Compressive Sensing of Stepped-Frequency Radar Based on Transfer Learning

Compressive sensing is a sub-Nyquist sampling technique for efficient signal acquisition and reconstruction of sparse or compressible signals. In order to account for the sparsity of the underlying signal of interest, it is common to use sparsifying priors such as Bernoulli–Gaussian-inverse Gamma (BGiG) and Gaussian-inverse Gamma (GiG) priors on the components of the signal. With the introduction of variational Bayesian inference, the sparse Bayesian learning (SBL) methods for solving the inverse problem of compressive sensing have received significant interest as the SBL methods become more efficient in terms of execution time. In this paper, we consider the sparse signal recovery problem using compressive sensing and the variational Bayesian (VB) inference framework. More specifically, we consider two widely used Bayesian models of BGiG and GiG for modeling the underlying sparse signal for this problem. Although these two models have been widely used for sparse recovery problems under various signal structures, the question of which model can outperform the other for sparse signal recovery under no specific structure has yet to be fully addressed under the VB inference setting. Here, we study these two models specifically under VB inference in detail, provide some motivating examples regarding the issues in signal reconstruction that may occur under each model, perform comparisons and provide suggestions on how to improve the performance of each model.

This paper introduces a novel reconstruction method of high-resolution range profile (HRRP) using sparse frequency-stepped chirp signal (FSCS). In the method, Compressed Sensing (CS) theory is utilized to reconstruct the moving target's HRRP. With this method even if the subpulses number of FSCS is incomplete, the HRRP can still be reconstructed successfully, and the final reconstructed ISAR image is clear enough for identification and classification of the target. Furthermore, the effect of noise is also analyzed by corrupting the data via an uncorrelated additive white Gaussian process.

Stepped-frequency chirp signal (SFCS) is an important signal form used for high-resolution radar. It contains several sub-chirp signals with stepped carrier frequency and can be synthesized to a wide-band chirp signal. Dechip method is a quick, effective method to process SFCS, it synthesizes the sub-pulses of SFCS in time-domain. The usual design rule of SFCS is the frequency step (Δf) must be less than the bandwidth of sub-chirp (B m ). If Δf > B m , the synthesized signal will have gaps, which leading to grating lobes in radar image. Here we proposed an algorithm by using Super-SVA to extrapolate each sub-pulse so as to fill the gaps and depress grating lobes. Without the restriction of Δf m , it will be very helpful for using less sub-chirps to obtain higher resolution and reducing the influence of target motion on the quality of synthesized signal. We presented simulations to verify the proposed algorithm. Furthermore the results imply that super-SVA can not only be used for frequency-domain extrapolation, but also time-domain extrapolation of chirp signal.

A classifier based on a syntactic approach is developed for High range resolution (HRR) radar target recognition. An attribute grammar is used to represent the structure of an HRR signature and an error-correcting parsing mechanism is implemented to extract peaks in the HRR profile and suppress the extraneous spikes. In the training phase, an error correcting grammatical inference technique is employed for structural inference of HRR signatures using a positive sample set. Recognition is done using a minimum distance classifier where Levenshtein error measure is used as the distance metric. The error-correcting parsing procedure for peak extraction is used to perform both inference and recognition. Experiments performed using public release MSTAR database indicate that this approach has sufficient discrimination power to perform target detection in HRR signatures.

Adaptive multinoulli-based Kalman filter with randomly unknown delayed and lost measurements

In many practical fields, the unknown time-varying measurement biases (additive and multiplicative bias) and heavy-tailed measurement noise caused by some unpredictable anomalous behaviors may degrade the performance of conventional Kalman filter seriously. To solve the state estimation problem of systems with time-varying measurement biases and heavy-tailed measurement noise, this paper proposes a new variational Bayesian (VB) based robust filter. Firstly, the non-Gaussian measurement likelihood probability density function (ML-PDF) with multiplicative and additive measurement bias is built. Then, the conjugate prior distributions for unknown bias and noise scale parameters are selected, and the VB method is utilized to jointly infer the system state, unknown measurement biases and inaccurate measurement noise covariance matrix. Finally, a VB based robust filter is derived and its effectiveness is verified by the numerical simulations.

A Gaussian-multivariate Laplacian mixture distribution based robust cubature Kalman filter

We investigate the performance of sparse signal recovery from a set of compressively sensed noisy measurements using variational Bayesian (VB) inference. The framework considered here is sparse Bayesian learning with no specific structure other than sparsity assumed on the signal. Two models are considered here, the issues of each model is studied, and the performances in reconstruction are compared. These models have been already applied to the compressive sensing problems implemented via expectation-maximization algorithm, Markov chain Monte Carlo (MCMC), etc., for different structures of the solution. However, the issues when using variational Bayes inference for only promoting the sparsity have not been investigated. This paper provides details on such modeling and the issues accompanied with each modeling are discussed.

