基于深度学习算法的极化合成孔径雷达通用分类器设计

General Purpose PolSAR Classifier Based on Deep Learning Algorithm

地物分类是PolSAR(极化合成孔径雷达)的重要应用方向。传统算法需要基于特定数据人工选取特征和设计分类器,而深度学习算法能够自行从海量数据中提取层次化特征。在深度学习算法总结的基础上,结合深度学习和PolSAR大数据,提出了一种高效率、高精度的通用分类器设计方法。使用人工标记的数据训练CNN(深度卷积网络),自动化地进行特征学习和提取,并实现高精度的地物自动分类。在具有不同分辨率的机载和星载PolSAR数据上对通用分类器进行测试,都能快速、准确地分类。研究成果可快速将PolSAR数据转译为更直观的地物分类结果,对海量数据,特别是GF-3卫星PolSAR图像的利用有一定的辅助价值。

Terrain classification is one of the most important applications of polarimetric synthetic aperture radar（PolSAR）data.The classic algorithms are limited by manual designed features and classifiers.However,deep learning can extract hierarchical features from big data.Open literatures of deep-learning based PolSAR data classification approaches are firstly reviewed,and one general purpose PolSAR image classifier is then presented based on deep learning and PolSAR big data.Manually labelled data are used for training,and experiments are carried out on both airborne and space-borne SAR data with variant resolution.The results show that the proposed classifier is highly accurate and efficient,which is helpful for big data utilization,especially for GF-3 PolSAR data.