A New Deep Learning Algorithm for SAR Scene Classification Based on Spatial Statistical Modeling and Features Re-Calibration.

Lifu Chen,Jin Xing,Zhihui Yuan,Xuemin Xing,Xianliang Cui,Zhenhong Li,Zhiwei Jia

doi:10.3390/s19112479

Abstract

Synthetic Aperture Radar (SAR) scene classification is challenging but widely applied, in which deep learning can play a pivotal role because of its hierarchical feature learning ability. In the paper, we propose a new scene classification framework, named Feature Recalibration Network with Multi-scale Spatial Features (FRN-MSF), to achieve high accuracy in SAR-based scene classification. First, a Multi-Scale Omnidirectional Gaussian Derivative Filter (MSOGDF) is constructed. Then, Multi-scale Spatial Features (MSF) of SAR scenes are generated by weighting MSOGDF, a Gray Level Gradient Co-occurrence Matrix (GLGCM) and Gabor transformation. These features were processed by the Feature Recalibration Network (FRN) to learn high-level features. In the network, the Depthwise Separable Convolution (DSC), Squeeze-and-Excitation (SE) Block and Convolution Neural Network (CNN) are integrated. Finally, these learned features will be classified by the Softmax function. Eleven types of SAR scenes obtained from four systems combining different bands and resolutions were trained and tested, and a mean accuracy of 98.18% was obtained. To validate the generality of FRN-MSF, five types of SAR scenes sampled from two additional large-scale Gaofen-3 and TerraSAR-X images were evaluated for classification. The mean accuracy of the five types reached 94.56%; while the mean accuracy for the same five types of the former tested 11 types of scene was 96%. The high accuracy indicates that the FRN-MSF is promising for SAR scene classification without losing generality.

Highlights

With the fast development of remote sensing technology, the variety of the acquired imagery datasets has been increasing, such as Hyperspectral images, Light Detection And Ranging (LiDAR)dense point clouds, and Synthetic Aperture Radar (SAR) images with different bands
Many scene classification approaches have been proposed, which can be categorized into three types according to the features extraction [16]: (1) methods based on handcrafted-feature, which usually use expertise and engineering skills to extract useful information to distinguish between different types of targets [17], such as texture, shape and spectral features; (2) methods based on unsupervised-feature-learning, which extract more discriminative features than manually designed features by learning from unlabeled input data automatically [18,19]; (3) methods based on deep-feature-learning, which extract high-level features of targets by learning from labeled input data [20,21], as proposed by Hinton and Salakhutdinov in 2006 [22]
The CAS-InSAR image was provided by the Chinese Academy of Science, and we selected an image taken at Weinan, China

Summary

Introduction

With the fast development of remote sensing technology, the variety of the acquired imagery datasets has been increasing, such as Hyperspectral images, Light Detection And Ranging (LiDAR)dense point clouds, and Synthetic Aperture Radar (SAR) images with different bands. As the number of the network layers increases, it can extract higher-level features and better interpret SAR images. It stems from an artificial neural network, which integrates low-level features to form abstract high-level features to determine the types of different targets. Hinton et al [23] presented a fast, greedy algorithm which used “complementary priors” to learn deep, Directed Belief Networks (DBN) one layer at a time. It achieves better digit classification result than the best discriminative learning algorithm, which brings some hope to solve the optimization problem of deep structures. In 2016, AlphaGo, an artificial intelligence robot developed by DeepMind, defeated the top player of mankind in the game

Methods

Results

Discussion

Conclusion