基于泛化深度迁移特征的高分遥感场景分类

doi:10.3969/j.issn.1001-506X.2018.03.30

系统工程与电子技术 ›› 2018, Vol. 40 ›› Issue (3): 682-691.doi: 10.3969/j.issn.1001-506X.2018.03.30

基于泛化深度迁移特征的高分遥感场景分类

罗畅, 王洁, 王世强, 史通, 任卫华

空军工程大学防空反导学院, 陕西西安 710051

出版日期:2018-02-26 发布日期:2018-02-26

General deep transfer features based high resolution remote scene classification

LUO Chang, WANG Jie, WANG Shiqiang, SHI Tong, REN Weihua

Air and Missile Defense College, Air Force Engineering University, Xi’an 710051, China

Online:2018-02-26 Published:2018-02-26

摘要/Abstract

摘要：

针对深度卷积神经网络(deep convolutional neural network, DCNN)迁移至高分辨率遥感场景分类的问题。设计了有效的网络结构用于增强DCNN在高分辨率遥感场景分类任务中的泛化能力。首先,线性主成分分析网络被用于整合高分辨率遥感图像的空间信息,减小DCNN在迁移过程中源数据集与目标数据集之间的空间差异。随后,经整合的图像输入预训练的DCNN,提取到更具泛化性能的全局特征表达。两个公开遥感数据集(UC Merced 21和WHU-RS 19)的试验结果表明,在不改变DCNN结构参数的情况下,相比现有方法,所设计的网络结构能够有效提升遥感场景分类精度。

Abstract:

Transferring deep convolutional neural network (DCNN) for high resolution remote scene classification is discussed. A promising architecture is proposed to enhance the generalization power of pre-trained DCNN for high resolution remote scene classification. Firstly, a linear principle component analysis network is designed to synthesize spatial information of high resolution remote sensing images. This design shortens the spatial distance between the target and source datasets for DCNN. Then pre-trained DCNN extract more general global features from the synthesized image. Experimental results of the two independent remote sensing datasets demonstrate that compared with state-of-the-art results, the proposed framework improves the accuracy of the remote scene classification without changing parameters in DCNN.

罗畅, 王洁, 王世强, 史通, 任卫华. 基于泛化深度迁移特征的高分遥感场景分类[J]. 系统工程与电子技术, 2018, 40(3): 682-691.

LUO Chang, WANG Jie, WANG Shiqiang, SHI Tong, REN Weihua. General deep transfer features based high resolution remote scene classification[J]. Systems Engineering and Electronics, 2018, 40(3): 682-691.

基于泛化深度迁移特征的高分遥感场景分类

General deep transfer features based high resolution remote scene classification

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 0

编辑推荐

Metrics

本文评价