Electronic Engineering Institution

Huangshan, China

Electronic Engineering Institution

Huangshan, China
SEARCH FILTERS
Time filter
Source Type

Wang W.,Electronic Engineering Institution | Zheng J.-J.,Hefei University of Technology | Liu H.,Electronic Engineering Institution | Yang J.-A.,Electronic Engineering Institution
2015 International Conference on Computer Science and Applications, CSA 2015 | Year: 2015

Image blur is a common type of image distortion in our daily life. Though traditional image quality assessment (IQA) is increasingly drawing the attention of the research community, the uncertainness and uniqueness which exists in blur images bring unexpected difficulties. We present a simple yet effective automatic method on assessing global invariant blurred images. Different from the traditional IQA methods, our method not only utilize the objective metrics which are closed related to image content, but also consider human perception to show the humanism care. Extensive experiments have been conducted over a dataset that consists of Gaussian blurred images. Our method is robust to using varying amounts of labeled data and works satisfactorily on these challenging image data. © 2015 IEEE.


Liu Z.,Electronic Engineering Institution | Yang J.-A.,Electronic Engineering Institution | Liu H.,Electronic Engineering Institution | Wang W.,Electronic Engineering Institution
2015 International Conference on Computer Science and Applications, CSA 2015 | Year: 2015

Laplacian support vector machine could utilize the unlabeled samples for semi-supervised learning by applying the manifold regularization term. But the data adjacent graph in the manifold regularization term couldn't take advantage of the label information and the empirical setting of heat kernel parameter would also degrade the learning performance. Inspired by human behavioral learning theory, a novel semi-supervised learning with local behavioral similarity was proposed in this paper to solve those problems. In detail, the new edge weight with label information was introduced and the local distribution parameter considering the underlying probability distribution in the neighborhood of a point was applied. Extensive experiments on public data sets show the good performance and validity of the new algorithm. © 2015 IEEE.


Yin H.,Electronic Engineering Institution | Yang J.-A.,Electronic Engineering Institution | Wang W.,Electronic Engineering Institution | Liu H.,Electronic Engineering Institution
IEICE Transactions on Information and Systems | Year: 2017

Transfer boosting, a branch of instance-based transfer learning, is a commonly adopted transfer learning method. However, currently popular transfer boosting methods focus on binary classification problems even though there are many multi-classification tasks in practice. In this paper, we developed a new algorithm called MultiTransferBoost on the basis of TransferBoost for multi-classification. MultiTransferBoost firstly separated the multi-classification problem into several orthogonal binary classification problems. During each iteration, MultiTransferBoost boosted weighted instances from different source domains while each instance's weight was assigned and updated by evaluating the difficulty of the instance being correctly classified and the "transferability" of the instance's corresponding source domain to the target. The updating process repeated until it reached the predefined training error or iteration number. The weight update factors, which were analyzed and adjusted to minimize the Hamming loss of the output coding, strengthened the connections among the sub binary problems during each iteration. Experimental results demonstrated that MultiTransferBoost had better classification performance and less computational burden than existing instance-based algorithms using the One-Against-One (OAO) strategy. © 2017 The Institute of Electronics, Information and Communication Engineers.


Yin H.,Electronic Engineering Institution | Yang Y.-A.,Electronic Engineering Institution
AIP Conference Proceedings | Year: 2017

In this paper, we propose a new transfer learning algorithm for online training. The proposed algorithm, which is called Online Transfer Extreme Learning Machine (OTELM), is based on Online Sequential Extreme Learning Machine (OSELM) while it introduces Semi-Supervised Extreme Learning Machine (SSELM) to transfer knowledge from the source to the target domain. With the manifold regularization, SSELM picks out instances from the source domain that are less relevant to those in the target domain to initialize the online training, so as to improve the classification performance. Experimental results demonstrate that the proposed OTELM can effectively use instances in the source domain to enhance the learning performance. © 2017 Author(s).


Huang Z.,Electronic Engineering Institution | Zheng Z.,Institute of North Electronic Equipment | Zhang J.,Electronic Engineering Institution
Dianbo Kexue Xuebao/Chinese Journal of Radio Science | Year: 2015

In order to solve the problem of worse angel estimation which caused by the power being transmitted dispersedly in traditional multiple-input multiple-output (MIMO) radar, a new method for pattern synthesis of MIMO radar is proposed. Firstly, the optimal model is constructed based on second-order cone programming (SOCP) by combining the focus range of the transmit power with the rotational invariance of the transmit steering vector. It can not only constrain the optimal error less than the given value, but also can minimize the peak side lobe power of the transmit pattern. Further, the diagonal elements of the covariance matrix are constrained, which can maximum the power efficiency. Moreover, the SOCP form of the optimal model is given so as to use the primal-dual interior point method effectively. Lastly, simulations illustrate that the beam weight matrix can make the transmit power focus within the desire space and constrain the transmit array power to be equal, which is propitious to the energy usage. Therefore, the angle estimation performance by using the proposed method in this paper is better than the traditional MIMO radar when the transmit power and angle estimation method are same. Copyright © 2015 by Editorial Department of Chinese Journal of Radio Science


Ma P.,Electronic Engineering Institution | Zhou Q.,Electronic Engineering Institution | Zhang J.,Institution of Engineering and Technology
Applied Mechanics and Materials | Year: 2012

Phase synchronization errors are practically inevitable, so in this paper the quantitative tool to asses the effect of antenna placement for localization performance with phased synchronization errors is provided. The lower bound of mean-square error (MSE) is set by the hybrid Cramer-Rao bound (HCRB) for the joint estimation of the target location and phase synchronization errors at the receivers. It is shown that HCRB follow up to a lower limit, determined by the synchronization error variance and the number of transmit and receive sensors. For uniform antenna distributing, symmetrical placement is optimal, and all forms give exactly the same performance. The localization performance of nonuniform arrays when the antennas are spread out as much as possible is a little better than symmetrical placement at high SNR. Simulation results verify the correctness of conclusions.


