Pingping Huang

Saliency detection based on distance between patches in polarimetric SAR images

In this paper, we propose a saliency detection model for polarimetric SAR images based on inter-patch distances. The model is biology-based as it takes the human visual properties into account. Our model consists of local and global saliency detection, which obtained by multi-scale information extraction. Whats more, inspired by the distance measures of different image patches, the model takes full use of the coherency matrix which includes the statistical information of pixels and precisely measures the similarity between patches. The experimental results demonstrate the effectiveness of our method.

Change Detection Using High Resolution Remote Sensing Images Based on Active Learning and Markov Random Fields

Change detection has been widely used in remote sensing, such as for disaster assessment and urban expansion detection. Although it is convenient to use unsupervised methods to detect changes from multi-temporal images, the results could be further improved. In supervised methods, heavy data labelling tasks are needed, and the sample annotation process with real categories is tedious and costly. To relieve the burden of labelling and to obtain satisfactory results, we propose an interactive change detection framework based on active learning and Markov random field (MRF). More specifically, a limited number of representative objects are found in an unsupervised way at the beginning. Then, the very limited samples are labelled as “change” or “no change” to train a simple binary classification model, i.e., a Gaussian process model. By using this model, we then select and label the most informative samples by “the easiest” sample selection strategy to update the former weak classification model until the detection results do not change notably. Finally, the maximum a posteriori (MAP) change detection is efficiently computed via the min-cut-based integer optimization algorithm. The time consuming and laborious manual labelling process can be reduced substantially, and a desirable detection result can be obtained. The experiments on several WorldView-2 images demonstrate the effectiveness of the proposed method.

Unsupervised classification of PolSAR data using large scale spectral clustering

In this paper, a spectral clustering based unsupervised classification scheme is proposed for processing large scale polarimetric synthetic aperture radar (PolSAR) data. Due to its high computational complexity, spectral clustering can hardly handle large PolSAR image. To overcome this bottleneck, a representative points based scheme is introduced. Instead of building pairwise affinity graph on the whole data set, we first build a bipartite graph between data points and a small set of selected representative points. Then an approximate large graph is constructed based on this bipartite graph. After that, spectral analysis on the approximate graph is solved efficiently by singular value decomposition (SVD). To integral context information, Markov random fields (MRF) model based smoothing is also performed to get the final clusters. We test the proposed approach on DLR ESAR data set. Experimental results demonstrate its effectiveness and efficiency.

Investigation of Azimuth Multichannel Reconstruction for Moving Targets in High Resolution Wide Swath SAR

The azimuth multichannel imaging scheme with the large receive antenna divided into multiple sub-apertures usually leads to azimuth non-uniform sampling, and echoes from all azimuth channels should be reconstructed based on the signal model before conventional SAR imaging. Unfortunately, the multichannel signal model of a moving target is different from that of a fixed target. This paper analyzes the multichannel signal model of the moving target and the effect of the target velocity on azimuth multichannel reconstruction. Based on the multichannel signal mode of the moving target, a new multichannel signal reconstruction algorithm is proposed. Furthermore, the slant range velocity is estimated by computing signal energy distribution. Simulation results on point targets validate the proposed approach.

Feature based decision methodology for vegetation classification

PolSAR features have great significance in application of vegetation classification, which can explain the scattering mechanism of the vegetation; the decision tree classifier not only can obtain good classification accuracy, but also can adjust the classification results, as well as make full use of PolSAR features to explain the scattering mechanism of the targets because of its simple and hierarchical classifier structure. Since all the classification methods are composed of two parts: feature selection and classifier selection, this paper established a classification method with PolSAR features as selected feature and decision tree as adopted classifier. As decision tree classifier is flexible in discriminant rules, the expected design of the experimental scheme introduces multiple data sources, multiple features and multiple classifiers into the framework of this classification method. In addition, discussion about how to improve the classification accuracy of the specific target has been made. The experiment of AIRSAR-Flevoland data illustrates the feasibility of this method.

