Table of Content

10 September 2019, Volume 39 Issue 9

Previous Issue    Next Issue

Artificial intelligence

Adversarial negative sample generation for knowledge representation learning

ZHANG Zhao, JI Jianmin, CHEN Xiaoping

2019, 39(9):  2489-2493.  DOI: 10.11772/j.issn.1001-9081.2019020357

Asbtract ( )   PDF (827KB) ( )  

References | Related Articles | Metrics

Knowledge graph embedding is to embed symbolic relations and entities of the knowledge graph into low dimensional continuous vector space. Despite the requirement of negative samples for training knowledge graph embedding models, only positive examples are stored in the form of triplets in most knowledge graphs. Moreover, negative samples generated by negative sampling of conventional knowledge graph embedding methods are easy to be discriminated by the model and contribute less and less as the training going on. To address this problem, an Adversarial Negative Generator (ANG) model was proposed. The generator applied the encoder-decoder pipeline, the encoder readed in positive triplets whose head or tail entities were replaced as context information, and then the decoder filled the replaced entity with the triplet using the encoding information provided by the encoder, so as to generate negative samples. Several existing knowledge graph embedding models were used to play an adversarial game with the proposed generator to optimize the knowledge representation vectors. By comparing with existing knowledge graph embedding models, it can be seen that the proposed method has better mean ranking of link prediction and more accurate triple classification result on FB15K237, WN18 and WN18RR datasets.

Gradual multi-kernel learning method for concept drift

BAI Dongying, YI Yaxing, WANG Qingchao, YU Zhiyong

2019, 39(9):  2494-2498.  DOI: 10.11772/j.issn.1001-9081.2019020299

Asbtract ( )   PDF (717KB) ( )  

References | Related Articles | Metrics

Aiming at the concept drift problem, a classification learning model with the characteristics of data changing progressively over time was constructed, and a Gradual Multiple Kenerl Learning method (G-MKL) based on Gradual Support Vector Machine (G-SVM) was proposed. In this method, with Support Vector Machine (SVM) used as the basic classifier, multi-interval sub-classifier coupling training was carried out and the incremental method of constraining sub-classifier was used to adapt the model to the gradual change of data. Finally, multiple kernels were integrated into SVM solution framework in a linear combination manner. This method integrated the advantages of different kernel functions and greatly improved the adaptability and validity of the model. Finally, the comparison experiments between the proposed algorithm and several classical algorithms were carried out on the simulated and real datasets with gradual characteristics, verifying the effectiveness of the proposed algorithm in dealing with non-stationary data problems.

Cost-sensitive active learning through farthest distance sum sampling

REN Jie, MIN Fan, WANG Min

2019, 39(9):  2499-2504.  DOI: 10.11772/j.issn.1001-9081.2019020763

Asbtract ( )   PDF (862KB) ( )  

References | Related Articles | Metrics

Active learning aims to reduce expert labeling through man-machine interaction, while cost-sensitive active learning focuses on balancing labeling and misclassification costs. Based on Three-Way Decision (3WD) methodology and Label Uniform Distribution (LUD) model, a Cost-sensitive Active learning through the Farthest distance sum Sampling (CAFS) algorithm was proposed. Firstly, the farthest total distance sampling strategy was designed to query the labels of representative samples. Secondly, LUD model and cost function were used to calculate the expected sampling number. Finally, k-Means algorithm was employed to split blocks obtained different labels. In CAFS, 3WD methodology was adopted in the iterative process of label query, instance prediction, and block splitting, until all instances were processed. The learning process was controlled by the cost minimization objective. Results on 9 public datasets show that CAFS has lower average cost compared with 11 mainstream algorithms.

Semantic segmentation method of road environment combined semantic boundary information

SONG Xiaona, RUI Ting, WANG Xinqing

2019, 39(9):  2505-2510.  DOI: 10.11772/j.issn.1001-9081.2019030488

Asbtract ( )   PDF (1018KB) ( )  

References | Related Articles | Metrics

Semantic segmentation is an important method to interpret the road semantic environment. The convolution, pooling and deconvolution in semantic segmentation of deep learning result in blur and discontinuous segmentation boundary, missing and wrong segmentation of small objects. These influence the outcome of segmentation and reduce the accuracy of segmentation. To deal with the problems above, a new semantic segmentation method combined semantic boundary information was proposed. Firstly, a subnet of semantic boundary detection was built in the deep model of semantic segmentation, and the feature sharing layers in the network were used to transfer the semantic boundary information learned in the semantic boundary detection subnet to the semantic segmentation network. Then, a new cost function of the model was defined according to the tasks of semantic boundary detection and semantic segmentation. The model was able to accomplish two tasks simultaneously and improve the descriptive ability of object boundary and the quality of semantic segmentation. Finally, the method was verified on the Cityscapes dataset. The experimental results demonstrate that the accuracy of the method proposed is improved by 2.9% compared to SegNet and is improved by 1.3% compared to ENet. It can overcome the problems in semantic segmentation such as discontinous segmentation, blur boundary of object, missing and wrong segmentation of small objects and low accuracy of segmentation.

Biogeography-based optimization algorithms based on improved migration rate models

WANG Yaping, ZHANG Zhengjun, YAN Zihan, JIN Yazhou

2019, 39(9):  2511-2516.  DOI: 10.11772/j.issn.1001-9081.2019020325

Asbtract ( )   PDF (713KB) ( )  

References | Related Articles | Metrics

Biogeography-Based Optimization (BBO) algorithm updates habitats through migration and mutation continuously to find the optimal solution, and the migration model affects the performance of the algorithm significantly. In view of the problem of insufficient adaptability of the linear migration model used in the original BBO algorithm, three nonlinear migration models were proposed. These models are based on Logistic function, cubic polynomial function and hyperbolic tangent function respectively. Optimization experiments were carried out on 17 typical benchmark functions, and results show that the migration model based on hyperbolic tangent function performs better than the linear migration model of original BBO algorithm and cosine migration model with good performance of improved algorithm. Stability test shows that the migration model based on hyperbolic tangent function performs better than the original linear migration model with different mutation rates on most test functions. The model satisfies the diversity of the solutions, and better adapts to the nonlinear migration problem with improved search ability.

Global optimal path planning for robots with improved A^* algorithm

WANG Zhongyu, ZENG Guohui, HUANG Bo, FANG Zhijun

2019, 39(9):  2517-2522.  DOI: 10.11772/j.issn.1001-9081.2019020284

Asbtract ( )   PDF (912KB) ( )  

References | Related Articles | Metrics

There are many redundant points and inflection points in the path planned by the traditional A^* algorithm. Therefore, an efficient path planning algorithm based on A^* algorithm was proposed. Firstly, the specific calculation method of the evaluation function was improved to reduce the calculation amount of the algorithm searching each interval, thereby reducing the path finding time and changing the generation path. Secondly, on the basis of improving the specific calculation method of the evaluation function, the weight ratio of the evaluation function was improved, and the redundant points and inflection points in the generation path were reduced. Finally, the path generation strategy was improved to delete the useless points in the generation path, improving the smoothness of the path. In addition, considering the actual width of the robot, the improved algorithm introduced an obstacle expansion strategy to ensure the feasibility of the planned path. The comparison of the improved A^* algorithm with three algorithms shows that the path of the improved A^* algorithm is more reasonable, the path finding time is shorter, and the smoothness is higher.

Autonomous obstacle avoidance of unmanned surface vessel based on improved fuzzy algorithm

LIN Zheng, LYU Xiafu

2019, 39(9):  2523-2528.  DOI: 10.11772/j.issn.1001-9081.2019020317

Asbtract ( )   PDF (930KB) ( )  

References | Related Articles | Metrics

In order to improve the performance of continuous obstacle avoidance ability of Unmanned Surface Vessel (USV) in unknown and complex environment, a fuzzy algorithm of obstacle avoidance with speed feedback was proposed. The USV utilized laser scanning radar and multi-channel ultrasonic sensors to perceive the surroundings and performed multi-sensor data fusion by grouping and setting the weight of the obstacle information, and the speed of USV was automatically adjusted according to the environmental situation based on fuzzy control. Then a more comprehensive fuzzy control rule table considering all the distribution of obstacles was proposed to enhance the adaptability of USV to complex environments. The experimental results show that the algorithm can make the USV successfully avoid obstacles and optimize the obstacle avoidance path by adjusting the speed through interaction with the environment, and has good feasibility and effectiveness.

