Time filter

Source Type

Gong X.,University of South China | Gong X.,Chongqing Key Laboratory of Computational Intelligence | Luo J.,Sichuan Academy of Medical science | Fu Z.,University of South China
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Year: 2013

This paper presents a framework for 3D face representation, including pose and depth image normalization. Different than a 2D image, a 3D face itself contains sufficient discriminant information. We propose to map the original 3D coordinates to a depth image using a specific resolution, hence, we can remain the original information in 3D space. 1) Posture correction, we propose 2 simple but effective methods to standardize a face model that is appropriate to handle in following steps; 2) create depth image which remain original measurement information. Tests on a large 3D face dataset containing 2700 3D faces from 450 subjects show that, the proposed normalization provides higher recognition accuracies over other representations. © Springer International Publishing 2013.

Wang J.,Chongqing University of Posts and Telecommunications | Wang J.,Chongqing Key Laboratory of Computational Intelligence | Jin L.,Chongqing University of Posts and Telecommunications | Sun K.,Chongqing University of Posts and Telecommunications
Jiangsu Daxue Xuebao (Ziran Kexue Ban)/Journal of Jiangsu University (Natural Science Edition) | Year: 2013

In order to improve the performance of Chinese text categorization, a Chinese text categorization method was proposed based on evolutionary hypernetwork. A Chinese Lexical Analysis System (ICTCLAS) was employed to take the words with parts of verb, noun and adjective as candidate features. The χ2-test method was used to realize feature selection, and the feature weight was calculated by Boolean weighting. The preprocessed data sets were divided into training set and testing set. A hyperedge replacement strategy was used to train hypernetwork classification model for classifying testing sets. The classification performances of the hypernetwork models with different orders were analyzed and compared with traditional KNN and SVM. The experimental results show that the proposed scheme can achieve 87.2% and 72.5% of macro precision, 86.9% and 70.5% of macro recall, 87.0% and 71.5% of macro F1 for Fudan University corpus and Sohu corpus, respectively. As an efficient tool for Chinese text classification, the proposed scheme is close to or better than KNN and SVM classification methods.

Zhu S.,Chongqing University of Posts and Telecommunications | Zhu S.,Chongqing Key Laboratory of Computational Intelligence | Zeng X.,Chongqing University of Posts and Telecommunications | Zeng X.,Chongqing Key Laboratory of Computational Intelligence | And 2 more authors.
Chinese Journal of Sensors and Actuators | Year: 2016

Location methods in wireless sensor networks(WSN)by received signal strength indicator(RSSI)are relatively inexpensive. Location Estimation-Locality Preserving Canonical Correlation Analysis(LE-LPCCA)can fit the structure of WSN approximately using RSSI similarity among nodes and achieve high localization accuracy,but it only uses data similarity while ignoring data dependency between signal and physical space,and employs rough centroid method. As to problems above,this paper proposes Location Estimation-Improved Locality Preserving Canonical Correlation Analysis(LE-ILPCCA)method. In training phrase,it combines data similarity and dependency using a balance parameter to compute more precise projection transformation of RSSI inner lower dimensional coordinates;in localization phrase,it calculates location of unknown nodes utilizing accurate transformational relation between position coordinates and RSSI inner lower dimensional coordinates of known nodes. Experimental results show that our method has a higher accuracy and stability than LE-LPCCA and LE-CCA. © 2016, The Editorial Office of Chinese Journal of Sensors and Actuators. All right reserved.