Detecting changes in high resolution remote sensing images using superpixels

In this paper, in order to detect changes in high resolution remote sensing images, we propose an MRF-based change detection method combined with the semantic information. Two temporal high resolution remote sensing images are represented by features of superpixels. For given images, we transform the change detection problem into a binary classification problem by combining differences in both low-level features and semantic information in MRF smoothing framework. All pixels are divided into two categories: changed or unchanged, so we can extract change information from classification result. Experimental results of two Geo-Eye1 high-resolution remote sensing images at different time demonstrate the efficiency of this proposed method. Detection combined with semantic information can significantly improve the result than only with low-level features. Adding Markov smoothing can also improve the detection results slightly.

Estimation and analysis soil moisture of hunshandake sandy land from polarimetric SAR data

At present, various models are developed for soil moisture retrieval, but the application of polarimetric SAR data to retrieve soil moisture in sandy land is rare. Therefore, it is necessary to develop a method for retrieving soil moisture in sandy land. In this paper, we proposed a model to estimate the soil moisture in Hunshandake Sandy Land. The model does not need to take into account the surface roughness, only using the VV and HH polarization backscattering coefficients can be used to retrieve soil moisture. In addition, diversity of the vegetation over the sandy land is difficult to predict. We selected five samples to analyze the effect of vegetation on the inversion results of soil moisture in sandy land.

Investigation of Wavenumber Domain Imaging Algorithm for Ground-Based Arc Array SAR

Ground-based synthetic aperture radar (GB-SAR) has become an important technique for remote sensing deformation monitoring. However, most of the existing GB-SAR systems realize synthetic aperture by exploiting two closely spaced horn antennas to move along a linear rail. In order to obtain higher data acquisition efficiency and a wider view angle, we introduce arc antenna array technology into the GB-SAR system, which realizes a novel kind of system: ground-based arc array SAR (GB-AA-SAR). In this paper, we analyze arc observation geometry and derive analytic expressions of sampling criteria. Then, we propose a novel wavenumber domain imaging algorithm for GB-AA-SAR, which can achieve high image reconstruction precision through numerical solutions in the wavenumber domain. The proposed algorithm can be applied in wide azimuth view angle scenarios, and the problem of azimuth mismatch caused by distance approximation in arc geometric efficient omega-k imaging can be solved successfully. Finally, we analyze the two-dimensional (2D) spatial resolution of GB-AA-SAR, and verify the effectiveness of the proposed algorithm through numerical simulation experiments.

Three-Dimensional Microwave Imaging for Concealed Weapon Detection Using Range Stacking Technique

Three-dimensional (3D) microwave imaging has been proven to be well suited for concealed weapon detection application. For the 3D image reconstruction under two-dimensional (2D) planar aperture condition, most of current imaging algorithms focus on decomposing the 3D free space Green function by exploiting the stationary phase and, consequently, the accuracy of the final imagery is obtained at a sacrifice of computational complexity due to the need of interpolation. In this paper, from an alternative viewpoint, we propose a novel interpolation-free imaging algorithm based on wavefront reconstruction theory. The algorithm is an extension of the 2D range stacking algorithm (RSA) with the advantages of low computational cost and high precision. The algorithm uses different reference signal spectrums at different range bins and then forms the target functions at desired range bin by a concise coherent summation. Several practical issues such as the propagation loss compensation, wavefront reconstruction, and aliasing mitigating are also considered. The sampling criterion and the achievable resolutions for the proposed algorithm are also derived. Finally, the proposed method is validated through extensive computer simulations and real-field experiments. The results show that accurate 3D image can be generated at a very high speed by utilizing the proposed algorithm.

Simultaneous SAR imaging and GMTI by fractional Fourier transform processing

Simultaneous synthetic aperture radar (SAR) imaging and ground moving target indication (GMTI) is of great importance in remote sensing applications, but it is difficult to be implemented for existing methods due to the cross-interferences between stationary targets/clutter and moving targets. Generally, the imaged moving targets may be displaced in azimuth according to their radial velocities and superimposed upon clutter at a wrong location. In this paper, we proposes a simple simultaneous SAR and GMTI approach by exploiting the fractional Fourier transform (FrFT) algorithm, different from existing methods that perform first stationary clutter suppression and thereafter handle GMTI via Doppler parameters estimation. The feasibility is verified by simulation results.