| 講演抄録/キーワード |
| 講演名 |
2018-11-08 14:50
A Novel Method Using Convolutional Neural Network for Polarimetric SAR ship detection ○Kan Jin・Junjun Yin・Jian Yang(Tsinghua Univ.) SANE2018-60 |
| 抄録 |
(和) |
This paper presents a novel approach for ship detection in polarimetric synthetic aperture radar (POLSAR) image with the help of convolutional neural network(CNN). Unlike optical images, lower resolution in POLSAR image leads directly to the result that ship especially the small one has fewer pixels than those in optical images. So architectures which pay more attention on the ship itself cannot do well in this work. The proposed convolutional neural network not only pay attention to the ship itself but also pay attention to its surroundings. A patch of sufficient size which contains both ship and surroundings is used as the input of the neural network to determine whether the pixel in the center of the patch is ship or not. In order to use contextual semantic information at all levels, the center pixels of all feature maps from top to down are added to get a fused feature. The proposed approach is tested on POLSAR images. The results show that the proposed approach can achieve a better performance than previous work on this problem. |
| (英) |
This paper presents a novel approach for ship detection in polarimetric synthetic aperture radar (POLSAR) image with the help of convolutional neural network(CNN). Unlike optical images, lower resolution in POLSAR image leads directly to the result that ship especially the small one has fewer pixels than those in optical images. So architectures which pay more attention on the ship itself cannot do well in this work. The proposed convolutional neural network not only pay attention to the ship itself but also pay attention to its surroundings. A patch of sufficient size which contains both ship and surroundings is used as the input of the neural network to determine whether the pixel in the center of the patch is ship or not. In order to use contextual semantic information at all levels, the center pixels of all feature maps from top to down are added to get a fused feature. The proposed approach is tested on POLSAR images. The results show that the proposed approach can achieve a better performance than previous work on this problem. |
| キーワード |
(和) |
ship detection / convolutional neural network / polarimetric synthetic aperture radar(POLSAR) / / / / / |
| (英) |
ship detection / convolutional neural network / polarimetric synthetic aperture radar(POLSAR) / / / / / |
| 文献情報 |
信学技報, vol. 118, no. 287, SANE2018-60, pp. 27-30, 2018年11月. |
| 資料番号 |
SANE2018-60 |
| 発行日 |
2018-11-01 (SANE) |
| ISSN |
Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
| PDFダウンロード |
SANE2018-60 |