Low probability of intercept (LPI) has currently become a central issue in modern electronic warfare. In this paper, the problem of task-dependent adaptive radar jamming waveform design based on LPI is investigated. Firstly, the achievable signal-to-interference-plus-noise ratio (SINR) and mutual information (MI) are derived to evaluate the target detection and parameter estimation performance, respectively. Then, regarding to the complexity and uncertainty of electromagnetic environment in the modern battlefield, the trapezoidal fuzzy number is utilized to describe the threshold of overall system performance based on the credibility theory, whose purpose is to minimize the total jamming power, while the achievable system performance outage probability is enforced to be greater than a specified confidence level. Finally, the fuzzy chance-constrained programming (FCCP) models are transformed to the crisp equivalent forms with the property of trapezoidal fuzzy number. Simulation results demonstrate that our proposed approaches can effectively achieve the optimal solutions and bring remarkable improvement on the LPI performance for radar jamming.
Sign Language Recognition (SLR) targets on facilitating the communication between deaf-mute people and ordinary people. This task is very challenging due to the complexity and large variations in hand postures. Some methods require user wear sensor gloves which can detect the position and angle of finger articulations. Others use RGB-D camera like Kinect to track hands and rely on complex algorithms to segment hands from background. However, all these methods have its own disadvantages. Sensor-based methods are not natural as the user must wear cumbersome instruments while camera-based methods have to design extra algorithms to track and segment hands from complex background. To address these problems, we propose a novel method for SLR which involves the use of the Real-Sense. It is a camera device which can detect and track the location of hands in a natural way. More powerful, it provides the 3D coordinates of finger joints in real time. We build a deep neural network (DNN) based on Real-Sense to recognize different signs. The DNN takes the 3D coordinates of finger joints as input directly without using any handcrafted features. The reason is that DNN, as a deep model, is capable of learning suitable features for recognition from raw data. In experiment, to demonstrate the effectiveness of Real-Sense, we collect two datasets by Real-Sense and Kinect respectively, then build DNNs based on each dataset for recognition. To validate the powerfulness of DNN, we compare the performance of DNN and support vector machine (SVM) on the same dataset.
This paper considers optimal multi-carrier (multiple messages) spread-spectrum (SS) data embedding on linearly-transformed host. We present information-theoretic security analysis for the optimal SS embedding. The security is quantified by both the Kullback-Leibler distance and Bhat-tacharyya distance between the cover and stego probability distributions. The main results of this paper permit to establish fundamental security limits for the optimal SS embedding. Theoretical analysis and experimental results show the impact of the number of embedding messages, the embedding distortion, and the host transformation in the security level.
Online review plays an important role when people are making decisions to purchase a product or service. It is shown that sellers can benefit from boosting their product review or downgrading their competitors' product review. Dishonest behavior on reviews can seriously affect both buyers and sellers. In this paper, we introduce a novel angle to detect dishonest reviews, called Equal Rating Opportunity (ERO) evaluation. The proposed ERO evaluation can detect embedded manipulation signals based on limited amount of data. Experiments based on real data are conducted. Four highly problematic products are successfully detected from 84 products.
Wireless Body Area Networks (WBAN) is a promising type of networks that mainly targets at applications in ubiquitous communication and e-Health services. To support long-term pervasive services, both reliability and energy efficiency are required in WBAN communications. In this paper, by adopting compressive sensing (CS) technology, a reliable transmission strategy for WBANs is proposed to alleviate the effect of packet loss and improve the data transmission reliability. Considering the transmission model with both direct link and relay link, two relay policies are present and the corresponding fusion rules are designed to merge the data from both links for better reconstruciton. A reconstruction algorithm, which exploits both block sparsity and temporal correlation, is proposed to reconstructed the merged data. The simulation results demonstrate the effectiveness of the proposed transmission strategy.
Sensor pattern noise (SPN) has been proved to be an inherent fingerprint of a camera, and it has been broadly used in the fields of image authentication and camera source identification. However, the SPN extracted using current denoising algorithm always contains image content residual, which would significatively influence the accuracy of camera source identification. In this paper, a novel patch-based (PB) sensor pattern noise algorithm for camera source identification is proposed to solve this problem. Low-complexity patches of images are selected to construct local reference SPN, which contains least image content residual. The global reference SPN is constituted with the block-wised local SPN. Similarly for the test image, SPN is extracted from low-complexity region, and making correlation with corresponding local reference SPN. Our experiments on the Dresden database demonstrate that the proposed approach outperforms two sensor pattern noise estimation methods on the literatures as baseline.
