines of the image are detected to calculate angles which could help obtain the true orientations of inshore ships. Finally, sliding windows with a certain angle is used to search regions of interest to get proposal targets which will be reclassified by another convolutional neural network. The experimental results on different harbors show that the proposed method can effectively detect the ships in different orientations and ships docked in parallel. |
| |
| Keywords: | inshore ship detection line detection YOLOv3 algorithm convolutional neural network remote sensing image |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
|
