Table of Content

31 May 2020, Volume 38 Issue 3

Previous Issue    Next Issue

Big Data

Analysis for Psychological Scale Big Data Based on Improved Ising Model

YAO Rujing, YANG Lei, YANG Tao, HU Yingxin, TIAN Qiang, WU Ou

2020, 38(3):  339-352.  doi:10.3969/j.issn.0255-8297.2020.03.001

Asbtract ( 709 )   PDF (10580KB) ( 138 )  

References | Related Articles | Metrics

In recent years, the quantitative measurement of individual psychology has attracted more and more attention of administrators and researchers. It has become a new trend to use Ising model for analyzing the psychological scale data. In this paper, aiming at the shortcomings of the existing Ising model, we propose a multi-class Ising model and an ordinal Ising model. By applying them to analyze a large-scale psychological scales data set, we verify the performance of the two improved Ising models, construct complex networks of psychological scales for different groups of people, and conduct the comparisons of various indicators. Some meaningful conclusions have been drawn from the constructed psychological networks, and how machine learning and big data can be better involved in the analysis of psychological scale big data is discussed as well.

Topic-Specific Assessment Approach for Social Network Influence Evaluation

JIANG Qinyin, ZHANG Xi

2020, 38(3):  353-366.  doi:10.3969/j.issn.0255-8297.2020.03.002

Asbtract ( 584 )   PDF (5803KB) ( 229 )  

References | Related Articles | Metrics

Previous studies on user influence modeling in social networks mostly depend on user friendship network structures and retweeting behaviors. It lacks of the consideration of contents and topics of the tweets, which may also play important roles. In addition, taking the interaction among topics into account would facilitate a more accurate user influence modeling. In this paper, we propose a semi-supervised topic extraction method, which brings in a set of seed words during initialization and assigns these seed words higher weights than other words, to improve the effectiveness of topic extraction. To better model the user influence, we involve the interactions among topics, and combine the similarity of topics together with the similarity of users. Experimental results on real-world datasets demonstrate the effectiveness of our proposed methods.

Lightweight Phytoplankton Detection Network Based on Knowledge Distillation

ZHANG Tongtong, DONG Junyu, ZHAO Haoran, LI Qiong, SUN Xin

2020, 38(3):  367-376.  doi:10.3969/j.issn.0255-8297.2020.03.003

Asbtract ( 584 )   PDF (8937KB) ( 199 )  

References | Related Articles | Metrics

Object detection framework based on convolution neural network usually uses a very deep convolution neural network to extract object features before detection. However, its huge network structure leads to the reduction of detection speed, thus, the model can hardly achieve real-time object detection and be put into embedded devices. Address to the problem, this paper applies a knowledge distillation method to feature extraction network of object detection network to improve the performance of shallow feature extraction network. In this way, the model can ensure the same performance with a big reduction on computational load and model scale. Experimental results show that the detection accuracy of feature extraction networks employing distilled shallow network is 11.7% higher than that of networks without teacher’s guidance. Moreover, we build a phytoplankton dataset in this paper, which can not only be used for the evaluation of the performance of object detection algorithms, but also will be helpful to the development of phytoplankton microscopic vision technology.

Improved Faster R-CNN Algorithm and Its Application on Vehicle Detection

WEI Ziyang, ZHAO Zhihong, ZHAO Jingjiao

2020, 38(3):  377-387.  doi:10.3969/j.issn.0255-8297.2020.03.004

Asbtract ( 854 )   PDF (18463KB) ( 312 )  

References | Related Articles | Metrics

In order to obtain an initial candidate frame that conforms to the morphological characteristics of vehicles more accurately, a vehicle detection algorithm based on the improved Faster R-CNN model is proposed. First, the coordinate values of target frames are extracted to get width and height values of labeled boxes, and then K-means algorithm is used to cluster the width and height values of all boxes. Second, by resetting the anchor box size and the anchor box ratio of region proposal network (RPN) according to the coordinates of the cluster center point, the three sizes and three ratios of the Faster R-CNN can be improved. Finally, vehicle data of four types including cars, SUVs, buses and trucks are selected to train both the unimproved and the improved Faster R-CNN models. At the same time, the performance of the two models in vehicle detection and vehicle identification tasks are compared. Experimental results show that the improved Faster R-CNN model can achieve 84.69% detection accuracy, which is 3.12% higher than the original model. The algorithm effectively improves the missed detection and false detection problems, and shows high robustness in bad weather and complex background.

