Journal of South China University of Technology(Natural Science Edition)

A Multi-feature Fusion-based Algorithm for Real-time Single Object Tracking

YANG Xiaowei HUANG Yingting

2019, 47(6): 1-9. doi:10.12141/j.issn.1000-565X.180439

Asbtract ( 400 )

PDF (3580KB) ( 683 )

References | Related Articles | Metrics

A multi-feature fusion-based algorithm was proposed for real-time single object tracking by using the bi- lateral weighted least squares fuzzy support vector machine． In the proposed algorithm，the bilateral weighted least squares fuzzy support vector machine was trained with local HOG feature and global color feature respectively，and object tracking was achieved by using the linear combination of the two classifiers． For the local HOG feature-based classifier，the multiple base samples based correlation filtering was adopted to overcome matrix inversion． For the global color feature based-classifier，the unique thermal coding were used to encode the feature to achieve fast cal- culation． The experimental results on the public data sets show that compared with the state-of-the-art trackers， the proposed algorithm shows better tracking performance in deformation，fast motion，motion blur，and in-plane/ out-of-plane rotation．

A Sentence Sentiment Classification Method with POS and Attention

SU Jindian YU Shanshan LI Pengfei

2019, 47(6): 10-17,30. doi:10.12141/j.issn.1000-565X.180486

Asbtract ( 336 )

PDF (957KB) ( 585 )

References | Related Articles | Metrics

Aiming at the problem that most of existing LSTM-based methods for sentence sentiment classification don’t take into account the part-of-speech (POS) information of words，a neural network model，PALSTM， which combines POS and self-attention mechanism，was proposed and applied to sentence sentiment classification． Firstly，PALSTM used pre-trained word vectors and POS tagging tool to give the semantic and POS word vector rep- resentations of words in the sentences，and then took them as the inputs of a LSTM so as to capture the long-term dependence of words on content and part-of-speech，which effectively compensates for the common LSTM networks relying solely on the co-occurrence information of words in pre-trained word vectors． Secondly，the self-attention mechanism was used to learn the position information about words in the sentences and build the corresponding po- sition weight matrix，which yields the final semantic representations of sentences． Finally，the results was classi- fied and outputted via a multi-layer perceptron． The experiments show that PALSTM outperforms common LSTM and attentional LSTM models on some open corpus，i. e． Movie Reviews，Internet Movie Database，Stanford Sen- timent Treebank binary and fine-grained classification．

An Efficient Incremental Attribute Reduction for Complex Information Systems

DUAN Hailing WANG Guangqiong

2019, 47(6): 18-30. doi:10.12141/j.issn.1000-565X.180409

Asbtract ( 300 )

PDF (1795KB) ( 561 )

References | Related Articles | Metrics

An incremental attribute reduction algorithm for complex information systems with discrete and continu- ous mixed attributes was proposed． Firstly，the knowledge granulation in the granular computing model was extend- ed under the mixed information system，the neighborhood knowledge granulation was proposed，and a non-incre- mental attribute reduction algorithm based on the neighborhood knowledge granulation was constructed． Then，the incremental computation of neighborhood knowledge granulation with the increase of objects was studied under the mixed information system，and the efficiency of the computation was proved theoretically． Finally，an incremental attribute reduction algorithm for mixed information systems based on neighborhood knowledge granulation was pro- posed． Experimental results of UCI datasets show that the proposed algorithm has high incremental attribute reduc- tion performance in mixed information systems．

Automatic Generation of Multi-modules ROP Based on Static Instructions Assignment

HUANG Ning HUANG Shuguang HUANG Hui DENG Zhaokun

2019, 47(6): 31-38. doi:10.12141/j.issn.1000-565X.180255

Asbtract ( 353 )

PDF (1197KB) ( 679 )

References | Related Articles | Metrics

Return Oriented Programming (ROP) is a kind of technology to bypass the Data Execution Prevention (DEP)． Existing technologies for automatic ROP generation can not optimize the progress of instructions assign- ment for modules switching in multi-modules ROP payload，leading to the problem of large amounts of memory space are occupied． In order to solve this problem，a new static instructions assignment rule SIA for ROP modules switching was designed based on the automatic ROP generation system Q． SIA constructs intermediate sequence of instructions by static instructions assignment，and finds the address and modifies the pointers by dynamic data fill- ing． Experimental results show that，compared with the existing technologies，the multi-modules ROP payload gen- erated by SIA needs less memory space so that improve the practicability of ROP．

