Land cover classification of RADARSAT-2 SAR data using convolutional neural network

Abstract

In this paper, we propose a convolutional neural network (CNN) based on deep learning method for land cover classification of synthetic aperture radar (SAR) images. The proposed method consists of convolutional layers, pooling layers, a full connection layer and an output layer. The method acquires high-level abstractions for SAR data by using a hierarchical architecture composed of multiple non-linear transformations such as convolutions and poolings. The feature maps produced by convolutional layers are subsampled by pooling layers and then are converted into a feature vector by the full connection layer. The feature vector is then used by the output layer with softmax regression to perform land cover classification. The multi-layer method replaces hand-engineered features with backpropagation (BP) neural network algorithm for supervised feature learning, hierarchical feature extraction and land cover classification of SAR images. RADARSAT-2 ultra-fine beam high resolution HH-SAR images acquired in the rural urban fringe of the Greater Toronto Area (GTA) are selected for this study. The experiment results show that the accuracy of our classification method is about 90% which is higher than that of nearest neighbor (NN).