PolSAR Image Classification with Lightweight 3D Convolutional Networks

被引:40
作者
Dong, Hongwei [1 ]
Zhang, Lamei [1 ]
Zou, Bin [1 ]
机构
[1] Harbin Inst Technol, Dept Informat Engn, Harbin 150001, Peoples R China
基金
中国国家自然科学基金;
关键词
deep learning; polarimetric synthetic aperture radar (PolSAR) classification; 3D convolution; pseudo-3D convolution; depthwise separable convolution;
D O I
10.3390/rs12030396
中图分类号
X [环境科学、安全科学];
学科分类号
08 ; 0830 ;
摘要
Convolutional neural networks (CNNs) have become the state-of-the-art in optical image processing. Recently, CNNs have been used in polarimetric synthetic aperture radar (PolSAR) image classification and obtained promising results. Unlike optical images, the unique phase information of PolSAR data expresses the structure information of objects. This special data representation makes 3D convolution which explicitly modeling the relationship between polarimetric channels perform better in the task of PolSAR image classification. However, the development of deep 3D-CNNs will cause a huge number of model parameters and expensive computational costs, which not only leads to the decrease of the interpretation speed during testing, but also greatly increases the risk of over-fitting. To alleviate this problem, a lightweight 3D-CNN framework that compresses 3D-CNNs from two aspects is proposed in this paper. Lightweight convolution operations, i.e., pseudo-3D and 3D-depthwise separable convolutions, are considered as low-latency replacements for vanilla 3D convolution. Further, fully connected layers are replaced by global average pooling to reduce the number of model parameters so as to save the memory. Under the specific classification task, the proposed methods can reduce up to 69.83% of the model parameters in convolution layers of the 3D-CNN as well as almost all the model parameters in fully connected layers, which ensures the fast PolSAR interpretation. Experiments on three PolSAR benchmark datasets, i.e., AIRSAR Flevoland, ESAR Oberpfaffenhofen, EMISAR Foulum, show that the proposed lightweight architectures can not only maintain but also slightly improve the accuracy under various criteria.
引用
收藏
页数:21
相关论文
共 48 条
[1]  
[Anonymous], IEEE T PATTERN ANAL, DOI DOI 10.1109/ICCICCT.2016.7987947
[2]  
[Anonymous], P 2 INT WORKSH POLIN
[3]  
[Anonymous], 2014, ARXIV14062199
[4]  
[Anonymous], ADV NEURAL INFORM PR
[5]   An Active Deep Learning Approach for Minimally Supervised PolSAR Image Classification [J].
Bi, Haixia ;
Xu, Feng ;
Wei, Zhiqiang ;
Xue, Yong ;
Xu, Zongben .
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, 2019, 57 (11) :9378-9395
[6]   A Graph-Based Semisupervised Deep Learning Model for PolSAR Image Classification [J].
Bi, Haixia ;
Sun, Jian ;
Xu, Zongben .
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, 2019, 57 (04) :2116-2132
[7]  
Brevdo E., 2016, TENSOR
[8]   Pixel-Wise PolSAR Image Classification via a Novel Complex-Valued Deep Fully Convolutional Network [J].
Cao, Yice ;
Wu, Yan ;
Zhang, Peng ;
Liang, Wenkai ;
Li, Ming .
REMOTE SENSING, 2019, 11 (22)
[9]   Slim and Efficient Neural Network Design for Resource-Constrained SAR Target Recognition [J].
Chen, Hongyi ;
Zhang, Fan ;
Tang, Bo ;
Yin, Qiang ;
Sun, Xian .
REMOTE SENSING, 2018, 10 (10)
[10]   PolSAR Image Classification Using Polarimetric-Feature-Driven Deep Convolutional Neural Network [J].
Chen, Si-Wei ;
Tao, Chen-Song .
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, 2018, 15 (04) :627-631