A Multi-level Deep Convolutional Neural Network for Image Emotion Classification

WANG Weining LI Lemin HUANG Jiexiong LUO Jiebo XU Xiangmin

2019, 47(6): 39-50. doi:10.12141/j.issn.1000-565X.180059

Asbtract ( 494 )

PDF (5163KB) ( 871 )

References | Related Articles | Metrics

Emotion classification of images is a challenging task regarding the complexity of various images and the subjectivity of human’s emotion perception． Most existing deep learning methods didn’t consider image prior in- formation fully． A new multi-level deep convolutional neural network was proposed to predict the emotion based on the multi-level prior information from global and local view． Extensive experiments on both the large scale and small scale emotion datasets demonstrate the effectiveness of our method． The average classification accuracy of our meth- od is 2. 8% higher than the state-of-art method，especially 15% higher in the category“disgust”．

Method of Network Compression and Hardware Acceleration Based on Tiny-yolo

HUANG Zhiyong WU Haihua YU Zhi ZHONG Yuanhong

2019, 47(6): 51-56. doi:10.12141/j.issn.1000-565X.180360

Asbtract ( 779 )

PDF (1096KB) ( 659 )

References | Related Articles | Metrics

Existing works based on Tiny-yolo often need large-scale network model，occupy more memories，rely on massive calculation and are not easy to deploy on embedded devices． To solve these problems，an efficient optimi- zation method on network compression and hardware acceleration was proposed． Firstly，connections which have less contribution to the network was pruned after analyzing the network connection relationship and sparse storage was adopted for the clipped weight matrix to reduce the memory consumption． Secondly，memory footprint and com- putational complexity within the guaranteed accuracy error was further reduced through quantifying the weight data and changing the number of digits． Finally，according to the characteristics of the Tiny-yolo network structure，a deep parallel-stream FPGA acceleration optimization scheme was proposed and the hardware acceleration of the Ti- ny-yolo network computation was achieved． Experiments demonstrate that the purposed method based on network pruning and quantization can achieve about 36X compression for network model and approximately 7X speedup compared with CPU by hardware acceleration．

Decision Analysis Method for Operation and Maintenance Management of Power Equipment Based on Data Mining

CAI Zexiang MA Guolong SUN Yuyan HUANG Yuhan

2019, 47(6): 57-64,71. doi:10.12141/j.issn.1000-565X.180399

Asbtract ( 357 )

PDF (1678KB) ( 622 )

References | Related Articles | Metrics

The operation and maintenance management of the power equipment (PE) mainly includes fault analy- sis，active early-warning and differentiated operation and maintenance． In the context of massive data with multiple time scales and multiple time and space dimensions in the process of grid operation，data mining technology was ap- plied for PE operation and maintenance management． The one-dimensional fault feature was extracted from fault in- formation by K-means clustering algorithm． Then，Apriori algorithm was employed to mine association rules of dif- ferent failure modes and establish key performance matrix． The spatial-temporal characteristics were analyzed based on high-dimensional random matrix theory (ＲMT)． Afterwards，one-dimensional and multi-dimensional fault fea- tures were combined based on D-S evidence theory so that the fault diagnosis criteria of PE was obtained． At the same time，comprehensively considering the PE operating state and the variation for power users，health index and importance index of equipment were established，which could help to significantly reduce the decision-making risk of PE operation and maintenance． The result of simulation proves the effectiveness of the proposed method．

Control Analysis of Fully-enclosed Wave Power Generation System Based on IDA-PBC

YANG Jinming JIANG Yuan XIE Zekun HUANG Xiuxiu

2019, 47(6): 65-71. doi:10.12141/j.issn.1000-565X.180298

Asbtract ( 305 )

PDF (1900KB) ( 607 )

References | Related Articles | Metrics

For the complex characteristics of wave and the fully-enclosed wave power generation system，a control algorithm which is robust to wave changes was proposed based on interconnection and damping assignment passivity based control (IDA-PBC)． Firstly，a Euler-Lagrange (EL) model was established for fully-enclosed wave power generation system，and the model was decomposed based on the structure and dynamic characteristics of the sys- tem． Then，according to the Lyapunov stability and damping injection，a passivity-based control law was acquired． In addition，in order to maximize power generation，the desired control objectives were deduced． Lastly，a fully-en- closed wave power generation model in Matlab/Simulink simulation environment was built，and simulation analysis of passivity based control was carried out． The result shows that asymptotically tracking control objectives can be a- chieved，and the control law is also effective in case of wave changes．

