Sparse Robust Filters for scene classification of Synthetic Aperture Radar (SAR) images

Abstract

With the increasing resolution of Synthetic Aperture Radar (SAR) images, extracting their discriminative features for scenes classification has become a challenging task, because SAR images are very sensitive to target aspect brought by shadowing effects, interaction of the signature with the environment, and so on. Moreover, SAR images are remarkably polluted by the multiplicative speckle noise, which makes the conventional feature extractors inefficient. In this paper we advance new Sparse Robust Filters (SRFs) for automatic learning of discriminant features of scenes. A Hierarchical Group Sparse Coding (HGSC) model is proposed to learn a set of sparse and robust filters, to capture the multiscale local descriptors that are robust to noises. Some experiments are taken on a TerraSAR-X images dataset (in the middle of the Swabian Jura, the Nördlinger Ries, HH, observed on July, 2007), and a Moving and Stationary Target Acquisition and Recognition (MSTAR) dataset, to evaluate the performance of our proposed method. The experimental results show that our method can achieve higher classification accuracy compared with other related approaches.

Compact Global-Local Convolutional Network With Multifeature Fusion and Learning for Scene Classification in Synthetic Aperture Radar Imagery

“…Currently, deep neural networks (DNNs) have proved their considerably strong power in the field of remote sensing image processing community, especially in case of SAR image classification. Several developed DNNs, including deep recurrent encoding neural networks [20], deep sparse tensor filtering network [21], discriminant deep belief network [22], patch-sorted deep neural network [23], and deep convolutional autoencoder [24], have performed well in SAR image classification. CNN, which is one of the DNNs, automatically learns high-level feature statistics via the back-propagation algorithm and has further boosted the representation power of deep feature statistics, i.e., VGGNet [25], ResNet [26], and DenseNet [27].…”

Fusing Deep Learning and Sparse Coding for SAR ATR

“…Feature-based solutions may rely on Krawtchouk moments [20] where features are derived from the discrete-defined Krawtchouk polynomials or on biologically inspired features. The latter can rely on episodic and semantic features [21] or sparse robust filters [22] that originate from the human cognition process. Other methods include binary operations [23], using the target's scattering centers [15], [24] or the azimuth and range target profiles fusion [25].…”

Scene Classification from Synthetic Aperture Radar Images Using Generalized Compact Channel-Boosted High-Order Orderless Pooling Network

“…SAR image scene classification, which is a direct understanding and interpretation of SAR data [6,7], has become an important technique in the extraction of the ground objects, target recognition and so on [8]. The description of SAR images is more difficult than high-resolution optical remote sensing sensing images.…”