Task planning algorithm of multi-AUV based on energy constraint

ZHAO Xuhao, WANG Yiqun, LIU Jian, XU Chunhui

2019, 39(9):  2529-2534.  DOI: 10.11772/j.issn.1001-9081.2019030539

Asbtract ( )   PDF (900KB) ( )  

References | Related Articles | Metrics

Autonomous Underwater Vehicle (AUV) task planning is the key technology that affects the level of cluster intelligence. In the existing task planning models, only the problem of homogeneous AUV cluster and single dive task planning are considered. Therefore, a multi-dive task planning model for AUV heterogeneous clusters was proposed. Firstly the model considered the energy constraints of AUV, the engineering cost of AUV multiple round-trip charging in mother ship, the efficiency difference between heterogeneous cluster individuals, and the diversity of tasks. Then in order to improve the efficiency of solving the problem model, an optimization algorithm based on discrete particle swarm was proposed. The algorithm introduced matrix coding for describing particle velocity and position and the task loss model for evaluating particle quality to improve the particle updating process, achieving efficient target optimization. Simulation experiments show that the algorithm not only solves the multi-dive task planning problem of heterogeneous AUV clusters, but also reduces the task loss by 11% compared with the task planning model using genetic algorithm.

Fine edge detection model based on RCF

JING Nianzhao, YANG Wei

2019, 39(9):  2535-2540.  DOI: 10.11772/j.issn.1001-9081.2019030462

Asbtract ( )   PDF (970KB) ( )  

References | Related Articles | Metrics

Aiming at the roughness and blur of edges generated by edge detection technology based on deep learning, an end-to-end fine edge detection model based on RCF (Richer Convolutional Features for edge detection) was proposed. In this model based on RCF model, attention mechanism was introduced in the backbone network, Squeeze-and-Excitation (SE) module was used to extract image edge features. In order to avoid excessive loss of detail information, two subsampling in the backbone network were removed. In order to increase the receptive field of the model, dilation convolution was used in the backbone. A residual module was used to fuse the edge images in different scales. The model was trained on the Berkeley Segmentation Data Set (BSDS500)and PASCAL VOC Context dataset by a multi-step training approach and was tested on the BSDS500. The experimental results show that the model improves the ODS (Optimal Dataset Scale) and OIS (Optimal Image Scale) to 0.817 and 0.838 respectively, and it not only generates finer edges without affecting real-time performance but also has better robustness.

Hyperspectral image unmixing algorithm based on spectral distance clustering

LIU Ying, LIANG Nannan, LI Daxiang, YANG Fanchao

2019, 39(9):  2541-2546.  DOI: 10.11772/j.issn.1001-9081.2019020351

Asbtract ( )   PDF (997KB) ( )  

References | Related Articles | Metrics

In order to solve the problem of the effect of noise on the unmixing precision and the insufficient utilization of spectral and spatial information in the actual Hyperspectral Unmixing (HU), an improved unmixing algorithm based on spectral distance clustering for group sparse nonnegative matrix factorization was proposed. Firstly, the HYperspectral Signal Identification by Minimum Error (Hysime) algorithm for the large amount of noise existing in the actual hyperspectral image was introduced, and the signal matrix and the noise matrix were estimated by calculating the eigenvalues. Then, a simple clustering algorithm based on spectral distance was proposed and used to merge and cluster the adjacent pixels generated by multiple bands, whose spectral reflectance distances are less than a certain value, to generate the spatial group structure. Finally, sparse non-negative matrix factorization was performed on the basis of the generated group structure. Experimental analysis shows that for both simulated data and actual data, the algorithm produces smaller Root-Mean-Square Error (RMSE) and Spectral Angle Distance (SAD) than traditional algorithms, and can produce better unmixing effect than other advanced algorithms.

Correntropy self-weighted based joint regularized nearest points for images set classification algorithm

REN Zhenwen, WU Mingna

2019, 39(9):  2547-2551.  DOI: 10.11772/j.issn.1001-9081.2019030463

Asbtract ( )   PDF (837KB) ( )  

References | Related Articles | Metrics

Image set classification algorithms, which make full use of the image set information to improve the recognition performance, have gained much attention. However, existing image set classification algorithms have the following problems:1) samples need to obey a certain probability and statistical distribution; 2) ignoring the mutual exclusion between classes in the gallery set; 3) without robustness against non-Gaussian noise. In order to solve the above problems, an image set classification algorithm based on Correntropy Self-weighted based joint Regularization of Nearest Points (SRNPC) was proposed. Firstly, the unique global joint regularization nearest point in the test set was found and the distance between this point and the regularization nearest point in each gallery set was minimized simultaneously. Then, to enhance the discrimination between classes and the robustness against non-Gaussian noise, a self-weighting strategy based on correntropy scale was introduced to update the correntropy weight between the test set and each gallery set iteratively. And the obtained weight was able to directly reflect the correlation between the test set and each gallery set. Finally, the classification result was obtained by using the minimum residual value between the test set and each gallery set. Experimental results on three open datasets UCSD/Honda, CMU Mobo and YouTube show that SRNPC has higher classification accuracy and better robustness than many state-of-the-art image classification algorithms.

Deep multi-scale encoder-decoder convolutional network for blind deblurring

JIA Ruiming, QIU Zhenzhi, CUI Jiali, WANG Yiding

2019, 39(9):  2552-2557.  DOI: 10.11772/j.issn.1001-9081.2019020373

Asbtract ( )   PDF (1078KB) ( )  

References | Related Articles | Metrics

Aiming at the heterogeneous blur of images caused by inconsistent motion of objects in the shooting scene, a deep multi-scale encoder-decoder convolutional network was proposed to improve the deblurring effect in complex motion scenes. A multi-scale cascade structure named "from coarse to fine" was applied to this network, and blind deblurring was achieved with the blur kernel unknown. In the encoder-decoder module of the network, a fast multi-scale residual block was proposed, which used two branches with different receptive fields to enhance the adaptability of the network to multi-scale features. In addition, skip connections were added between the encoder and the decoder to enrich the information of the decoder. The Peak Signal-to-Noise Ratio (PSNR) value pf this network is 0.06 dB higher than that of the Scale-recurrent Network proposed on CVPR(Conference on Computer Vision and Pattern Recognition)2018; the PSNR and Mean Structural Similarity (MSSIM) values are increased by 1.4% and 3.2% respectively compared to those of the deep multi-scale convolution network proposed on CVPR2017. The experimental results show that the proposed network can deblur the image quickly and restore the edge structure and texture details of the image.

Methods of training data augmentation for medical image artificial intelligence aided diagnosis

WEI Xiaona, LI Yinghao, WANG Zhenyu, LI Haozun, WANG Hongzhi

2019, 39(9):  2558-2567.  DOI: 10.11772/j.issn.1001-9081.2019030450

Asbtract ( )   PDF (1697KB) ( )  

References | Related Articles | Metrics

For the problem of time, effort and money consuming to obtain a large number of samples by conventional means faced by Artificial Intelligence (AI) application research in different fields, a variety of sample augmentation methods have been proposed in many AI research fields. Firstly, the research background and significance of data augmentation were introduced. Then, the methods of data augmentation in several common fields (including natural image recognition, character recognition and discourse parsing) were summarized, and on this basis, a detailed overview of sample acquisition or augmentation methods in the field of medical image assisted diagnosis was provided, including X-ray, Computed Tomography (CT), Magnetic Resonance Imaging (MRI) images. Finally, the key issues of data augmentation methods in AI application fields were summarized and the future development trends were prospected. It can be concluded that obtaining a sufficient number of broadly representative training samples is the key to the research and development of all AI fields. Both the common fields and the professional fields have conducted sample augmentation, and different fields or even different research directions in the same field have different sample acquisition or augmentation methods. In addition, sample augmentation is not simply to increase the number of samples, but to reproduce the existence of real samples that cannot be completely covered by small sample size as far as possible, so as to improve sample diversity and enhance AI system performance.