Qin H.,Chongqing Key Laboratory of Computational Intelligence | Qin H.,Chongqing University of Posts and Telecommunications | Ye B.,Chongqing Key Laboratory of Computational Intelligence | Ye B.,Chongqing University of Posts and Telecommunications | And 2 more authors.
Computerized Medical Imaging and Graphics | Year: 2015

Volume visualization is a very important work in medical imaging and surgery plan. However, determining an ideal transfer function is still a challenging task because of the lack of measurable metrics for quality of volume visualization. In the paper, we presented the voxel vibility model as a quality metric to design the desired visibility for voxels instead of designing transfer functions directly. Transfer functions are obtained by minimizing the distance between the desired visibility distribution and the actual visibility distribution. The voxel model is a mapping function from the feature attributes of voxels to the visibility of voxels. To consider between-class information and with-class information simultaneously, the voxel visibility model is described as a Gaussian mixture model. To highlight the important features, the matched result can be obtained by changing the parameters in the voxel visibility model through a simple and effective interface. Simultaneously, we also proposed an algorithm for transfer functions optimization. The effectiveness of this method is demonstrated through experimental results on several volumetric data sets. © 2014 Elsevier Ltd.

Zeng X.,Chongqing University of Posts and Telecommunications | Zeng X.,Chongqing Key Laboratory of Computational Intelligence | Tang S.,Chongqing University of Posts and Telecommunications | Tang S.,Chongqing Key Laboratory of Computational Intelligence
Chinese Journal of Sensors and Actuators | Year: 2013

Aimed at those problems that Fast MDS-MAP (Multidimensional Scaling Map) localization algorithm has the high location error in irregularly shaped wireless sensor networks and is unable to select different granular levels of the network to locate, a multi-granularity-based manifold learning method was proposed for localization in wireless sensor networks, abbreviated as MG-MDS. Different granular framework nodes can be obtained by selecting the different filter radius, and a new strategy is introduced for transforming relative coordinates to absolute coordinates. Experimental results show that the MG-MDS algorithm can get the higher locating accuracy than the Fast MDS-MAP algorithm in irregular wireless sensor networks, and the localization error will be smaller when the granularity size of the network becomes finer.

Xiao D.,Beihang University | He T.,Beihang University | Pan Q.,Beihang University | Liu X.,Beihang University | And 2 more authors.
Ultrasonics | Year: 2014

A novel acoustic emission (AE) source localization approach based on beamforming with two uniform linear arrays is proposed, which can localize acoustic sources without accurate velocity, and is particularly suited for plate-like structures. Two uniform line arrays are distributed in the x-axis direction and y-axis direction. The accurate x and y coordinates of AE source are determined by the two arrays respectively. To verify the location accuracy and effectiveness of the proposed approach, the simulation of AE wave propagation in a steel plate based on the finite element method and the pencil-lead-broken experiment are conducted, and the AE signals obtained from the simulations and experiments are analyzed using the proposed method. Moreover, to study the ability of the proposed method more comprehensive, a plate of carbon fiber reinforced plastics is taken for the pencil-lead-broken test, and the AE source localization is also realized. The results indicate that the two uniform linear arrays can localize different sources accurately in two directions even though the localizing velocity is deviated from the real velocity, which demonstrates the effectiveness of the proposed method in AE source localization for plate-like structures. © 2013 Elsevier B.V. All rights reserved.

Wu W.,Zhejiang Ocean University | Wu W.,Chongqing Key Laboratory of Computational Intelligence | Gao C.,Zhejiang Ocean University | Li T.,Zhejiang Ocean University
Jisuanji Yanjiu yu Fazhan/Computer Research and Development | Year: 2014

Granular computing, which imitates human being's thinking, is an approach for knowledge representation and data mining. Its basic computing unit is called granule, and its objective is to establish effective computation models for dealing with large scale complex data and information. In order to study knowledge acquisition in ordered information systems with multi-granular labels, rough set approximations based on ordered granular labeled structures are explored. The concept of ordered labeled structures is first introduced. A dominance relation on the universe of discourse from an ordered labeled structure is also defined. Dominated labeled blocks determined by the dominance relation are constructed. Ordered lower approximations and ordered upper approximations, as well as ordered labeled lower approximations and ordered labeled upper approximations of sets based on dominance relations, are then proposed. Properties of approximation operators are examined. It is further proved that the qualities of lower and upper approximations of a set derived from an ordered labeled structure are a dual pair of necessity measure and possibility measure. Finally, multi-scale ordered granular labeled structures are defined and relationships among rough approximations with different scales induced from multi-scale ordered granular labeled structures are discussed. ©, 2014, Science Press. All right reserved.