Research and Application of Improved CRNN Model in Classification of Alarm Texts

WANG Mengxuan, ZHANG Sheng, WANG Yue, LEI Ting, DU Wen

2020, 38(3):  388-400.  doi:10.3969/j.issn.0255-8297.2020.03.005

Asbtract ( 923 )   PDF (10161KB) ( 135 )  

References | Related Articles | Metrics

Aiming at classifying the police text descriptions of city’s public security for police stations, this paper builds a text classification of police descriptions based on the existing text classification methods used in other industries. By demonstrating the applicable occasions of common classification networks and their advantages and disadvantages, and combining with the text characteristics of the police case description data, a network structure based on Improved convolutional reccurrent neural network (CRNN) is proposed. The proposed structure provides an optimization key feature extraction process to make up the insufficiency of the existing model in the extraction of short-text feature. Through the comparison test between the proposed model and the existing common classification model, the proposed model not only shows an improved classification accuracy, 2%～3% higher than the existing model, but also provides effective guarantee on the relevance of local features of the data. The model can achieve accurate type classification of police descriptions, thus improving the automation efficiency of the police station.

Communication Engineering

Monitoring Microstructural Variations of Plant Tissues by Mueller Matrix Imaging

LIU Gang, ZHANG Yali, ZHAO Jingjing, WANG Chunhua

2020, 38(3):  401-409.  doi:10.3969/j.issn.0255-8297.2020.03.006

Asbtract ( 534 )   PDF (21656KB) ( 52 )  

References | Related Articles | Metrics

Mueller matrix imaging, which contents the full polarization information of samples under test, has become an important way to investigate the inner micro-structure of the sample. Based on dual rotating waveplates, a Mueller matrix polarization microscopy imaging system is designed to measure the Mueller matrix images of an anisotropic plant tissue, combining with a novel data processing algorithm. The Mueller matrix imaging and polar decomposition parameters are analyzed, and the changes of plant tissues are monitored by observing the imaging of decomposition parameters. It shows that Mueller matrix imaging are more sensitive to the micro-structure of plant samples, comparing with the traditional intensity images.

Research on Intelligent Sensing of Radio Signals in Cognitive Networks

HUANG Tangsen, LI Xiaowu, CAO Qingjiao

2020, 38(3):  410-418.  doi:10.3969/j.issn.0255-8297.2020.03.007

Asbtract ( 507 )   PDF (6689KB) ( 218 )  

References | Related Articles | Metrics

In the case of noise fluctuation, the performance of the radio signal detection needs to be improved. In this paper, a method for cognitive users to automatically adjust the detection threshold according to the changes of the radio environment is proposed. The fusion center applies coordinate search algorithm to provide the optimal control parameters to cognitive users. Cognitive users set the detection threshold according to the optimal parameters and autonomously learn the optimal threshold for a specific radio environment. In addition, by taking a full consideration of the distinctions and sensing contributions of cognitive users, a new weight calculation method to reflect the distinctions is designed. Simulation results show that the spectrum sensing method has excellent robustness to noise fluctuation. It performs a much higher detection probability than the traditional sensing methods as signal-to-noise ratio (SNR) is below -15 dB.

Random Interpolation Method for Data Hiding in Encrypted Images

SUN Ronghai, SHI Linfu, YU Chunqiang, LAO Huan, TANG Zhenjun

2020, 38(3):  419-430.  doi:10.3969/j.issn.0255-8297.2020.03.008

Asbtract ( 659 )   PDF (23975KB) ( 128 )  

References | Related Articles | Metrics

When traditional data interpolation methods are applied to data hiding in encrypted images, they will reduce security of the data hiding system. To address this problem, we exploit random weight strategy to design a random interpolation method for encrypted images. The proposed method firstly generates the initial interpolated image, which is twice the size of the encrypted image. For the pixels in the odd rows and the odd columns of the interpolated image, the proposed method fills them with the corresponding pixels of the encrypted image. For the rest pixels of the interpolated image, the proposed method generates random values by pseudo-random function and calculates the interpolated results in terms of their detailed locations. Experimental results show that the histograms of interpolated encrypted images calculated by our random interpolation are approximately uniformly distributed. Comparison results demonstrate that our random interpolation outperforms three popular existing interpolation methods.