Real-time facial expression recognition based on convolutional neural network with multi-scale kernel feature

LI Minze, LI Xiaoxia, WANG Xueyuan, SUN Wei

2019, 39(9):  2568-2574.  DOI: 10.11772/j.issn.1001-9081.2019030540

Asbtract ( )   PDF (1097KB) ( )  

References | Related Articles | Metrics

Aiming at the problems of insufficient generalization ability, poor stability and difficulty in meeting the real-time requirement of facial expression recognition, a real-time facial expression recognition method based on multi-scale kernel feature convolutional neural network was proposed. Firstly, an improved MSSD (MobileNet+Single Shot multiBox Detector) lightweight face detection network was proposed, and the detected face coordinates information was tracked by Kernel Correlation Filter (KCF) model to improve the detection speed and stability. Then, three linear bottlenecks of three different scale convolution kernels were used to form three branches. The multi-scale kernel convolution unit was formed by the feature fusion of channel combination, and the diversity feature was used to improve the accuracy of expression recognition. Finally, in order to improve the generalization ability of the model and prevent over-fitting, different linear transformation methods were used for data enhancement to augment the dataset, and the model trained on the FER-2013 facial expression dataset was migrated to the small sample CK+ dataset for retraining. The experimental results show that the recognition rate of the proposed method on the FER-2013 dataset reaches 73.0%, which is 1.8% higher than that of the Kaggle Expression Recognition Challenge champion, and the recognition rate of the proposed method on the CK+ dataset reaches 99.5%. For 640×480 video, the face detection speed of the proposed method reaches 158 frames per second, which is 6.3 times of that of the mainstream face detection network MTCNN (MultiTask Cascaded Convolutional Neural Network). At the same time, the overall speed of face detection and expression recognition of the proposed method reaches 78 frames per second. It can be seen that the proposed method can achieve fast and accurate facial expression recognition.

Classification Method Based on Hierarchical Features of Fundus Images

YU Linfang, DENG Fuhu, QIN Shaowei, QIN Zhiguang

2019, 39(9):  2575-2579.  DOI: 10.11772/j.issn.1001-9081.2019030511

Asbtract ( )   PDF (843KB) ( )  

References | Related Articles | Metrics

To solve the problem of retinal vascular structure division in fundus images, an adaptive breadth-first search algorithm was proposed. Firstly, based on the structure of retinal vessels, the concept of hierarchical features was proposed and feature extraction was carried out. Then, the segmented retinal vessels were analyzed and processed, and several undirected subgraphs were extracted. Finally, the adaptive breadth-first search algorithm was used to classify the hierarchical features in each subgraph. The division of retinal vascular structure was transformed into the classification of hierarchical features. By classifying the hierarchical features of retinal vessels, the hierarchical structures of retinal vascular segments containing these hierarchical features were able to be determined, thus realizing the division of retinal vascular structures. The algorithm has excellent performance when applied to public fundus image databases.

Welding ball edge bubble segmentation for ball grid array based on full convolutional network and K-means clustering

ZHAO Ruixiang, HOU Honghua, ZHANG Pengcheng, LIU Yi, TIAN Zhu, GUI Zhiguo

2019, 39(9):  2580-2585.  DOI: 10.11772/j.issn.1001-9081.2019030523

Asbtract ( )   PDF (1006KB) ( )  

References | Related Articles | Metrics

For inaccurate segmentation results caused by the existence of edge bubbles in welding balls and the grayscale approximation of background due to the diversity of image interference factors in Ball Grid Array (BGA) bubble detection, a welding ball bubble segmentation method based on Fully Convolutional Network (FCN) and K-means clustering was proposed. Firstly, a FCN network was constructed based on the BGA label dataset, and trained to obtain an appropriate network model, and then the rough segmentation result of the image were obtained by predicting and processing the BGA image to be detected. Secondly, the welding ball region mapping was extracted, the bubble region identification was improved by homomorphic filtering method, and then the image was subdivided by K-means clustering segmentation to obtain the final segmentation result. Finally, the welding balls and bubble region in the original image were labeled and identified. Comparing the proposed algorithm with the traditional BGA bubble segmentation algorithm, the experimental results show that the proposed algorithm can segment the edge bubbles of complex BGA welding balls accurately, and the image segmentation results highly match the true contour with higher accuracy.

Data science and technology

Improved K-means algorithm with aggregation distance coefficient

WANG Qiaoling, QIAO Fei, JIANG Youhao

2019, 39(9):  2586-2590.  DOI: 10.11772/j.issn.1001-9081.2019030485

Asbtract ( )   PDF (670KB) ( )  

References | Related Articles | Metrics

Initial centers and K value are determined randomly in the traditional K-means algorithm, which makes clustering results uncertain and with low precision. Therefore, an improved K-means algorithm based on aggregation distance was proposed. Firstly, high-quality cluster centers were filtered out based on the aggregation distance coefficient as the initial centers of the K-means algorithm. Secondly, Davies-Bouldin Index (DBI) was introduced as the criterion function of the algorithm, and the clustering was cyclically updated until the criterion function converged. Finally, the clustering was completed. The proposed algorithm provides good initial clustering centers and K value, avoiding the randomness of clustering results. The clustering results of two-dimensional numerical simulation data show that the improved algorithm can still maintain a good clustering effect when the number of data samples reaches 10000. For the adjusted Rand coefficients of the two UCI standard datasets named Iris and Seg, the improved algorithm respectively improves the performance of clustering by 83.7% and 71.0% compared to the traditional algorithm. It can be seen that the improved algorithm can increase the accuracy of the clustering result compared with the traditional algorithm.

Improved SMOTE unbalanced data integration classification algorithm

WANG Zhongzhen, HUANG Bo, FANG Zhijun, GAO Yongbin, ZHANG Juan

2019, 39(9):  2591-2596.  DOI: 10.11772/j.issn.1001-9081.2019030531

Asbtract ( )   PDF (981KB) ( )  

References | Related Articles | Metrics

Aiming at the low classification accuracy of unbalanced datasets, an unbalanced data classification algorithm based on improved SMOTE (Synthetic Minority Oversampling TEchnique) and AdaBoost algorithm (KSMOTE-AdaBoost) was proposed. Firstly, a noise sample identification algorithm was proposed according to the idea of K-Nearest Neighbors (KNN). The noise samples in the sample set were accurately identified and filtered out by the number of heterogeneous samples included in the K neighbors of the sample. Secondly, in the process of oversampling, the sample set was divided into different sub-clusters based on the idea of clustering. According to the cluster center of the sub-cluster and the number of samples the sub-cluster contains, the synthesis of new samples was performed between the samples in the cluster and the cluster center. In the process of sample synthesis, the data imbalance between classes as well as in the class was fully considered, and the samples were corrected in time to ensure the quality of the synthesized samples and balance the sample information. Finally, using the advantage of AdaBoost algorithm, the decision tree was used as the base classifier and the balanced sample set was trained and iterated several times until the termination condition was satisfied, and the final classification model was obtained. The comparative experiments were carried out on 6 KEEL datasets with G-mean and AUC selected as evaluation indicators. The experimental results show that compared with the classical oversampling algorithm SMOTE and ADASYN (ADAptive SYNthetic sampling approach), G-means and AUC have the highest of 3 groups in 4 groups. Compared with the existing unbalanced classification models SMOTE-Boost, CUS (Cluster-based Under-Sampling)-Boost and RUS (Random Under-Sampling)-Boost, among the 6 groups of data:the proposed classification model has higher G-means than CUS-Boost and RUS-Boost, and 3 groups are lower than SMOTE-Boost; AUC is higher than SMOTE-Boost and RUS-Boost, and one group is lower than CUS-Boost. It is verified that the proposed KSMOTE-AdaBoost has better classification effect and the model has higher generalization performance.

