| 一种改进深度卷积神经网络的海岛识别方法 |
| |
| 引用本文： | 王振华,曲念毅,钟元芾,何婉雯,宋巍,黄冬梅.一种改进深度卷积神经网络的海岛识别方法[J].上海海洋大学学报,2020,29(3):474-480. |
| |
| 作者姓名： | 王振华 曲念毅 钟元芾 何婉雯 宋巍 黄冬梅 |
| |
| 作者单位： | 上海海洋大学 信息学院,上海 201306;上海电力大学,上海 200090 |
| |
| 基金项目： | 国家自然科学基金资助项目 Nos.41501419,41671431 |
| |
| 摘 要： | 受不规律潮汐的影响,现有的海岛地物类别自动识别方法存在精度低和时效性差等问题,通过改进深度卷积神经网络提出了一种基于遥感影像的海岛快速识别方法:(1)在深度卷积神经网络的卷积层中增设1×1的卷积核作为瓶颈单元,对多波段的遥感影像进行降维;(2)在池化层引入了重采样方法,基于灰度值对海量的遥感影像进行特征压缩。以300景Landsat-8遥感影像为源数据,分别采用CNN、RCNN和本文改进的深度卷积神经网络对遥感影像中的海岛进行识别,实验结果表明:(1)改进的深度卷积神经网络降低了海岛识别的计算耗时,其计算耗时仅为CNN的4.56%和RCNN的5.60%;(2)改进的深度卷积神经网络较CNN和RCNN提高了海岛识别的精度,识别精度分别为96.0%、93.3%和95.0%。结果说明,改进的深度卷积神经网络适用于面向遥感影像的海岛自动识别。
|
| 关 键 词： | 深度卷积神经网络 遥感影像 海岛识别 卷积运算 |
| 收稿时间： | 2019/4/1 0:00:00 |
| 修稿时间： | 2019/9/16 0:00:00 |
A method for identification of island by improving deep convolutional neural network |
| |
| WANG Zhenhu,QU Nianyi,ZHONG Yuanfu,HE Wanwen,SONG Wei,HUANG Dongmei.A method for identification of island by improving deep convolutional neural network[J].Journal of Shanghai Ocean University,2020,29(3):474-480. |
| |
| Authors: | WANG Zhenhu QU Nianyi ZHONG Yuanfu HE Wanwen SONG Wei HUANG Dongmei |
| |
| Institution: | College of Information Science,Shanghai Ocean University,Shanghai,College of Information Science,Shanghai Ocean University,Shanghai,College of Information Science,Shanghai Ocean University,Shanghai,College of Information Science,Shanghai Ocean University,Shanghai,College of Information Science,Shanghai Ocean University,Shanghai,Shanghai University of Electric Power,Shanghai |
| |
| Abstract: | Remote sensing technology has been widely applied in island identification in recent years, but the automatic identification method for island identification has several problems, such as low precision and poor timeliness. Because of these problems, this paper proposed a method for rapid identification of island by improving deep convolutional neural network (DCNN). The improved method contains two aspects. Firstly, adding a 1×1 convolution kernel as the bottleneck unit in the convolutional layer, it reduced the dimension of remote sensing images. Secondly, a resampling method has introduced in the pooling layer to perform feature compression on the target features. Taking 300 scenes of Landsat-8 remote sensing image as an example data, the improved method was compared with CNN model and RCNN model by identifying the islands, the results showed that the improved method reduced the computational time of island identification and improved the accuracy of island identification. Based on the experimental results, the model in this paper is more suitable for automatic island identification of remote sensing images. |
| |
| Keywords: | deep convolutional neural network remote sensing image island identification convolution operation |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
| 点击此处可从《上海海洋大学学报》浏览原始摘要信息 |
| 点击此处可从《上海海洋大学学报》下载免费的PDF全文 |
|