Optimization Design of Polar Codes Based on MI Heterogeneity in MLC NAND Flash Channel

ZHANG Siqi, KONG Lingjun, ZHANG Shunwai, ZHANG Nan

2020, 38(3):  431-440.  doi:10.3969/j.issn.0255-8297.2020.03.009

Asbtract ( 642 )   PDF (7544KB) ( 189 )  

References | Related Articles | Metrics

In order to further improve the durability and reliability of multi-level-cell (MLC) flash memory, a polar code optimization method based on mutual information (MI) heterogeneity in the MLC flash channel is proposed. By exploiting the differences of log-likelihood ratio (LLR) distribution between MLC flash channels and AWGN (additive white Gaussian noise) channels, and employing MI re-fitting for obeying Gaussian distribution in AWGN channels, the method obtains its equivalent variance of AWGN channels. Thereafter, the polar code optimization design in the high-density storage system is performed according to the obtained new variance. This paper also analyzes the effects of other different polar code construction methods on the error correction of multi-level memory cells, and compares them with the proposed construction method. Simulation results show that the optimization method is better than the traditional construction methods in AWGN channels. It improves bit error rate (BER) by more than 2 orders of magnitudes compared with Monte-Carlo method when program-and-erase (P/E) cycles is 21 000, and it can increase the lifetime of MLC flash memory up to 6 800 P/E cycles at the BER of 2×10^-5.

Robust Coverless Data Hiding Based on Texture Classification and Synthesis

SI Guangwen, QIN Chuan, YAO Heng, HAN Yanfang, ZHANG Zhichao

2020, 38(3):  441-454.  doi:10.3969/j.issn.0255-8297.2020.03.010

Asbtract ( 523 )   PDF (49959KB) ( 371 )  

References | Related Articles | Metrics

Aiming at the problem that the embedding rate and robustness of coverless information hiding cannot be well balanced, a robust coverless information hiding scheme based on texture feature classification and synthesis is proposed. In this scheme, texture image features are extracted with spatial pyramid algorithm, and classification models are obtained by supervised classification training. A mapping dictionary is constructed according to the classification of image blocks and different location information. The sender chooses image blocks based on secret information and combines all image blocks into one image according to public key, then generates complex lines through reversible deformation. The texture image can be restored to image blocks by using the key, and the classification model is used to identify the classification of image blocks and determine the location information. Finally, secret information is extracted based on the mapping dictionary. Experimental results show that the proposed scheme has strong robustness against JPEG compression, Gaussian noise, salt and pepper noise and other typical attacks, and the embedding capacity can be further improved with the increase of image category number.

TOA Estimation Based on Narrowband Interference Mitigation Technique

JING Yanliang, LUO Xuetao, WANG Xue, NIE Hong

2020, 38(3):  455-465.  doi:10.3969/j.issn.0255-8297.2020.03.011

Asbtract ( 448 )   PDF (8494KB) ( 183 )  

References | Related Articles | Metrics

Focused on the low accuracy of time of arrive (TOA) estimation in traditional energy detection methods with the presence of narrowband interference, a square filtering technique combining square-law device and band-pass filter is used to eliminate the influence of narrowband interference on TOA estimation in this paper. Then the TOA estimation algorithm is applied to get output estimation results with the IEEE 802.15.4 CM3 channels and CM4 channels. Theoretical analysis and simulation results show that the proposed ED method using square filtering technique has higher TOA estimation accuracy than the traditional ED method in the presence of strong NBI. In the line-of-sight (LOS) environment, the TOA estimation accuracy can be improved from 2.8 ns to 0.5 ns after applying the square filtering scheme, and in non-line-of-sight (NLOS) environment, it can be improved from 6 ns to 1 ns.