Cyber security

Dynamic trust evaluation method for IoT nodes

XIE Lixia, WEI Ruixin

2019, 39(9):  2597-2603.  DOI: 10.11772/j.issn.1001-9081.2019020315

Asbtract ( )   PDF (1070KB) ( )  

References | Related Articles | Metrics

In order to solve the problem that the existing Internet of Things (IoT) trust evaluation method ignores the impact of the timeliness of trust and non-intrusion factors on direct trust evaluation, and is lack of reliability evaluation of trust recommendation nodes, which lead to low trust evaluation accuracy and low capability to deal with malicious nodes, an IoT node Dynamic Trust Evaluation Method (IDTEM) was proposed. Firstly, the quality of service persistence factor for nodes was introduced to evaluate node behavior and the dynamic trust attenuation factor of nodes was used to express the timeliness of trust, improving the Bayesian-based direct trust evaluation method. Secondly, the reliability of recommended node was evaluated from three aspects:recommended node value, evaluation difference and trust value of the node itself, and was used to optimize the recommendation trust weight calculation method. At the same time, recommendation trust feedback mechanism was designed to suppress collaborative malicious recommendation nodes by the feedback error between the actual trust of the service provided node after providing service and the recommendation trust. Finally, the adaptive weights of direct and recommendation trust of the node were calculated based on the entropy to obtain the comprehensive trust value of the node. Experimental results show that compared with the Reputation-based Framework for high integrity Sensor Network model (RFSN) based on Bayesian theory and the Behavior-based IoT Trust Evaluation Method (BITEM), IDTEM has certain advantages in dealing with malicious services and malicious recommendation behaviors, and has lower transmission energy consumption.

Network intrusion detection model based on improved convolutional neural network

YANG Hongyu, WANG Fengyan

2019, 39(9):  2604-2610.  DOI: 10.11772/j.issn.1001-9081.2019020327

Asbtract ( )   PDF (1032KB) ( )  

References | Related Articles | Metrics

Aiming at the problems of deep learning based network intrusion detection technology such as low detection efficiency, easy over-fitting and weak generalization ability of model training, an Improved Convolutional Neural Network (ICNN) based Intrusion Detection Model (IBIDM) was proposed. Different from the traditional "convolution-pooling-full connection" cascading network design method, the model adopted the design method of cross-layer aggregation network. Firstly, the pre-processed training set data was forwardly propagated as input data and the network features were extracted, and the merge operation was performed on the output data of the cross-layer aggregation network. Then, the training error was calculated according to the classification result and the model was iteratively optimized to convergence by the back propagation process. Finally, a classification test experiment was performed on the test dataset using the trained classifier. The experimental results show that IBIDM has high intrusion detection accuracy and true positive rate, and its false positive rate is low.

CP-ABE access control scheme based on proxy re-encryption in cloud storage

WANG Haiyong, PENG Yao, GUO Kaixuan

2019, 39(9):  2611-2616.  DOI: 10.11772/j.issn.1001-9081.2019020356

Asbtract ( )   PDF (968KB) ( )  

References | Related Articles | Metrics

Focused on the large user's decryption overhead of the Ciphertext Policy Attribute-Based Encryption (CP-ABE) access control scheme in cloud storage, a CP-ABE Access Control Scheme Based on Proxy Re-Encryption (CP-ABE-BPRE) was proposed, and the key generation method was improved. Five components were included in this scheme:trusted key authority, data owner, cloud service provider, proxy decryption server and data visitor. The cloud server re-encrypted the data, and the proxy decryption server performed most of the decryption calculation. The proposed scheme reduces the user's decryption overhead effectively,and solves the data leakage problem caused by illegal stealing of the user's private key in the traditional CP-ABE scheme, and the direct revocation of user attributes is provided while the fine-grained access control is ensured in the scheme. A comparison with other CP-ABE schemes demonstrates that this scheme has better decryption performance for users when accessing cloud data.

Blockchain shard storage model based on threshold secret sharing

ZHANG Guochao, WANG Ruijin

2019, 39(9):  2617-2622.  DOI: 10.11772/j.issn.1001-9081.2019030406

Asbtract ( )   PDF (894KB) ( )  

References | Related Articles | Metrics

To solve the problem that blockchain technology is difficult to be used in large-scale business scenarios due to storage constraints, a blockchain shard storage model based on threshold secret sharing was proposed. Firstly, the transaction data to be placed in blockchain was processed into shards by consensus nodes using improved Shamir's threshold secret sharing. Secondly, consensus nodes constructed different blocks based on data shards and distributed them to other nodes existing in the blockchain network for storage. Finally, when a node wanted to read transaction data, the node would request data from k of the n nodes with transaction data shards, and use Lagrange interpolation algorithm to recover the original transaction data. The experimental results show that the model not only guarantees the security, reliability and privacy of data to be placed in blockchain, but also effectively reduces the amount of data stored by each node to 1/(k-1), which is conducive to blockchain technology using in large-scale business scenarios.

Secure ranked search scheme based on Simhash over encrypted data

LI Zhen, YAO Hanbing, MU Yicheng

2019, 39(9):  2623-2628.  DOI: 10.11772/j.issn.1001-9081.2019020269

Asbtract ( )   PDF (873KB) ( )  

References | Related Articles | Metrics

Concerning the large computation and low search efficiency in ciphertext retrieval, a secure ranked search scheme based on Simhash was proposed. In this scheme, a Secure Multi-keyword Ranked search Index (SMRI) was constructed based on the dimensionality reduction idea of Simhash, the documents were processed into fingerprints and vectors, the B+ tree was built with the segmented fingerprints and encrypted vectors and the "filter-refine" strategy was adopted to searching and sorting. Firstly, the candidate result set was obtained by matching the segmented fingerprints to perform the quick retrieval, then the top-k results were ranked by calculating the Hamming distance and the vector inner product between candidate result set and query trapdoor, and the Simhash algorithm with secret key and the Secure k-Nearest Neighbors (SkNN) algorithm ensured the security of the retrieval process. Simulation results show that compared with the method based on Vector Space Model (VSM), the SMRI-based ranked search scheme has lower computational complexity, saves time and space cost, and has higher search efficiency. It is suitable for fast and secure retrieval of massive ciphertext data.

Threshold signature scheme suitable for blockchain electronic voting scenes

CHENG Yage, JIA Zhijuan, HU Mingsheng, GONG Bei, WANG Lipeng

2019, 39(9):  2629-2635.  DOI: 10.11772/j.issn.1001-9081.2019030513

Asbtract ( )   PDF (1051KB) ( )  

References | Related Articles | Metrics

When traditional signature algorithms such as blind signature and group signature applied to heterogeneous networks of blockchain, they might have problems like relying on trusted centers or low efficiency. Aiming at the problems, a threshold signature scheme suitable for blockchain electronic voting scenes was proposed. The proposed scheme was based on the Asmuth-Bloom secret sharing scheme and did not need a trusted center. Firstly, the signature was generated by the collaboration of blockchain nodes, implementing mutual verification between nodes and improving the node credibility. Secondly, a mechanism of nodes joining and exiting was established to adapt to the high mobility of the blockchain nodes. Finally, the node private keys were updated regularly to resist mobile attacks and make them forward-secure. Security analysis shows that the security of the scheme is based on the discrete logarithm problem, so that the scheme can effectively resist mobile attacks and is forward-secure. The performance analysis shows that compared with other schemes, this scheme has lower computational complexity in the signature generation and verification phases. The results show that the proposed scheme can be well applied to blockchain electronic voting scenes.

Privacy preserving Hamming distance computing problem of DNA sequences

MA Minyao, XU Yi, LIU Zhuo

2019, 39(9):  2636-2640.  DOI: 10.11772/j.issn.1001-9081.2019020247

Asbtract ( )   PDF (816KB) ( )  

References | Related Articles | Metrics

DNA sequences carry important biological information of human bodies, how to compare multiple DNA sequences correctly with privacy preserving is an important problem. To a certain extent, Hamming distance characterizes the similarity between two DNA sequences. Therefore, the privacy preserving Hamming distance computing problem of DNA sequences was researched. First of all, the "0-1 Coding" of the DNA sequence was defined, which codes the DNA sequence with length n to a 0-1 string with length 4n, proving that the Hamming distance of two DNA sequences is a half of the Hamming distance of their "0-1 Coding" strings. Then, with the help of this conclusion, with the Goldwasser-Micali (GM) encryption algorithm taken as main encryption tool, a secure two-party computation protocol for computing the Hamming distance of two DNA sequences was proposed. It was shown that the protocol is both secure and correct under semi-honest attacker model. The proof of security based on a simulator was given. After then, the efficiency of the protocol was analyzed.