Wang J.,Chongqing Key Laboratory of Computational Intelligence | Wang J.,Inha University | Lee C.-H.,Inha University
Journal of Central South University | Year: 2014

A virtual reconfigurable architecture (VRA)-based evolvable hardware is proposed for automatic synthesis of combinational logic circuits at gate-level. The proposed VRA is implemented by a Celoxica RC1000 peripheral component interconnect (PCI) board with an Xilinx Virtex xcv2000E field programmable gate array (FPGA). To improve the quality of the evolved circuits, the VRA works through a two-stage evolution: finding a functional circuit and minimizing the number of logic gates used in a feasible circuit. To optimize the algorithm performance in the two-stage evolutionary process and set free the user from the time-consuming process of mutation parameter tuning, a self-adaptive mutation rate control (SAMRC) scheme is introduced. In the evolutionary process, the mutation rate control parameters are encoded as additional genes in the chromosome and also undergo evolutionary operations. The efficiency of the proposed methodology is tested with the evolutions of a 4-bit even parity function, a 2-bit multiplier, and a 3-bit multiplier. The obtained results demonstrate that our scheme improves the evolutionary design of combinational logic circuits in terms of quality of the evolved circuit as well as the computational effort, when compared to the existing evolvable hardware approaches. © 2014 Central South University Press and Springer-Verlag Berlin Heidelberg.

Ji L.-H.,Chongqing University | Ji L.-H.,Chongqing Key Laboratory of Computational Intelligence | Liao X.-F.,Chongqing University | Chen X.,Chongqing University
Chinese Physics B | Year: 2013

In this paper the pinning consensus of multi-agent networks with arbitrary topology is investigated. Based on the properties of M-matrix, some criteria of pinning consensus are established for the continuous multi-agent network and the results show that the pinning consensus of the dynamical system depends on the smallest real part of the eigenvalue of the matrix which is composed of the Laplacian matrix of the multi-agent network and the pinning control gains. Meanwhile, the relevant work for the discrete-time system is studied and the corresponding criterion is also obtained. Particularly, the fundamental problem of pinning consensus, that is, what kind of node should be pinned, is investigated and the positive answers to this question are presented. Finally, the correctness of our theoretical findings is demonstrated by some numerical simulated examples. © 2013 Chinese Physical Society and IOP Publishing Ltd.

Zou L.,Dalian Jiaotong University | Yang X.,Dalian Jiaotong University | Sun Y.,Dalian Jiaotong University | Deng W.,Dalian Jiaotong University | Deng W.,Chongqing Key Laboratory of Computational Intelligence
Hanjie Xuebao/Transactions of the China Welding Institution | Year: 2015

An integrated model of rough set and neural network (RS_RBFNN) was proposed for predicting fatigue life of titanium alloy welded joints. The fatigue data were discretized by using the entropy-based algorithm, and the fatigue evaluation indices were reduced without information loss through a genetic algorithm. The reduced indices were used to develop the rules for fatigue life of welded joints and to train the RBF neural network. The rough set theory was used to determine the category of fatigue life for the test samples which matched the rules in the rule-base. The neural network was applied to those test samples which did not match any rules in the rule-base. Experimental results based on the fatigue data of titanium alloy show that the RS_RBFNN model for fatigue analysis of welded joints had improved fault tolerance and precision. Therefore this model is of practical significance for predicting fatigue life of titanium alloy welded joints. ©, 2015, Harbin Research Institute of Welding. All right reserved.

Loading Chongqing Key Laboratory of Computational Intelligence collaborators
Loading Chongqing Key Laboratory of Computational Intelligence collaborators