Load Aggregators Joint Scheduling Strategy Considering Multiple Demand Responses to Physical Constraints

DONG Ping WU Zidong CHEN Yongchun DUAN Qinggang

2019, 47(6): 72-79,93. doi:10.12141/j.issn.1000-565X.180515

Asbtract ( 327 )

PDF (1694KB) ( 598 )

References | Related Articles | Metrics

With the advancement of electricity market reforms，load aggregators have gained significant opportuni- ties due to their important role as a bridge between connected systems and end users． The physical constraints and the daily load cap constraints of Load Curtailment，Load Shifting，Utilizing Onsite Generation was considered in the load integrator scheduling optimization． The branch and bound method was used to ensure that the optimal solution was found，so as to determine the optimal scheduling strategy of the load integrator in the energy market． Finally， examples were given to analyze the impact of load integrator scheduling on physical constraints and joint scheduling constraints，which provide a theoretical basis for the actual scheduling of load aggregators．

Analysis of Dynamic Characteristics and Parameter Influence of High Energy Pipe Whip in a Nuclear Power Plant

ZHAO Wensheng LUO Xiangyu WANG Pengfei JIANG Jin SHANG Yibo

2019, 47(6): 80-86. doi:10.12141/j.issn.1000-565X.180491

Asbtract ( 408 )

PDF (1671KB) ( 633 )

References | Related Articles | Metrics

The pipe whip behaviour of a high energy pipe in a nuclear power plant under assumption of postulated ruptures was investigated by using the finite element method． A three-dimensional finite element model of the pipe and U-bolt restraint was built，the structural parameters of them were tested experimentally，the thrust force caused by jet flow was calculated based on the jet thrust equation，and the accuracy of the model was verified with reported experimental results． Then，the time traces of restraint force，displacement，velocity and the energy of the system were analyzed． Finally，the effects of overhang length，friction coefficient and clearance on the pipe whip behaviour of the pipe-restraints system were discussed． The research results show that the pipe oscillates with small amplitude after impacting on the U-bolt restraint，the impact velocity increases suddenly to a maximum val- ue，and then decreases rapidly to zero，and the impact force undergoes 150Hz periodic oscillation． The work done by thrust force is almost converted to strain energy． The vertical displacement of pipe increases with increasing o- verhang length，whipping force increases with increasing whipping friction coefficient and increasing primary clea- rance．

Study on Horizontal Curve Index Satisfying the Safe Passing of Articulated Train

YANG Yonghong DENG Zhuo

2019, 47(6): 87-93. doi:10.12141/j.issn.1000-565X.180467

Asbtract ( 468 )

PDF (1200KB) ( 498 )

References | Related Articles | Metrics

The articulated train is a very important transportation tool for special equipment in China’s infrastruc- ture construction． It has high requirements for the passability and safety of road design． However，the current specification doesn’t fully consider the safe traffic demand of articulated trains． The design standard of such spe- cial roads needs to be improved． According to the principle of rear wheel steering of the articulated train，and con- sidering the vehicle size and the horizontal curve index，the wheel trajectory was analyzed and the road widening value model was improved． It proposed a method of selecting circular curve radius limited by the road widening value when the filling and excavating conditions are limited． Taking the road widening less than 3 meters as an ex- ample，the method suggested the recommended values of the turning radius of the two types of vehicles “steering rear wheel”and “non-steering rear wheel”under different wheelbase and corner conditions are 20 ～80m． In or- der to ensure the safety of the vehicle in the process of turning and the inside and outside of the road，two safety design indexes，namely “lateral clear distance”and “over-length permissible value”were proposed to analyze the demand and difference types of vehicles． The research results will provide engineers with accurate design refer- ences，reduce the cost of road reconstruction and expansion，and improve the transportation safety of large car- goes．

Analysis of 3D Reconstruction of Ancient Bridge Based on Consumer-grade Unmanned Aerial Vehicle

MA Rujin DONG Yiqing PAN Zichao PAN Yue

2019, 47(6): 94-100,126. doi:10.12141/j.issn.1000-565X.180390

Asbtract ( 441 )