Advanced computing

Multi-scale quantum harmonic oscillator algorithm without energy level stability criterion

WANG Dezhi, WANG Peng

2019, 39(9):  2641-2645.  DOI: 10.11772/j.issn.1001-9081.2019020339

Asbtract ( )   PDF (852KB) ( )  

References | Related Articles | Metrics

Multi-scale quantum harmonic oscillator algorithm is an intelligent optimization algorithm based on quantum theory. The energy level stabilization process is one of the core iterative processes of the algorithm, and the energy level stability criterion is the condition for judging whether the algorithm reaches metastable state. Through the analysis of the physical model of the algorithm, it was considered that each iteration of the algorithm in the initial sampling stage is a process of energy level descent, so that the algorithm without energy level stability criterion was also able to realize the evolution from high energy state to metastable state until ground state. The results of the algorithm without energy level stability criterion on six standard test functions show the excellent performance of the algorithm in terms of solution accuracy, success rate and number of iterations. The wave function of the algorithm shows that the algorithm without energy level stability criterion can still converge from the high-energy state to the ground state, and is simpler in structure, easier to use and less difficult to implement. Multi-scale quantum oscillator algorithm without energy level stability criterion can be applied in a more concise and efficient way.

Method for solving color images quantization problem of color images

LI He, JIANG Dengying, HUANG Zhangcan, WANG Zhanzhan

2019, 39(9):  2646-2651.  DOI: 10.11772/j.issn.1001-9081.2019030384

Asbtract ( )   PDF (947KB) ( )  

References | Related Articles | Metrics

For the color quantization problem of color images, the K-means clustering algorithm has strong dependence on initial conditions and is easy to fall into local optimum, and the traditional intelligent optimization algorithms only consider the mutual competition between individuals in the population layer and ignores the mutual cooperation between the population layers. To solve the problems, a K-means-based PES (Pyramid Evolution Strategy) color image quantization algorithm was proposed. Firstly, the clustering loss function in K-means clustering algorithm was used as the fitness function of the new algorithm; secondly, PES was used for the population initialization, layering, exploration, acceleration and clustering of the colors; finally, the new algorithm was used to quantify four standard color test images at different color quantization levels. The experimental results show that the proposed algorithm can improve the defects of the K-means clustering algorithm and the traditional intelligent algorithm. Under the criterion of intra-class mean squared error, the average distortion rate of the image quantized by the new algorithm is 12.25% lower than that quantized by the PES-based algorithm, 15.52% lower than that quantized by the differential evolution algorithm, 58.33% lower than that quantized by the Particle Swarm Optimization (PSO) algorithm, 15.06% lower than that quantized by the K-means algorithm; and the less the color quantization levels, the more the image distortion rate reduced quantized by the new algorithm than that quantized by other algorithms. In addition, the visual effect of the image quantized by the proposed algorithm is better than that quantized by other algorithms.

Network and communications

Low complexity narrowband physical downlink control channel blind detection algorithm based on correlation detection

WANG Dan, LI Anyi, YANG Yanjuan

2019, 39(9):  2652-2657.  DOI: 10.11772/j.issn.1001-9081.2019020262

Asbtract ( )   PDF (1016KB) ( )  

References | Related Articles | Metrics

In NarrowBand Internet of Things (NB-IoT) systems, the Internet of Things (IoT) terminals should decode Downlink Control Information (DCI) quickly to receive resource allocation and scheduling information of the data channel correctly. Therefore, a low complexity Narrowband Physical Downlink Control Channel (NPDCCH) blind detection algorithm using correlation detection was proposed for NPDCCH with search space size being greater than or equal to 32. By employing two correlation judgments on the data in a possible minimum repetition transmission unit of NPDCCH, the invalid data in searching space was removed to reduce the computation complexity. Then, the repetition periods with the valid data were combined and decoded to improve the blind detection performance. Finally, theoretical and simulation analysis of two correlation thresholds used in correlation detection were carried out. Results show that compared with conventional exhaustive blind detection algorithm, the decoding complexity of the proposed algorithm is reduced by at least 75% and the detection performance gain is increased by 2.5 dB to 3.5 dB. The proposed algorithm is more beneficial for engineering practice.

Secure optimized transmission scheme of artificial noise assisted time inversion in D2D cross-cell communication

LI Fangwei, ZHOU Jiawei, ZHANG Haibo

2019, 39(9):  2658-2663.  DOI: 10.11772/j.issn.1001-9081.2019020298

Asbtract ( )   PDF (857KB) ( )  

References | Related Articles | Metrics

In order to solve the problem of intercellular eavesdropping in Device to Device (D2D) communication, an anti-eavesdropping secure transmission scheme based on artificial noise assisted Time-Reversal (TR) was proposed. Firstly, the interference between cells was eliminated under the cross-cell channel model. Secondly, the ability of eavesdropping users to steal information was deteriorated by sending artificial noise to the base station to assist TR technology. Finally, in order to meet the needs of cellular users for service quality and maximize the system traversal secrecy rate, the power control allocation scheme with convex optimization was adopted to maximize the transmitted power of D2D users. Through simulation experiment analysis, compared with the artificial noise scheme, this scheme improves the achievable secrecy rate of 0.8 b·s^-1·Hz^-1 under the same Signal-to-Noise Ratio (SNR). In addition, with the increase of the number of neighboring cells, this scheme has improvement on the reachable secrecy rate more and more obvious.

Content offloading scheme of greedy strategy in mobile edge computing system

YUAN Peiyan, CAI Yunyun

2019, 39(9):  2664-2668.  DOI: 10.11772/j.issn.1001-9081.2019030509

Asbtract ( )   PDF (743KB) ( )  

References | Related Articles | Metrics

Content offloading technology based on mobile edge computing can effectively reduce the traffic pressure on the backbone network and improve the end user's experience. A content offloading scheme of greedy strategy was designed for the heterogeneous contact rate between end users and small base stations. Firstly, the content optimal offloading problem was transformed into the content maximum delivery rate problem. Secondly, the maximum delivery rate problem was proved to satisfy the submodularity. On this basis, the greedy algorithm was used to deploy the content. The algorithm was able to guarantee its optimality with the probability (1-1/e). Finally, the impacts of content popularity index and cache size on different offloading schemes were analyzed in detail. The experimental results show that the proposed scheme improves the content delivery rate and reduces the content transmission delay at the same time.

Important node identification method for dynamic networks based on H operation

SHAO Hao, WANG Lunwen, DENG Jian

2019, 39(9):  2669-2674.  DOI: 10.11772/j.issn.1001-9081.2019020324

Asbtract ( )   PDF (850KB) ( )  

References | Related Articles | Metrics

Focused on the issue that the traditional important node identification method for K-shell networks needs global topology during iteration and cannot be used in dynamic networks, an important node identification method for dynamic networks based on neighborhood priority asynchronous H operation was proposed. Firstly, the algorithm was proved to converge to Ks (K-shell) value; then the degree of each node was taken as the initial value of h-index, and the nodes to be updated were selected by the h-index ranking of the node and the h-index change of the neighbor nodes; meanwhile the h-index was modified to adapt to the topology change according to the number change and maximum degree of the dynamic network nodes, finally the algorithm converged to the Ks and the important nodes were found. The simulation results show that the algorithm can find important nodes effectively by local information of neighbor nodes with less convergence time. Compared with the random selection algorithm and the neighborhood-variety selection algorithm, the convergence time of the proposed algorithm decreases by 77.4% and 28.3% respectively in static networks and 84.3% and 38.8% respectively in dynamic networks.