Signal and Information Processing

Urban Spatial Form Analysis of GBA Based on “LJ1-01” Nighttime Light Remote Sensing Images

ZHANG Yuxin, LI Xi, SONG Yang, LI Changhui

2020, 38(3):  466-477.  doi:10.3969/j.issn.0255-8297.2020.03.012

Asbtract ( 21412 )   PDF (15211KB) ( 2538 )  

References | Related Articles | Metrics

In this paper, “LJ1-01” nighttime light (NTL) images are used to extract urban built-up areas of Guangdong-Hong Kong-Macao greater bay area (GBA) by employing simple threshold method and vegetation adjusted NTL urban index (VANUI). Comparing the two methods, VANUI is capable of reducing the over-saturations in LJ1-01 images, thus reducing misclassifications caused by “blooming”. The landscape indices of the urban builtup areas in GBA are calculated and analyzed. It is found that there are different patterns in distribution of built-up areas in different cities. As the cores of the development of GBA, Guangzhou, Shenzhen and Hong Kong have expanding urban areas. The urban built-up areas, like Dongguan, Foshan, Macao, Zhongshan and Zhuhai, are highly compact and integrated in spatial distribution. And the urban built-up areas of less developed cities, including Zhaoqing, Jiangmen and Huizhou, are small and separated. This study proves that the LJ1-01 nighttime light images can effectively reveal the urban spatial form of GBA, providing a basis for urban planning policy of GBA.

Computer Science and Applications

Collaborative Filtering Recommendation Model Based on Hybrid Neural Network

MA Xin, WU Yun, LU Zeguang

2020, 38(3):  478-487.  doi:10.3969/j.issn.0255-8297.2020.03.013

Asbtract ( 782 )   PDF (7717KB) ( 279 )  

References | Related Articles | Metrics

In the recommendation system, data sparsity is one of the important factors that seriously affect the accuracy of recommendation results. Aiming at the data sparsity, this paper proposes a hybrid neural network collaborative filtering score prediction model convolutional-denosing auto-encoder (CDAE) to perform prediction scoring for solving the problem of data sparsity. The CDAE model combines a convolutional neural network (CNN) and a denoising auto-encoder (DAE) neural network. Firstly, the vectorized user review data is trained by the CNN to obtain a user feature vector matrix. Secondly, the user feature vector matrix is used as the initial weight of the DAE neural network, and the user-item prediction score is obtained by training the auto-encoder neural network in combination with user rating data. Accordingly, user-based collaborative filtering recommendations can be made. In the paper, the proposed convolutional-denosing auto-encoder collaborative filtering (CDAECF) model is experimentally verified by comparing with the experimental data set of movielens-1M. Experiment results prove that the CDAECF model can effectively combine implicit and explicit feedback data, and performs a higher recommendation accuracy rate.

Data Center Energy Consumption Measurement Based on Virtual Machine

CHEN Jun, LI Ya, ZHANG Jie

2020, 38(3):  488-495.  doi:10.3969/j.issn.0255-8297.2020.03.014

Asbtract ( 484 )   PDF (5664KB) ( 110 )  

References | Related Articles | Metrics

Based on computation intensive mode and I/O intensive mode and combined with the running state parameters of typical equipment, a mathematical model for the dynamic energy consumption measurement of virtual machine granularity is proposed for the experimental environment of data center. The power consumption model takes CPU usage and CPU frequency in the computation intensive mode and takes the total read and write bytes of hard disk and memory in the I/O intensive mode to measure the power consumption, accordingly, deriving the energy consumption of the cloud platform by integrating the power consumption. Compared with the conventional method, the experimental method further refines the measurement granularity, and obtains an average precision of 0.062 5 as it measures energy consumption of test nodes with Wordcount computing task and Sort computing task. The proposed method performs a finer-grained energy consumption measurement than conventional methods with the same measurement accuracy.

Text Detection in Natural Scene Based on Visual Attention Model and Multi-scale MSER

WANG Daqian, CUI Rongyi, JIN Jingxuan

2020, 38(3):  496-506.  doi:10.3969/j.issn.0255-8297.2020.03.015

Asbtract ( 564 )   PDF (18664KB) ( 227 )  

References | Related Articles | Metrics

Aiming at the low accuracy of current natural image detection algorithms, which is induced by the influence of illumination, complex background, multi-language and variety of font and size, a natural image text detection algorithm based on Itti visual salience model and multi-scale maximally stable extremal region (MSER) is proposed. First, we extract a text feature map from the improved Itti visual attention model, and obtain the text saliency maps of different scales by using different combination strategies. Then three kinds of text candidate regions can be figured out by combining with the multiscale MSER region, and text lines can be obtained by the text candidate regions according to these geometric rules of text and generated text boxes. Finally, the text area is obtained by using the random forest classifier to remove the non-text regions. Experimental results show that the text detection algorithm proposed in this paper has high detection accuracy and robustness under the influences of multi-language, text distortion and variety of size.