Jubo Zhu

Superresolution imaging by dynamic single-pixel compressive sensing system

Abstract. We present a superresolution imaging method based on the dynamic single-pixel compressive sensing (CS) system. Different from the traditional static CS, this system is slowly moving in parallel with the scene during the compressive sampling, implying that the measurements are possible to contain the information about the scene with the subpixel resolution. Here we first build the dynamic compressive sampling model and give the recovery method via traditional CS scheme, and then we propose the image superresolution recovery method in the CS framework, where a subdivision scheme is used. The proposed method not only has remarkable superresolution performance, but also has low requirements on the imaging system, since it is associated with the single-pixel imager, which is one of the simplest systems in the existing CS imaging architectures. The feasibility of the proposed method is demonstrated by the numerical simulations as well as the optical experiments.

Direction-of-Arrival Tracking via Low-Rank Plus Sparse Matrix Decomposition

In this letter, a novel method is proposed to track the direction-of-arrival (DOA) of moving sources using a uniform linear array and is named as Tracking via Low-rank and Sparse Recovery (TvLSR). Based on the low-rank property of stationary sources and the sparsity of moving sources, TvLSR tracks the DOA of sources over all snapshots integrally via a convex optimization, instead of via a series of DOA estimation snapshot-by-snapshot as some existing method. This method can track DOA efficiently with several other merits: applicability to complex scenarios, including multiple intersecting trajectories, no requirement of numbers of moving sources and stationary sources, and robustness to noisy perturbation. Numerical simulations are carried out to demonstrate the excellent performance of TvLSR for DOA tracking.

Missile target automatic recognition from its decoys based on image time-series

As the development of the technology for radar target recognition, missile target automatic recognition has received considerable attention in recent years. Missile target, compared with the plane target, is hard to recognize for its smallness, feebleness and maneuver. In this paper, a new recognition method based on radar image time-series, which can significantly reduce the recognition time and classification error, is proposed. The image time-series are produced by range instantaneous Doppler imaging algorithm firstly, and then cross-range scaling of the images is processed. In particular, the inertia ratio, extracted from the obtained image time-series, is introduced to distinguish the missile from decoys. Furthermore, the effectiveness of this method is demonstrated by application to simulated data and it has been shown that this method has the potential to be used in a number of real-time applications.

Random phase modulation and sparse sampling-based optical imaging system

We describe an optical imaging system based on random phase modulation and sparse sampling. The system is also an application of compressive sensing in the field of imaging. Different from the Rice compressing imaging system, the imaging system needs only a signal exposure. Meanwhile, the space-bandwidth product of the system is extended, which means the pixels of imaging sensor can be reduced without a loss of images resolution. A random phase modulation is employed to make the energy of optical field spread out across the entire modulated image, which can facilitate the sampling process. A random sparse sampling scheme is designed to effectively reduce the pixels of imaging sensor, and the sensing matrix is uncorrelated with arbitrary sparse representation matrix. The feasibility of the proposed system is validated by numerical experiments.

Dehazing for single image with sky region via self-adaptive weighted least squares model

Abstract. The physical imaging model, which is based on atmospheric absorption and scattering, plays an important role in single-image dehazing. It is critical that the transmission is accurately estimated for the dehazing algorithm based on the physical imaging model. A self-adaptive weighted least squares (AWLS) model is proposed to refine the rough transmission, which is extracted by the dark channel (DC) model. In our model, the gray-world hypothesis and a smoothing technique with edge preservation are integrated to optimize the transmission and remove the artifacts which are brought by the DC model. The self-AWLS model has higher computational efficiency and can prevent the distortion of the recovered image better when the hazy image contains sky region, while many other dehazing techniques are not applicable for such images. Experimental results show that the proposed model is both effective and efficient for the haze removal application.

Super-resolution DOA estimation using single snapshot via compressed sensing off the grid

A super-resolution approach for direction of arrival (DOA) estimation is proposed in this paper. This approach can achieve the super-resolution performance using only one snapshot. It deals with the DOA estimation problem on the continuous bearing space rather than a discretization of this space, so it can eliminate the model mismatch and off-grid effect induced by the spatial discretization and thus improves the accuracy of DOA estimation. Moreover, we provide the conditions that guarantee the exact DOA estimation using one snapshot, including the theoretical low-bound of the required number of sensors and the minimum angle-distance between two consecutive sources. The experimental results demonstrate the superiority of the proposed approach on the super-resolution performance.

Salt-and-pepper noise removal based on image sparse representation

A new method for removing salt-and-pepper noise from corrupted images is presented. We employ an efficient impulse noise detector based on the image sparse representation to detect the noisy pixels, and a local median filter to estimate the intensity values of noisy pixels. Experimental results demonstrate that the proposed method obtains better performance in terms of both qualitative and quantitative evaluations than those de-noising techniques such as the median filter, the peak-and-valley filter, the detail preserving filter, and the boundary discriminative noise detection filter, etc. Especially, the proposed method provides high detection rate and preserves the detail very well.

Sparse Signal Recovery for Direction-of-Arrival Estimation Based on Source Signal Subspace

After establishing the sparse representation of the source signal subspace, we propose a new method to estimate the direction of arrival (DOA) by solving an -norm minimization for sparse signal recovery of the source powers. Second-order cone programming is applied to reformulate this optimization problem, and it is solved effectively by employing the interior point method. Due to the keeping of the signal subspace and the discarding of the noise subspace, the proposed method is more robust to noise than many other sparsity-based methods. The real data tests and the numerical simulations demonstrate that the proposed method has improved accuracy and robustness to noise, and it is not sensitive to the knowledge about the number of sources. We discuss the computational cost of our method theoretically, and the experiment results verify the computational effectiveness.

A New SVM Multi-Class Classification Method Based on Error-Correcting Code

Traditional SVM (support vector machine) multi-class classification methods are mainly based on one-to-one and one-to-multi, which both have disadvantages in applications: slow computational speed and low classification precision. This paper introduces a new method based on error correcting code to reduce the training time and improve the classification precision. In view of the relations among the length, the Hsmming distance and the order of the code and the generalization ability of each SVM, we propose the principles of code table-designing and the center-range method that ascertains the code order to eliminate the problem caused by error correcting code in factual application. Finally the results of experiments of HRRP recognition show this improved method has high computational efficiency and batter generalization ability.

A novel proposal distribution for particle filter

In order to solve the problem of degeneracy in particle filtering algorithm, a novel proposal distribution is designed in this paper. The principal idea of the proposal distribution is to fuse the latest observations together with the previous filtering estimate and the prior model information. In that case, the one-step smoothing estimate of the state is employed. Simulation results show that the improved particle filtering algorithm based on this proposal distribution is more accuracy than that of standard particle filter, the extended Kalman particle filter and the unscented particle filter. Besides, the particles drawn from the distribution proposed is more efficient.