Computer software technology

SOA based education informatization driven by master data management

MEI Guang, ZOU Henghua, ZHANG Tian, XU Weisheng

2019, 39(9):  2675-2682.  DOI: 10.11772/j.issn.1001-9081.2019030418

Asbtract ( )   PDF (1271KB) ( )  

References | Related Articles | Metrics

The existence of heterogeneous information systems in colleges and universities hinders data assets integration and information interaction. The emergence of Service Oriented Architecture (SOA) and its widespread adoption in enterprises provide ideas for solving this problem, while it is difficult to implement SOA and form an SOA-based informational ecosystem in universities. In response to these problems, an SOA construction scheme driven by master data management was proposed. Firstly, a master data management platform was used to model and integrate the core data assets at the data level. In order to realize data synchronization and consumption, and solve the problem of protocol conversion and service authentication in the process, an enterprise service bus based solution was proposed. Then, in order to the transform the legacy "information island" systems to SOA, a construction solution driven by master data was proposed. The experimental results show that the average latency with concurrency single user, 10 users, 100 users and 10000 users is 8, 11, 59 and 18 ms respectively, which indicates that the performance of the proposed scheme meets the need in different concurrent scenarios. The implementation results show that the data assets integration and information interaction problems have been solved, which proves that the scheme is feasible.

Virtual reality and multimedia computing

Visual analysis method for pilot eye movement data based on user-defined interest area

HE Huaiqing, ZHENG Liyuan, LIU Haohan, ZHANG Yumin

2019, 39(9):  2683-2688.  DOI: 10.11772/j.issn.1001-9081.2019030494

Asbtract ( )   PDF (922KB) ( )  

References | Related Articles | Metrics

Focused on the issue that the traditional interest area based visualization method can not pay attention to the details in the process of analyzing pilot eye movement data, a visual analysis method of eye movement data based on user-defined interest area was proposed. Firstly, according to the specific analysis task, the self-divison and self-definition of the background image of the task were introduced. Then, multiple auxiliary views and interactive approaches were combined, and an eye movement data visual analysis system for pilot training was designed and implemented to help analysts analyze the difference of eye movement between different pilots. Finally, through case analysis, the effectiveness of the visual analysis method and the practicability of the analysis system were proved. The experimental results show that compared with the traditional method, in the proposed method, the analysts' initiative in the analysis process is increased. The analysts are allowed to explore the local details of the task background in both global and local aspects, making the analysts' analyze the data in multi-angle; the analysts are allowed find the flight students' cognitive difficulties in the training process as a whole, so as to develop more targeted and more effective training courses.

Design and implementation of Chinese architecture history teaching system based on mixed reality technology

YAO Luji, ZHANG Li

2019, 39(9):  2689-2694.  DOI: 10.11772/j.issn.1001-9081.2019030545

Asbtract ( )   PDF (1150KB) ( )  

References | Related Articles | Metrics

The teaching of Chinese architecture history has building structures too complex, is limited to 2D planar teaching and is not easy for students to master and apply, therefore an implementation method of Chinese architecture history teaching system based on mixed reality technology was proposed. The wooden structure system of Baoguo Temple in Ningbo was taken as an example, and the mixed reality device Microsoft HoloLens was used as the teaching platform. Firstly, 3ds Max was applied to the 3D simulation modeling of the wooden structure system of Baoguo Temple based on the collected data, and a building model library was built. Then, the 3D human-computer interface of the virtual teaching system was constructed in unity3D, the key technologies were used including environment understanding and human-computer interaction based on C# scripts, and a Chinese architectural history teaching system using HoloLens was implemented with core functions of building structure recognition and cultural cognition. The results show that the system has good 3D visual effects and natural effective human-computer interaction, which can improve the efficiency of knowledge transfer and the initiative of students.

Semi-supervised image segmentation based on prior Laplacian coordinates

CAO Yunyang, WANG Tao

2019, 39(9):  2695-2700.  DOI: 10.11772/j.issn.1001-9081.2019030543

Asbtract ( )   PDF (1037KB) ( )  

References | Related Articles | Metrics

Focusing on the issue that classic semi-supervised image segmentation methods have difficulty in accurately segmenting scattered or small regions, a semi-supervised segmentation algorithm based on label prior and Laplacian Coordinates (LC) was proposed. Firstly, the Laplacian coordinates model was extended, and further the relationship between unlabeled pixels and labeled pixels accurately characterized by introducing the label prior. Secondly, based on the derivation of matrix equation, the posterior probability that the pixel belongs to the label was able to be effectively estimated, thus achieving the segmentation of the image. Thanks to the introduction of the label prior, the algorithm was more robust to the segmentation of scattered and small regions. Lastly, the experimental results on several public semi-supervised segmentation datasets show that the segmentation accuracy of the proposed algorithm is significantly improved compared with that of the Laplacian coordinates algorithm, which verifies the effectiveness of the proposed algorithm.

Image fusion quality evaluation algorithm based on TV-L¹ structure and texture decomposition

ZHANG Bin, LUO Xiaoqing, ZHANG Zhancheng

2019, 39(9):  2701-2706.  DOI: 10.11772/j.issn.1001-9081.2019020302

Asbtract ( )   PDF (1039KB) ( )  

References | Related Articles | Metrics

In order to objectively and accurately evaluate the image fusion algorithms, an evaluation algorithm based on TV-L¹ (Total Variation regularization) structure and texture decomposition was proposed. According to the studies on human visual system, human's perception to image quality mainly comes from the underlying visual features of image, and structure features and texture features are the most important features of underlying visual feature of image. However, the existed image fusion quality evaluation algorithms ignore this fact and lead to inaccurate evaluation. To address this problem, a pair of source images and their corresponding fusion results were individually decomposed into structure and texture images with a two-level TV-L¹ decomposition. Then, According to the difference of image features between the structure and texture images, the similarity evaluation was carried out from the decomposed structure image and the texture image respectively, and the final evaluation score was obtained by integrating the scores at all levels. Based on the dataset with 30 images and 8 mainstream fusion algorithms, compared with the 11 existing objective evaluation indexes, the Borda counting method and Kendall coefficient were employed to verify the consistency of the proposed evaluation algorithm. Moreover, the consistency between the proposed objective evaluation index and the subjective evaluation is verified on the subjective evaluation image set.

Image forgery detection based on local intensity order and multi-support region

YAN Pu, SU Liangliang, SHAO Hui, WU Dongsheng

2019, 39(9):  2707-2711.  DOI: 10.11772/j.issn.1001-9081.2019020306

Asbtract ( )   PDF (810KB) ( )  

References | Related Articles | Metrics

Image forgery detection is currently one of the research focuses of digital image processing, and copy-move forgery is the most frequently used techniques in it. The forgery region is subjected to the operations of scale, rotation, JPEG compression, adding noise and so on before the image moved in, thus detecting the forgery becomes hard. Aimming at the image copy-move forgery technology, an image forgery detection algorithm based on Local Intensity Order Pattern (LIOP) and multiple support regions was proposed. Firstly, the affine invariant regions were detected as support regions by Maximally Stable Extremal Region (MSER) algorithm. Secondly, multiple support regions of different scales, resolutions and directions were obtained by NonSubsampled Contourlet Transform (NSCT). Thirdly, the LIOP descriptors invariant to monotonic intensity change and image rotation were extracted on each support region, and the initial feature matching was implemented via bidirectional distance ratio method. Fourthly, spatial clustering was used to classify the matching features, and geometric transformation parameters were estimated for each cluster by using RANdom SAmple Consensus (RANSAC) algorithm. Finally, the essential operations like post-processing were performed to detect the forgery regions. The experimental results show that the proposed algorithm has higher forgery detection accuracy and reliability.