Online social networks (OSNs) have reduced global boundaries, with Twitter enabling perspective sharing. Bot profile‐propagated false information misuse raises serious concerns. Considering this issue, we present our research on classifying Twitter accounts as “human” or “bot” using deep neural networks and transfer learning. Our proposed approach, TL‐PBot, stands for bot profile detection using transfer learning. The TL‐PBot framework utilizes Twitter account metadata such as follower count. Our TL‐PBot also incorporates text data from the Twitter description field as a feature. Word representation of the text data is achieved using Global Vectors (GloVe), a pre‐trained model. By employing user profile‐based features, we significantly reduce the overhead of feature engineering. The hybrid nature of the model enables it to effectively handle mixed‐type features, including text, binary, and numerical data. We design the network using long‐short‐term memory (LSTM) units. DNN model layers were trained, and the weights of the pre‐trained model layers were frozen to apply the transfer learning, resulting in reduced training time and improved bot profile detection accuracy. The performance of the proposed TL‐PBot is evaluated using publicly available datasets. The proposed approach is trained and tested on the same datasets and further evaluated on the validation datasets that were not used in the training phase, which is also a novelty in our approach. Comparative analysis with state‐of‐the‐art approaches demonstrates that the TL‐PBot approach achieves a higher accuracy of 98.07%, while excelling in precision of 99%, recall of 98%, f measure of 98.32%, and AUC of 0.99. Employing the transfer learning strategy resulted in an accelerated detection rate of 5.04 milliseconds, attesting to the effectiveness of this approach in enhancing computational efficiency.

Implementation of time-efficient adaptive sampling function design for improved undersampled MRI reconstruction

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) needs high temporal and spatial resolution to accurately estimate quantitative parameters and characterize tumor vasculature. Compressed Sensing (CS) has the potential to accomplish this mutual importance. However, the randomness in CS under-sampling trajectory designed using the traditional variable density (VD) scheme may translate to uncertainty in kinetic parameter estimation when high reduction factors are used. Therefore, accurate parameter estimation using VD scheme usually needs multiple adjustments on parameters of Probability Density Function (PDF), and multiple reconstructions even with fixed PDF, which is inapplicable for DCE-MRI. In this paper, an under-sampling trajectory design which is robust to the change on PDF parameters and randomness with fixed PDF is studied. The strategy is to adaptively segment k-space into low-and high frequency domain, and only apply VD scheme in high-frequency domain. Simulation results demonstrate high accuracy and robustness comparing to VD design.

Incomplete mixed data-driven outlier detection based on local–global neighborhood information

We propose to use a low order autoregressive (AR) model to describe the temporal structure of source. Then we adopt Variational Bayesian (VB) method for source separation which models additive noise into the mixing system. The approach integrates the source probabilistic model and noise probabilistic model. Its goal is to approximate the actual probability density function of hidden variables and parameters using its approximating posterior distribution by minimizing the misfit between them. The advantage of our VB-AR algorithm is that it can exploit the temporal nature of source signals and avoid overfitting in the separating process. This algorithm can also identify the AR order. Experiments on artifact and real-world speech signals are used to verify our proposed algorithms. As a result, the lower AR source model improves the separation. The performance of the algorithm is compared with that of i.i.d. separation algorithm and the second-order separation algorithm.

This paper investigates a new multifrequency compressed sensing (CS) model for 2-D near-field microwave and millimeter-wave synthetic aperture radar (SAR) imaging system, which usually collects multifrequency sparse data. Spatial data of each frequency are represented as a hierarchical tree structure under a wavelet basis and spatial data of different frequencies are modeled as a joint structure, because they are highly correlated. Based on the developed multifrequency CS model, a new CS approach is proposed by exploiting both the intrafrequency and interfrequency correlations, and enriches the existing CS approaches for 2-D near-field microwave and millimeter-wave SAR image reconstruction from undersampled measurements. Combining a splitting Bregman update with a variation of the parallel Fast Iterative Shrinkage-Thresholding Algorithm-like proximal algorithm, the proposed CS approach minimizes a linear combination of five terms: a least squares data fitting, a $\textrm {multi-}\ell _{1}$ norm, a multitotal variation norm, a joint-sparsity $\ell _{21}$ norm, and a tree-sparsity overlapping $\ell _{21}$ norm. Simulation and experimental results demonstrate the superior performance of the proposed approach in terms of both efficiency and convergence speed.

Particle filters for probability hypothesis density filter with the presence of unknown measurement noise covariance