PDF (1856KB) ( 654 )

References | Related Articles | Metrics

There are a large number of ancient bridges in China． However，documentations of ancient bridges are scarce，and the details of them are rare． To provide solution to the lack of data on ancient bridges，3D recon- struction based on consumer-grade unmanned aerial vehicle was proposed． Configuration of consumer-grade un- manned aerial vehicle was first introduced． Parameters of the camera were resolved． Reasonable flight path was then considered． Algorithms of 3D reconstruction was studied． Hongde Bridge in Shanghai was experimented，re- sulting in an optimized 3D model with refine details． The results indicate that 3D reconstruction of ancient bridge based on consumer-grade unmanned aerial vehicle is feasible，complete and of high-fidelity． Thus，the model is particularly suitable for documentation of ancient bridge，leading to digitization and informatization of ancient bridges．

Mechanical Properties of Two Kinds of Lined Reinforced Concrete Pipe Culvert

LI Baijian ZHU Liangsheng FU Xinsha LI Yong

2019, 47(6): 101-107. doi:10.12141/j.issn.1000-565X.180320

Asbtract ( 424 )

PDF (1610KB) ( 837 )

References | Related Articles | Metrics

With the increase of highway extension and expansion project，a large number of concrete pipe culverts need to be reinforced，rebuilt and extended． In order to study the mechanical properties of the reinforced concrete (RC) pipes with slip-liners，the RC pipes (inner diameter of 1. 2 m) reinforced by HDPE pipe and shape steel bracket were studied by two-point loading experiments，and the load versus diameter change curve was obtained． Through analyzing the load-diameter change curve，it finds that the load-carrying capacity of the pipe is improved after reinforcement，the HDPE pipe increases 1. 50 times and the shape steel bracket pipe increases 1. 31 times， whereas the corresponding stiffness and the ductility of the rehabilitated pipes are markedly unchanged． It is proved that，according to the experimental phenomenon and the calculation results，the rehabilitated pipe is a“pipe with- in a pipe system”，so the load sharing depends on the stiffness coefficient of the pipe． The formula proposed in this study can be used to calculate the ultimate load-carrying capacity of the rehabilitated pipes，and the error is less than 5%． The short-term or long-term stiffness should be used to calculate the stiffness factor of the RC pipe． when the RC pipe is damaged，the grout plays an important role in distributing loads，and the liner shares only a small part of the load． If the rehabilitated pipe contains only the RC pipe and the liner，the load is shared between the RC pipe and the liner，and when the RC pipe is completely damaged，the liner carries most of the load．

Evaluation Method of Bridge Risk Probability in Vehicle Fire Accidents

ZHANG Xiaodong MA Rujin CHEN Airong

2019, 47(6): 108-118. doi:10.12141/j.issn.1000-565X.180575

Asbtract ( 271 )

PDF (2212KB) ( 632 )

References | Related Articles | Metrics

Based on the vulnerability analysis and the hazard analysis，a numerical calculation method for risk probability of bridge exposed to fires caused by vehicles was developed． According to the statistical data of vehicle fire accidents，the probability density function of maxim heat release rate was estimated using the principle of maxi- mum entropy． A vulnerability analysis method of bridge member based on RSM-MCS was proposed，and a vulner- ability analysis method of structural system was given by the research about structural system，correlation and dam- age criterion of system． The risk probability of a simply supported beam bridge under fires caused by vehicles was provided to illustrate the method． The results show that the suggested method can effectively evaluate the fire resist- ance of a bridge．

Analysis of Bus Users’Behavior Based on Latent Class

LIU Jianrong LIU Zhiwei

2019, 47(6): 119-126. doi:10.12141/j.issn.1000-565X.180412

Asbtract ( 304 )

PDF (839KB) ( 742 )