Single image shadow removal based on attenuated generative adversarial networks

LIAO Bin, TAN Daoqiang, WU Wen

2019, 39(9):  2712-2718.  DOI: 10.11772/j.issn.1001-9081.2019020321

Asbtract ( )   PDF (1327KB) ( )  

References | Related Articles | Metrics

Shadow in an image is important visual information of the projective object, but it affects computer vision tasks. Existing single image shadow removal methods cannot obtain good shadow-free results due to the lack of robust shadow features or insufficiency of and errors in training sample data. In order to generate accurately the shadow mask image for describing the illumination attenuation degree and obtain the high quality shadow-free image, a single image shadow removal method based on attenuated generative adversarial network was proposed. Firstly, an attenuator guided by the sensitive parameters was used to augment the training sample data in order to provide shadow sample images agreed with physical illumination model for a subsequent generator and discriminator. Then, with the supervision from the discriminator, the generator combined perceptual loss function to generate the final shadow mask. Compared with related works, the proposed method can effectively recover the illumination information of shadow regions and obtain the more realistic shadow-free image with natural transition of shadow boundary. Shadow removal results were evaluated using objective metric. Experimental results show that the proposed method can remove shadow effectively in various real scenes with a good visual consistency.

Adaptive intensity fitting model for segmentation of images with intensity inhomogeneity

ZHANG Xuyuan, WANG Yan

2019, 39(9):  2719-2725.  DOI: 10.11772/j.issn.1001-9081.2019020364

Asbtract ( )   PDF (1104KB) ( )  

References | Related Articles | Metrics

For the segmentation of images with intensity inhomogeneity, a region-adaptive intensity fitting model combining global information was proposed. Firstly, the local and global terms were constructed based on local and global image information respectively. Secondly, an adaptive weight function was defined to indicate the deviation degree of the gray scale of a pixel neighborhood by utilizing the extreme difference level in the pixel neighborhood. Finally, the defined weighting function was used to assign weights to local and global terms adaptively to obtain the energy functional of the proposed model and the iterative equation of the model's level set function was deduced by the variational method. The experimental results show that the proposed model can segment various inhomogeneous images stably and accurately in comparison with Region-Scalable Fitting (RSF) model and Local and Global Intensity Fitting (LGIF) model, which is more robust in the position, size and shape of initial contour of evolution curve.

Video colorization method based on hybrid neural network model of long short term memory and convolutional neural network

ZHANG Zheng, HE Shan, HE Jingqi

2019, 39(9):  2726-2730.  DOI: 10.11772/j.issn.1001-9081.2019020264

Asbtract ( )   PDF (985KB) ( )  

References | Related Articles | Metrics

A video can be seen as a sequence formed by continuous video frames of images, and the colorization process of video actually is the colorization of images. If the existing image colorization method is directly applied to video colorization, it tends to cause flutter or twinkle because of long-term sequentiality of videos. For this problem, a method based on Long Short Term Memory (LSTM) cells and Convolutional Neural Network (CNN) was proposed to colorize the grayscale video. In the method, the semantic features of video frames were extracted with CNN and the time sequence information of video was learned by LSTM cells to keep the time-space consistency of video, then local semantic features and time sequence features were fused to generate the final colorized video frames. The quantitative assessment and user study of the experimental results show that this method achieves good performance in video colorization.

Medical image super-resolution reconstruction based on depthwise separable convolution and wide residual network

GAO Yuan, WANG Xiaochen, QIN Pinle, WANG Lifang

2019, 39(9):  2731-2737.  DOI: 10.11772/j.issn.1001-9081.2019030413

Asbtract ( )   PDF (1073KB) ( )  

References | Related Articles | Metrics

In order to improve the quality of medical image super-resolution reconstruction, a wide residual super-resolution neural network algorithm based on depthwise separable convolution was proposed. Firstly, the depthwise separable convolution was used to improve the residual block of the network, widen the channel of the convolution layer in the residual block, and pass more feature information into the activation function, making the shallow low-level image features in the network easier transmitted to the upper level, so that the quality of medical image super-resolution reconstruction was enhanced. Then, the network was trained by group normalization, the channel dimension of the convolutional layer was divided into groups, and the normalized mean and variance were calculated in each group, which made the network training process converge faster, and solved the difficulty of network training because the depthwise separable convolution widens the number of channels. Meanwhile, the network showed better performance. The experimental results show that compared with the traditional nearest neighbor interpolation, bicubic interpolation super-resolution algorithm and the super-resolution algorithm based on sparse expression, the medical image reconstructed by the proposed algorithm has richer texture detail and more realistic visual effects. Compared with the super-resolution algorithm based on convolutional neural network, the super-resolution neural network algorithm based on wide residual and the generative adversarial-network super-resolution algorithm, the proposed algorithm has a significant improvement in PSNR (Peak Signal-to-Noise Ratio) and SSIM (Structural SIMilarity index).

Underwater image super-resolution reconstruction method based on deep learning

CHEN Longbiao, CHEN Yuzhang, WANG Xiaochen, ZOU Peng, HU Xuemin

2019, 39(9):  2738-2743.  DOI: 10.11772/j.issn.1001-9081.2019020353

Asbtract ( )   PDF (893KB) ( )  

References | Related Articles | Metrics

Due to the characteristics of water itself and the absorption and scattering of light by suspended particles in the water, a series of problems, such as low Signal-to-Noise Ratio (SNR) and low resolution, exist in underwater images. Most of the traditional processing methods include image enhancement, restoration and reconstruction rely on degradation model and have ill-posed algorithm problem. In order to further improve the effects and efficiency of underwater image restoration algorithm, an improved image super-resolution reconstruction method based on deep convolutional neural network was proposed. An Improved Dense Block structure (IDB) was introduced into the network of the method, which can effectively solve the gradient disappearance problem of deep convolutional neural network and improve the training speed at the same time. The network was used to train the underwater images before and after the degradation by registration and obtained the mapping relation between the low-resolution image and the high-resolution image. The experimental results show that on a self-built underwater image training set, the underwater image reconstructed by the deep convolutional neural network with IDB has the Peak Signal-to-Noise Ratio (PSNR) and Structural SIMilarity (SSIM) improved by 0.38 dB and 0.013 respectively, compared with SRCNN (an image Super-Resolution method using Conventional Neural Network) and proposed method can effectively improve the reconstruction quality of underwater images.

Insect sound feature recognition method based on three-dimensional convolutional neural network

WAN Yongjing, WANG Bowei, LOU Dingfeng

2019, 39(9):  2744-2748.  DOI: 10.11772/j.issn.1001-9081.2019030481

Asbtract ( )   PDF (794KB) ( )  

References | Related Articles | Metrics

The quarantine of imported wood is an important task for the customs, but there are problems such as low accuracy and poor robustness in the insect sound detection algorithm. To solve these problems, an insect sound detection method based on Three-Dimensional Convolutional Neural Network (3D CNN) was proposed to detect the presence of insect sound features. Firstly, the original insect audio was framed and pre-processed, and Short-Time Fourier Transform (STFT) was operated to obtain the spectrogram of the insect audio. Then, the spectrogram was used as the input of the 3D CNN consisting three convolutional layers. Network training and testing were conducted by setting inputs with different framing lengths. Finally, the analysis of performance was carried out using metrics like accuracy, F₁ score and ROC curve. The experiments showed that the test results were best when the overlap framing length was 5 seconds. The best result of the 3D CNN model on the test set achieved an accuracy of 96.0% and an F₁ score of 0.96. The accuracy was increased by nearly 18% compared with that of the two-dimensional convolutional neural network (2D CNN) model. It shows that the proposed model can extract the insect sound features from the audio signal more accurately and complete the insect identification task, which provides an engineering solution for customs inspection and quarantine.