A new policy-iteration algorithm based on neural networks (NNs) is proposed in this paper to synthesize optimal control laws online for continuous-time nonlinear systems. Latest advances in this field have enabled synchronous policy iteration but require an additional tuning loop or a logic switch mechanism to maintain system stability. A new algorithm is thus derived in this paper to address this limitation. The optimal control law is found by solving the Hamilton-Jacobi-Bellman (HJB) equation for the associated value function via synchronous policy iteration in a critic-actor configuration. As a major contribution, a new form of NN approximation for the value function is proposed, offering the closed-loop system asymptotic stability without additional tuning scheme or logic switch mechanism. As a second contribution, an extended Kalman filter is introduced to estimate the critic NN parameters for fast convergence. The efficacy of the new algorithm is verified by simulations.
Conventional online education platforms implement a combination of static and PTZ (Pan-Tilt-Zoom) cameras which are not only costly but also require operators and high computational power. In this paper, we establish a real-time face tracking system based on a single PTZ camera with automatic operation as a cost effective solution. The proposed system takes into account the camera movement and network delays in the tracking algorithm design while maintaining cost and flexibility in practical implementation.
We consider joint adaptive transmit(TX) and receive (RX) beamforming for interference mitigation in array sensing systems. Unlike conventional designs, which only employ adaptive processing for the RX beamforming, we propose a fully adaptive approach that jointly selects the transmit correlation matrix and RX beamformer by maximizing the signal-to-interference-plus-noise ratio (SINR). Due to imprecise knowledge of the interference (e.g., because of limited training data), employing only adaptive RX beamforming may be inadequate for effective interference cancellation, whereas joint adaptive transmit and RX beamforming can afford a stronger ability to handle the interference. Numerical examples are presented to evaluate the performance of the proposed joint beamforming approach.
Compressive sensing (CS) based dynamic MRI techniques have been proposed to improve the imaging speed and spatiotemporal resolution. However, existing CS recovery methods haven't exploited the rich redundancy among the spatial and temporal dimensions. In this paper, we address the CS recovery of dynamic MRI from partially sampled k-t space using the nonlocal low-rank regularization (NLR). To exploit the nonlocal redundancy in the spatial-temporal dimension, the dynamic MRI sequence is divided into overlapping 3D patches along both the spatial and temporal directions. We exploit the fact that the matrix that consists of a sufficient number of similar patches is low-rank. To effectively approximate the low-rank matrix, the non-convex surrogate function logdet (·) is used instead of the convex nuclear norm. Experimental results show that our proposed method can outperform existing state-of-the-art dynamic MRI reconstruction methods.
With the development of multiple-input multiple-output (MIMO) radar, in addition to satisfy general function in target detection, most systems also involve the function of target recognition, parameter estimation, etc. This requires that the MIMO radar works in wideband mode. This paper proposes a way to detect moving target using Doppler effect correction, while the airborne MIMO radar transmits wideband frequency-division linear frequency modulation (FDLFM) signals. Both moving target detecting (MTD) processing and joint domain localized (JDL) processing are used to verify the proposed method. Simulation results also demonstrate the validity of the method.
Information fusion is a key research area widely applied to various multimedia analysis tasks such as artificial intelligence, humancomputer interaction, robotics, distributed computing, financial systems and security/surveillance. Feature level fusion has been considered as the most promising fusion method due to the rich information presented at this level. A critical operation of feature level fusion is the projection of the features onto a space which best presents the information for recognition. However, the identification of the optimal projection of the multimodal features onto the projected space remains a difficult task. This paper presents a graph representation approach for selecting optimal projection in information fusion which substantially minimizes the effort of finding the optimal or near-optimal dimension of the features in the projected space. The effectiveness of the proposed method is demonstrated through numerous experimentation on handwritten digit recognition and audio based emotion recognition problems.
