Hyperspectral image classification has attracted considerable interest in recent years. The previous classification methods are usually based on single-kernel or composite-kernel machines. In this paper, a novel regularization framework referred to least-squares support vector machine in sum space (LS-SVM-SS) of reproducing kernel Hilbert space (RKHS) is proposed for the classification of hyperspectral images using spectral signatures or local binary pattern features. The method is designed to simultaneously approximate the low- and high-frequency components of the target classification function with multiscale kernels. In the newly proposed scheme, LS-SVM-SS carries out the supervised learning and train a closed-form discriminant function to directly implement the multiclass classification. In contrast to multiple-kernel learning (MKL) by one-step method or two-step method, we can obtain a noniterative optimization procedure. Experiments are conducted on three real hyperspectral datasets. The corresponding experimental results demonstrate that the LS-SVM-SS method achieves good generalization performance in most cases compared with several state-of-the-art methods, which is better than that in any single-kernel RKHS. In addition, the proposed framework of multiscale kernels classifier opens a wide field for future developments in hyperspectral image classification.

In this article, we propose a total variation (TV) regularization approach for the reconstruction of super-resolution synthetic aperture radar (SAR) image based on gradient profile prior or other texture image prior in the maximum a posteriori framework. We also design a novel super-resolution reconstruction algorithm via split Bregman iteration with the known degradation matrix, thereby enhancing the resolution of the SAR image. The parameter adaptation of the TV regularization is performed based on the high-resolution (HR) SAR image at each step. Several evaluation indices are tested on SAR images for objective assessment of the performance of SAR image super-resolution reconstruction. This computationally efficient algorithm is robust to noise in SAR scenes in HR image estimation. Experimental results show that the proposed split Bregman super-resolution approach can effectively avoid the speckle noise generated due to some strange textures and has good effect of noise suppression, while effectively maintaining the SAR image content, the structure of the SAR image is more apparent. Additionally, the experimental results on real SAR scenes also demonstrate the effectiveness of the proposed algorithm and demonstrate its superiority to other super-resolution algorithms.

Recently, the semisupervised learning for hyperspectral image classification (HIC) has attracted increasing attention. The previous semisupervised methods were usually based on reproducing kernel Hilbert spaces with a single kernel. In this paper, we propose a novel Laplacian least squares support vector machine in sum space (Lap-LSSVM-SS) for semisupervised HIC with multiscale kernels, which can simultaneously deal with the high and low frequency components of the target classification function with small and large scale kernels, respectively. Furthermore, due to its representation theorem, a noniterative algorithm based on the closed-form classification function of Lap-LSSVM-SS is provided to make a rapid semisupervised HIC possible in the multiple kernel learning framework. The proposed method can solve the multiclass problem of hyperspectral dataset directly. We further reduce the storage and computation cost of the proposed Lap-LSSVM-SS method with the random sampling strategy, i.e., Lap-LSSVM-SS-RS. We also show that the Lap-LSSVM-SS-RS method has near-optimal classification performance when the number of local Lap-LSSVM-SS machines is not too large. The experimental results on two real hyperspectral datasets demonstrate that Lap-LSSVM-SS-RS yields superior classification performance over the conventional semisupervised classifiers (e.g., LapSVM, S4VM, and BagSVMs, etc.) in challenging small training labeled samples and a large amount of unlabeled samples.

PolSAR image classification has attracted extensive significant research in recent decades. Aiming at improving PolSAR classification performance with speckle noise, this paper proposes an active complex-valued convolutional-wavelet neural network by incorporating dual-tree complex wavelet transform (DT-CWT) and Markov random field (MRF). In this approach, DT-CWT is introduced into the complex-valued convolutional neural network to suppress the speckle noise of PolSAR images and maintain the structures of learned feature maps. In addition, by applying active learning (AL), we iteratively select the most informative unlabeled training samples of PolSAR datasets. Moreover, MRF is utilized to obtain spatial local correlation information, which has been proven to be effective in improving classification performance. The experimental results on three benchmark PolSAR datasets demonstrate that the proposed method can achieve a significant classification performance gain in terms of its effectiveness and robustness beyond some state-of-the-art deep learning methods.

BingSat-Tomographic Observation of Polar Ice Sheets (BingSat-TOPIS) is a spaceborne multistatic radar sounding system, which can achieve high resolution and stereoscopic observation. It is designed to penetrate ice sheets and acquire tomographic image. The satellite group fly over the polar regions, which can form about 6.56 km cross-track baseline. This cross-track baseline is formed by one master satellite and forty slave CubeSats. In this paper, we propose a tomographic imaging algorithm for orbital radar sounding of the Earth's ice sheets. First, we give the method to calculate the two-way slant range in air and ice. The phase errors of the method are smaller than π/4 rad. Second, we express the radar data cube for a point target in ice sheets. The radar data cube is made of forty bistatic synthetic aperture radar (SAR) echo signals. At last, the echoes are simulated for TOPIS system, and a matched filter is used for pulse compression in range. Then, back projection (BP) algorithm is used in track and cross-track imaging. The experimental results demonstrate that our tomographic imaging algorithm is effective and reliable.