Search

In this paper, the application of advanced Support Vector Regression (SVR) approach in predicting the electromagnetic (EM) response of graphene reconfigurable patch antenna is mainly introduced. First, the EM response evaluation problem is recast as a vector regression problem, and then SVR is used to find an alternative model of a microstrip patch antenna instead of a full-wave (FW) simulation tool. With patch geometry, frequency and chemical potential as input and reflection coefficient as output, the EM response of the next generation patch antenna was estimated by using SVR through learning the training samples. Compared with the results of FW simulation, the effectiveness and accuracy of the proposed method are shown.

The problem of the estimation of the direction-of-arrival(DoA) and the bandwidth(BW) of wideband signals impinging on time-modulated linear array(TMLA) of electromagnetic sensors is addressed in this work. Starting from the voltages measured at the output of the TMLA elements at multiple time instant, in order to recover the DoAs and BW, a customized strategy using Multi-Tasks Bayesian Compressive Sensing (MT-BCS) based on Laplace prior is proposed to improve the accuracy of the joint DoA and BW estimation, thanks to the statistical correlation of multiple frequency data samples acquired at consecutive time instants. Besides, the pulse durations of linear array is obtained by artificial bee colony-differential evolution(ABC-DE). The reconstruction algorithm starts with the sparsity of the signal source, divides the space angle into I part. Because TMLA works at periodic T, the received data can be expanded by Fourier series. We take the first Q-order sidebands and L times sampling, can get received signal matrix Y. The DoAs of signal is determined by sampling at different frequencies, and the bandwidth of signal is judged according to the presence or absence of the DoA. Simulation results is reported to show the behavior and the validity of the proposed approach.

The amount of academic articles in digital libraries is increasing exponentially. This growth of scientific papers’ growth made it difficult for researchers to obtain related papers from their queries. Recommendation systems can help them resolve the problem of information overload. However, existing paper recommender methods generally rely on the simple citation network, which ignores the semantic of papers and has the problem of cold start. In this paper, a hybrid paper recommendation approach AMHG is proposed which is based on a multi-level citation heterogeneous graph. Unlike existing works which only use the reference relationship, we consider the same or similar authors’ papers to alleviate the cold start problem of zero-citation and newly published papers. Besides, the metadata information of papers is also incorporated into a representation model to generate better recommender results to alleviate the cold start problem. We use the authors’ influence factors to reorder the candidate list outputting by MLP to obtain high-quality articles. Through experiments, we compare our model with several methods on the DBLP-REC dataset to demonstrate that AMHG outperforms state-of-the-art performance and the effectiveness of recommender.

Based on the multi-task Bayes Compressive Sensing (MT-BCS), a Direction-Of-Arrival (DOA) and bandwidth (BW) estimation strategy impinging on a linear array using multiple snapshots data is proposed. The DOA estimation is using as the reconstruction of sparse signal constrained by the Laplace prior through multi-task Bayes Compressive Sensing. Receiving wideband signal data through linear array, and the space is divided into I parts according to the equal interval. The data of interest are assumed to be represented as I-dimensional vector S, the wideband signal can be reconstructed accurately using only a small number. The receiving antenna operates in the frequency range [f min , f max ]. Starting from the voltages measured at the output of the array elements at a multiple time instant at f p = f min + ∆f, p = 1,…, P, the retrieval of the DoAs is addressed by means of a customized strategy based on MT-BCS in order to correlate the solutions obtained over different frequency samples. The bandwidth of the signals is obtained as a by product by identifying at which frequencies (f p , p = 1,…, P) the MT-BCS estimations include a signal along the i-th (i = 1,…, I) sampling direction. From the outputs of different frequency we can know the DoA and BW of signals. A preliminary numerical result is reported to show the behavior of the proposed approach in different snapshots and noise, and also using RMSE to present numerical results more clearly.

In this letter the application of advanced machine learning method to predict the electromagnetic (EM) response of reflectarray antenna elements with complex shapes is explored. The calculation of the scattering coefficients of passive reflectarray elements with periodic intervals of arbitrary degrees of freedom is firstly transformed into a regression estimation problem, and then an analysis model is established by the sample learning method based on support vector regression (SVR). Through a series of numerical experiments to evaluate the accuracy and the calculation efficiency show that, compared with the traditional full-wave (FW) simulation method, SVR can accurately and quickly predict the EM response of the reflectarray antenna elements.