In this paper, we present a printed image forensics method based on detection of halftone dots arrangement. The proposed method can be applied to identify printers and expose image forgery. The printed image is composed of halftone dots and these dots have regular arrangement. For a halftone dot, its adjacent halftone dots can be seen as the vertexes of a hexagon, we model the printer by 6 distance and 6 angles based on the halftone dot positions in the hexagon. In examination step, we apply Euclidean distance and K-means to identify printer and expose the forgeries, respectively. The experimental results have demonstrated the performance of proposed method both on printer identification and forgery detection.
In this paper, a new algorithm is proposed for the object tracking in the complex background, in which discrete cosine transform (DCT) features of image are used for expressing candidate objects and the templates, casting tracking as a sparse approximation problem. Firstly, the location of object is labeled and templates of DCT features are constructed in the first frame. Secondly, the candidate objects are searched by random particles in the next frame and the sparsity is achieved by solving an h-regularized least-squares problem. The candidate with the smallest object templates projection error is chosen as the tracking object. Finally, the DCT feature templates are updated using the most recent tracking results to capture changes of the object appearance. Experimental results show the effectiveness of the proposed method comparing with other tracking algorithms.
In practical automatic speech recognition (ASR) systems, it is difficult to recognize words that are with low-frequency in the language model (LM) training data. Ironically, these words tend to be highly important as they are often domain-specific name entities. In order to meet this challenge, we present a novel approach that enhances the weights of these words by borrowing information from some high-frequency words that are similar to the target words. Experimental results demonstrated that our method can significantly improve ASR performance on low-frequency words and does not impact performance on high-frequency words. Additionally, this method can be easily extended to deal with new words that are absent in the LM training data.
The micro-motion characteristics of warheads have been utilized to discriminate false warheads from true ones. To obtain accurate dimensional measurements, a multiple-input multiple-output (MIMO) radar is adopted to observe the kinetic information of the warhead. A distributed state space model (SSM) is built and the differences between the true and false warheads are characterized as different system parameters of the SSM. In this paper, we extend the locally optimal unknown direction (LOUD) detector, which has shown its effectiveness for hypothesis testing, to the underlying distributed detection problem, and a novel consensus-based LOUD detector is proposed. The superior detection performance of the proposed detection algorithm in identifying the true and false warheads is verified using simulation results.
Consider an uplink SIMO network consisting of multiple base stations (BSs) and multiple users, where each user has an individual transmit power constraint, and the BSs are allowed to cooperate in data receiving. To guarantee fairness among users and avoid heavy burden of backhaul data exchange, we maximize the minimum SINR based on joint BS selection and beamforming. We formulate this problem from the perspective of sparse beamforming. However, the scaling ambiguity of the receive beamformers will make the sparse constraints trivial. Inspired by the observation that the transmit beamformer is immune to the scaling ambiguity, we apply the duality theorem to the uplink max-min problem under per-user power constraints, and reformulate it as an equivalent downlink problem. An iterative two-stage algorithm is then developed to solve the sparse beamforming problem efficiently. The effectiveness of the proposed algorithm is demonstrated by numerical simulations.
Joint precoding for multiple-input multiple-output (MIMO) two-way relay networks has recently received much attention. In this paper, we present a coherent way to iteratively optimize the source and relay precoding matrices based on convex programming. Although the proposed total mean-square error (MSE) minimization problem is not joint convex for both the sources and relay, the subproblems for separate optimization can be both formulated as a standard convex programming problem. Numerical results further show that the proposed iterative precoding algorithm converges very fast.
Learning a deep architecture involves a tough issue called hyperparameter search. This is especially the case for convolutional neural networks with a large number of hyperparameters. To solve this problem, we propose a tensor completion method to predict the best architecture configurations for convolutional neural networks. This method is based on a hypothesis that the generalization performance of a deep architecture is controlled by several effect factors, each of which is a function of hyperparameter of the deep architecture. Predicted generalization accuracy of the best configurations are checked by carrying out deep learning computation. Since generalization performance for a practical recognition task is always data- and code-dependent, we tried out our method on an open deep learning platform named Caffe, and we increased the generalization accuracy from 98.97% to around 99.25% on MNIST by replacing only five numbers.