Huang Z.-R.,Electronic Engineering Institution | Niu Z.-Y.,Electronic Engineering Institution | Zhang J.-Y.,Electronic Engineering Institution
Xi Tong Gong Cheng Yu Dian Zi Ji Shu/Systems Engineering and Electronics | Year: 2015

A method for the orthogonal serial phase code waveform design of multiple-input multiple-output radar based on sequential cone programming is proposed, which minimizes the auto-correlation peak sidelobe level (APSL) and cross-correlation peak level (CCPL) of transmit signal as its objective function. Taylor approximation is adopted to transform the original problem which is non-convex into a series second-order cone programming problems via sequential cone programming framework in each iterative point. Moreover, in order to improve the optimal results, the threshold of phase increment is constrained as a linear function. Finally, simulations show that the orthogonal waveform performance is better than the reference method when the number of the transmit array and the code is same. At the same time, the APSL and CCPL can be accurately controlled respectively by using the proposed method, which is very flexible for the orthogonal waveform design.


Liu Z.,Electronic Engineering Institution | Yang J.,Electronic Engineering Institution | Liu H.,Electronic Engineering Institution | Wang W.,Electronic Engineering Institution
ICIC Express Letters, Part B: Applications | Year: 2016

As an important semi-supervised learning, Laplacian support vector machine utilizes the unlabeled data for learning by adding the manifold regularizer into the objective function. However, the data adjacent graph in the manifold regularizer was not good at data structure representation as the label information was neglected. Moreover, the heat kernel parameter is usually empirical Fixed and neglected the local distribution information, which might also degrade the learning performance. Inspired by human behavioral learning theory, a novel semi-supervised learning with local behavioral similarity was proposed to solve those problems. In detail, a new data adjacent graph considering label information was constructed by introducing behavioral similarity based edge weight. Besides, a local distribution parameter considering the underlying probability distribution in the neighborhood was applied. Extensive experiments on public datasets show the good performance and validity of the new algorithm. © 2016 ICIC International.


Huang Z.-R.,Electronic Engineering Institution | Zhang J.-Y.,Electronic Engineering Institution | Zhou Q.-S.,Electronic Engineering Institution
Dianzi Yu Xinxi Xuebao/Journal of Electronics and Information Technology | Year: 2015

In order to solve the problem that the power is transmitted dispersedly in traditional bistatic MIMO radar, a method about the transmit power focus of the bistatic MIMO radar is proposed. Firstly, the mathematic model is constructed based on an effective optimal criterion that can not only constrain the maximum error of the optimal beam and the desire one less than the given value, but also minimize the peak side lobe power of the transmit pattern. At the same time, a special beam matrix of transmitting terminal is constructed, not only making equivalent transmit/receive steering vector possessing the character of rotational invariance property, but also simplifying the original optimized model in order to be fast solved via second-order cone programming theory. Secondly, the transmitting and receiving angle of space target is estimated by utilizing improved PARallel FACtor (PARFAC) algorithm. The initial iteration point in the least square algorithm of PARFAC interior is improved by integrating into rotation invariance of transmitting/receiving steering vector, which can effectively decrease the number of iteration. Furthermore, the Cramer-Rao Bound of multi-target angle estimation under bistatic MIMO radar transmitting power focusing is derived which prove the superiority of the proposed method. Finally, the simulation results show the effectiveness of the theoretical analysis. ©, 2015, Science Press. All right reserved.


Huang Z.,Electronic Engineering Institution | Liu C.,Electronic Engineering Institution | Wang T.,Electronic Engineering Institution
Shengxue Xuebao/Acta Acustica | Year: 2014

In order to solve the problem of the high side lobe level in the circular array, a pattern synthesis method is proposed. It firstly makes the peak side lobe level as its fitness function to optimize the locations and coefficients of the array elements based on genetic algorithm, which is modified in order to avoid premature convergence. It accords with the academic global optimization and can greatly improve the algorithm search performance. Secondly, it makes the optimal results in the first step as a good initial iterative point and adopts Taylor approximation to transform the original pattern synthesis problem which is non-convex into sequential cone programming framework near the good initial point. After this it can be solved efficiently by the convex optimization. Due to the late variability capacity in the operation, it can effectively improve the optimization performance. Lastly simulations demonstrate that when the array number is constant, this method can not only reduce the side lobe level but also minimize its dynamic range. And it achieves a better performance of circular array pattern in comparison with the reference method. ©, 2014, Institute of Acoustics, CAS. All right reserved.

Loading Electronic Engineering Institution collaborators
Loading Electronic Engineering Institution collaborators