To reconstruct sparsely distributed inhomogeneous objects, a Bayesian compressed sensing microwave imaging method based on Gauss prior is proposed. In the first order Born approximation, a sparse sensing model is established based on the electric field integral equation and the mesh discretization in the imaging region. The Bayesian probability density function based on the Gauss prior is constructed. The objective function is optimized by using the relevance vector machine method. The simulation imaging of multi-target and non-uniform target is studied, and the influence of noise is considered. The results show that the reconstruction results of Bayesian compressed sensing method based on Gauss prior are better than conjugate gradient iteration algorithm and orthogonal matching pursuit compressed sensing algorithm, which verify the effectiveness and robustness of the algorithm.

Facial expression is an important feedback of human feelings, and has always been the focus of human-computer interaction research. In recent years, with the rapid rise of artificial intelligence technology and the dramatic improvement of computing technology, the method of facial expression recognition has changed greatly, from the traditional machine learning method of image preprocessing, feature extraction and classification into classifier to the deep learning method based on large data cloud computing, and the recognition rate of facial expression has been remarkable. Raise. This paper introduces the common facial expression database, facial expression recognition based on classical machine learning method and popular deep learning method. Finally, the research of facial expression recognition is summarized and prospected.

In this letter the application of the Least Squares Support Vector Machines (LS-SVM) to soil moisture inversion based on active and passive microwave data is explored. The integral equation model (IEM) is chosen to obtain the backscattering coefficients which serve as the active microwave data for simulation. Similarly, the emissivity model is applied to perform simulations under a large range of soil moisture and surface roughness conditions in order to acquire the brightness temperature and generate the datasets. The frequencies of interest include 1.4 GHz (L-band), 6.9 GHz (C-band) as well as 10.7 GHz (X-band). The effectiveness of this application is assessed by implementing various input combinations (i.e. different microwave sensor frequencies, polarization status and incident angles). This study demonstrates the great potential of LS-SVM in the inversion of soil moisture based on microwave remotely sensed data.

In this paper, we explore the use of the Least Squares Support Vector Machines (LS-SVM) technique for modeling vegetation surface parameters inversion. The methodology has been applied to two sets of data to retrieve soil moisture content and vegetation water content from simulated microwave data. Based on the SMAP mission, we simulated the backscattering coefficients and the brightness temperature of vegetation surface by combining active and passive sensor at the frequencies of 1.26 and 1.41GHz, respectively, at incidence angles of 30, 40 degree. To assess the effectiveness of the method, the Gaussian distributed noise are added. A set of preliminary numerical results is provided in order to assess the performance of the proposed technique. This study demonstrates the great potential of LS-SVM in the estimation of soil moisture from microwave remotely sensed data.

In this paper, the two-dimensional target microwave imaging is studied by using compressive sensing method. Based on the electric field integral equation and the discrezation of the imaging region, a compressive sensing observation matrix is established, and multiple scattering field observation points are set up and the scattering electric field to be obtained based on the moment method. Finally, the reconstruction method is used to perform microwave imaging on the target position and dielectric constant. Reconstruction example includes the effects of different target numbers, different size targets, inhomogeneous targets, and noise. The numerical simulation results verify the effectiveness of the proposed method, and show that the compressive sensing method has important potential application in the field of microwave imaging.

Based on the D5000 online analysis system, an intelligent AGC initiative coordinated control system which composed by multi-modules is proposed. The main program module carry out AGC adjustable judgment, start/stop unit screening, real-time output adjustment automatically after distributed load synchronization ultra short-term forecasting. The auxiliary module proceeds pre-conditioning before switching and emergency adjustment after accidents. The intelligent AGC system can preferred and remote start/stop hydropower units, and optimal control units output automatically according to load fluctuation, besides, assist dispatchers to perform pre-conditioning before equipments blackout and emergency adjustment after accident trip. The simulation example shows that, the system can not only effectively improve implementation rate of optimal recently generation scheme and optimize performance indicators, but also promote the power grid stabilization adjustment, and reduce the labor intensity of dispatchers.

For a complex multitask and multiple low power mode processor, to make the processor enter a suitable low power mode in free time to reduce the power consumption, we can forecast the free time of the processor by Back Propagation (BP) neural network algorithm. But as a local search method and having the disadvantage being sensitive to initial weights, BP neural network has the low forecasting accuracy. In this paper, one dimensional convolution neural network is chosen to establish the forecasting model of free time, and some improvements are made based on two-dimensional convolution neural network to improve the suitableness for forecasting one-dimensional array. The simulation experiment shows that the accuracy of the data predicted by one dimensional convolutional neural network is higher than that of the BP neural network.