References | Related Articles | Metrics

Different types of bus users have different opinions on public transport and their willingness to use public transport． Different types of bus users have different opinions on public transport and their willingness to use public transport． Based on bus users’satisfaction and their traveling behavior，bus users were classified with the latent class regression model． Then based on the classification，factors that affect the overall satisfaction and the willing- ness to traveling by bus were analyzed among different subgroup． This result shows that bus users can be classified into two subgroups: the occasional bus users and the loyal bus users，and that private car ownership and income have a great influence on the classification． That is users with private car are more likely to be occasional bus us- ers，and along with income increases，the user is more likely to be a occasional bus user． The analysis result of factors affecting satisfaction shows that the evaluation on bus speed，time reliability，waiting time and ticket price have stronger influence on occasional bus users’overall satisfaction than that on loyal bus users’． As to the effect of recommendation to use bus，it is found that the overall satisfaction of occasional bus users significantly affect the recommendation effect，and the recommendation effect increases as the overall satisfaction increases． Also，travel- ers’traveling by bus decreases after they have private cars and the occasional bus users’frequency of traveling by bus increases as the overall satisfaction increases．

Study on Travel Time Reliability Considering Route Travel Time Boundary

ZHAO Lei GUAN Hongzhi ZHANG Xinjie ZHAO Pengfei WANG Pengfei

2019, 47(6): 127-135. doi:10.12141/j.issn.1000-565X.180466

Asbtract ( 275 )

PDF (933KB) ( 501 )

References | Related Articles | Metrics

Through assuming the route-free flow travel time as the lower bound of route travel time，the travel time budget model with truncated normal distribution was proposed． And the model was compared with the travel time budget model with general normal distribution． In the multi-class user stochastic network，travelers were assumed to be heterogeneous with respect to their desired on-time arrival probabilities． Then the reliability-based multi-class stochastic user equilibrium model was developed，its equivalent variational inequality formulation was given． The results show that the budget time of travelers under truncated normal distribution is larger than that under general normal distribution． In this case，the risk aversion levels of travelers are context-based adaptive． In addition，the results also demonstrate that the punctual travelers tend to choose the routes with the better reliability． Neverthe- less，the risky travelers prefer to choose the routes with the less average travel time．

Feasibility Evaluation of Dissimilarity Algorithm and Adjacency Method for Constructing Evolutionary Tree

LIU Xuemei LI Wen HUANG Guanda HUANG Tianlai YE Yuzhong XU Guohao

2019, 47(6): 136-141,148. doi:10.12141/j.issn.1000-565X.180313

Asbtract ( 595 )

PDF (2116KB) ( 880 )

References | Related Articles | Metrics

In phylogenetic，the algorithm of constructing evolutionary tree is the basis of studying evolutionary rela- tionship． The feasibility of constructing evolutionary tree based on k-mer dissimilarity algorithm d2S and adjacency method NJ was discussed and verified． The evolutionary tree was constructed for 12 16S rRNA and 36 bacterio- phage molecule sequences in two evolutionary scenarios (k =4，6，8，10)． Three types of 16S rRNA were clear- ly classified according to archaea，bacteria and eukaryotes，and the results are basically consistent with Woese cata- logue． At the same time，four types of bacteriophages can be distinguished according to the original evolutionary relationship，and the influence of parameter k is small，so the classification effect is very good． This demonstrates the powerful function of the dissimilarity algorithm d2S． Using the dissimilarity algorithm d2S and the adjacency method (NJ) to construct the evolutionary tree will provide a new algorithm and idea for the phylogenetic relation- ship of the system．

Comparisons of Humanized AGR2 Monoclonal Antibodies Expressions in Different Host Cells

XIE Qiuling TANG Jie HUANG Jiahui LU Jia LIU Chenxuexuan

2019, 47(6): 142-148. doi:10.12141/j.issn.1000-565X.180454

Asbtract ( 445 )

PDF (1220KB) ( 504 )

References | Related Articles | Metrics

Anterior gradient-2 (AGR2) gene is expressed in many organs of human body，which is closely related to tumor growth，metastasis，invasion and drug resistance． In this study，we successfully developed a humanized monoclonal antibody，rhAGR2-mAb，which can specifically binds with AGR2 to be a new anti-tumor drug． In or- der to evaluate the activity of this antibody，rhAGR2-mAb was expressed with HEK293F cells and CHO cells by transient transfections and this protein with 95% purity was obtained． Then the properties of rhAGR2-mAb inclu- ding molecular weights，isoelectric points，glycan structures expressed in CHO cells and HEK cells were com- pared． The results show that there are no detected differences between HEK293F cells and CHO cells on molecular weights，isoelectric points，glycosylation types． However，the product of two expressions are found different in glycosylation types proportion，which makes them differ in affinities with AGR2． This provides data for selecting a suitable expression system to transiently and stably express rhAGR2-mAb in the future．

Table of Content