基于散射图卷积网络的PolSAR影像地物分类

Scattering graph convolutional network-based PolSAR image classification

本文针对极化合成孔径雷达(polarimetric synthetic aperture radar,PolSAR)图像解译中特征提取不足与目标分类困难的问题,进行了深入的研究,提出了一种基于散射图卷积网络的PolSAR影像分类方法.在特征提取方面,本文给出了极化散射编码的一维表现形式;同时,考虑目标散射的特性和像素间的复杂关系,结合图论理论,提出了一种新的散射机制的图表示模型,来刻画复杂的极化散射机理;最后,将这种新的散射建模方法和图卷积网络结合,提出了PolSAR图像分类的新方法,从而更加高效、高精度地完成分类任务.实验结果表明,在5幅公开的PolSAR图像上(PolSF数据集),本文提出的算法具有良好的分类性能.

In this paper, we address the problem of insufficient feature extraction and the difficulty in interpreting the polarimetric synthetic aperture radar(PolSAR) images with all-day and all-weather capability and propose a novel scattering graph convolution network for PolSAR image classification. In terms of feature extraction, this paper presents a one-dimensional representation of polarimetric scattering coding. Simultaneously,considering the target scattering characteristics and the complex relationship among pixels, combined with the graph theory, a novel graph representation model of scattering mechanism is proposed to describe the complex polarization scattering mechanism. Finally, a polarimetric SAR image classification model based on the scatter graph convolution network is proposed to improve the decoding and classification of the PolSAR data. The experimental results regarding the five images of the PolSF dataset demonstrate that the proposed algorithm exhibits superior performance.