Frontier & interdisciplinary applications

Railway crew routing plan based on improved ant colony algorithm

WANG Dongxian, MENG Xuelei, QIAO Jun, TANG Lin, JIAO Zhizhen

2019, 39(9):  2749-2756.  DOI: 10.11772/j.issn.1001-9081.2019020368

Asbtract ( )   PDF (1297KB) ( )  

References | Related Articles | Metrics

In order to improve the quality and efficiency of railway crew routing plan, the problem of crew routing plan was abstracted as a Multi-Traveling Salesman Problem (MTSP) with single base and balanced travel distance, and a equilibrium factor was introduced to establish a mathematical model aiming at less crew routing time and balanced tasks between sub-crew routings. A dual-strategy ant colony optimization algorithm was proposed for this model. Firstly, a solution space satisfying the space-time constraints was constructed and pheromone concentration was set for the node of the crew section and the continuation path respectively, then the transitional probability of the dual-strategy state was adopted to make the ant traverse all of the crew segments, and finally the sub-crew routings that meet the crew constraint rules were found. The designed model and algorithm were tested by the data of the intercity railway from Guangzhou to Shenzhen. The comparison with the experimental results of genetic algorithm shows that under the same model conditions, the number of crew routing in the crew routing plan generated by double-strategy ant colony optimization algorithm is reduced by about 21.74%, the total length of crew routing is decreased by about 5.76%, and the routing overload rate is 0. Using the designed model and algorithm to generate the crew routing plan can reduce the crew routing time of crew plan, balance the workload and avoid overload routing.

High-speed train connection optimization for large passenger transport hub based on transfer orientation

QIAO Jun, MENG Xuelei, WANG Dongxian, TANG Lin

2019, 39(9):  2757-2764.  DOI: 10.11772/j.issn.1001-9081.2019020350

Asbtract ( )   PDF (1248KB) ( )  

References | Related Articles | Metrics

In view of the optimization of high-speed train connection in passenger transport hub under the condition of high-speed railway network, the concept of transfer satisfaction of medium and long distance passenger flow was proposed by analyzing the passenger transfer process in hub, and a high-speed train connection optimization model for large passenger transport hub based on transfer orientation was proposed with the average transfer satisfaction and the arrival and departure equilibrium of trains at hub stations as the optimization objective and with the constraint conditions of reasonable originating time of large stations, reasonable terminating time, station operation interval time, passenger transfer time and station arrival and departure line capacity. A genetic algorithm with improved chromosome coding mode and selection strategy was designed to solve the example. Compared with the basic genetic algorithm and the basic simulated annealing algorithm, the improved genetic algorithm increases the average transfer satisfaction in the objective function by 5.10% and 2.93% respectively, and raises the equilibrium of arrival and departure of trains at hub stations by 0.27% and 2.31% respectively. The results of the example verify the effectiveness and stability of the improved genetic algorithm, which indicates that the proposed method can effectively optimize the quality of the high-speed train connection in large passenger transport hub.

Multi-objective optimization model and solution algorithm for emergency material transportation path

LI Zhuo, LI Yinzhen, LI Wenxia

2019, 39(9):  2765-2771.  DOI: 10.11772/j.issn.1001-9081.2019020270

Asbtract ( )   PDF (983KB) ( )  

References | Related Articles | Metrics

For the actual background of the shortage of self-owned vehicles of the transporters in the early stage of emergency, the combinatorial optimization problem of hybrid vehicle paths with transportation mode of joint distribution of self-owned vehicles and vehicles rented by third-party was studied. Firstly, with the different interests between demand points and transporters considered, a multi-objective hybrid vehicle routing optimization model with soft time windows was established with the goal of maximizing system satisfaction and minimizing system delivery time and total cost. Secondly, the shortcomings of NSGA-Ⅱ algorithm in solving this kind of problems such as poor convergence and uneven distribution of Pareto frontiers were considered, the heuristic strategy and pheromone positive feedback mechanism of ant colony algorithm were used to generate offspring population, non-dominated sorting strategy model was used to guide the multi-objective optimization process, and the variable neighborhood descent search was introduced to expand the search space. A multi-objective non-dominated sorting ant colony algorithm was proposed to break through the bottleneck of the original algorithm. The example shows that the proposed model can provide reference for decision makers to choose reasonable paths according to different optimization objectives in different situations, and the proposed algorithm shows better performance in solving different scale problems and different distribution type problems.

Whole-set order problem of hybrid flow shop based on heuristic-genetic algorithms

JIA Yeling, Dong Shaohua

2019, 39(9):  2772-2777.  DOI: 10.11772/j.issn.1001-9081.2019030468

Asbtract ( )   PDF (882KB) ( )  

References | Related Articles | Metrics

A heuristic-genetic algorithm based on batch scheduling strategy was proposed for the whole-set order problem in hybrid flow shop environment with process constrainted parallel machine. Firstly, a mathematical model was established with the objective of maximizing the number of weighted whole-set orders, and initial scheduling was generated by inner genetic algorithm applying to workpieces in batches. Then, the target was transformed to the maximum weighted whole-set order quantity by outer heuristic rules, and an order evaluation index was designed to break the delivery time bottleneck. Finally, the inner and outer algorithms were optimized circularly until there was no bottleneck, which means the satisfactory solution was obtained. Examples show that heuristic-genetic algorithm can obtain the optimal scheduling within 20 generations, and the probability of obtaining the optimal solution is 70% when the population size is larger than 50. The experimental results show that when the scale of the problem increases to 40 workpieces, the solving time of genetic algorithm increases significantly, and the number of whole-set orders optimized by Smallest Critical Ratio (SCR) rule is smaller than the heuristic-genetic algorithm in different problem sizes. Heuristic-genetic algorithm can increase the quantity of weighted whole-set orders to more than 1.5 times in practical engineering, and shorten the processing time by 5.1% on average. The results show that the heuristic-genetic algorithm can solve the problem that the whole-set order problems are easy to fall into local optimum in the hybrid flow shop environment, and can realize the synchronization of planning and production in the large-scale and complex hybrid flow shop ordering enterprises, which has practical significance.

Task assignment based on discrete cuckoo search algorithm in mobile crowd sensing system

YANG Zhengqing, ZHOU Zhaorong, YUAN Shu

2019, 39(9):  2778-2783.  DOI: 10.11772/j.issn.1001-9081.2019020365

Asbtract ( )   PDF (886KB) ( )  

References | Related Articles | Metrics

Considering the problems of low-enthusiasm workers and task expiration in the mobile crowd sensing system, a task assignment algorithm based on initial cost and soft time window was proposed. As the corresponding task assignment problem belongs to the category of NP-hard problems and the computationally efficient optimal algorithm cannot be found, thus, an algorithm was developed based on Discrete Cuckoo Search Algorithm (DCSA). Firstly, the corresponding global search process and local search process were designed respectively according to the problem characteristics. Secondly, to derive the better solution, the priorities of tasks with respect to the distance between tasks and workers' starting positions as well as the size of time windows were analyzed. Finally, feasible operations were executed to guarantee that the related constraints were satisfied by each task assignment. Compared with genetic algorithm and greedy algorithm, the simulation results show that DCSA-based task assignment algorithm can improve the enthusiasm of workers to participate, solve the problem of task expiration, and ultimately reduce the total system cost.

Application of KNN algorithm based on value difference metric and clustering optimization in bank customer behavior prediction

LI Bo, ZHANG Xiao, YAN Jingyi, LI Kewei, LI Heng, LING Yulong, ZHANG Yong

2019, 39(9):  2784-2788.  DOI: 10.11772/j.issn.1001-9081.2019030571

Asbtract ( )   PDF (806KB) ( )  

References | Related Articles | Metrics

In order to improve the accuracy of loan financial customer behavior prediction, aiming at the incomplete problem of dealing with non-numerical factors in data analysis of traditional K-Nearest Neighbors (KNN) algorithm, an improved KNN algorithm based on Value Difference Metric (VDM) distance and iterative optimization of clustering results was proposed. Firstly the collected data were clustered by KNN algorithm based on VDM distance, then the clustering results were analyzed iteratively, finally the prediction accuracy was improved through joint training. Based on the customer data collected by Portuguese retail banks from 2008 to 2013, it can be seen that compared with traditional KNN algorithm, FCD-KNN (Feature Correlation Difference KNN) algorithm, Gauss Naive Bayes algorithm, Gradient Boosting algorithm, the improved KNN algorithm has better performance and stability, and has great application value in the customer behavior prediction from bank data.

Duplicate detection algorithm for massive images based on pHash block detection

TANG Linchuan, DENG Siyu, WU Yanxue, WEN Liuying

2019, 39(9):  2789-2794.  DOI: 10.11772/j.issn.1001-9081.2019020792

Asbtract ( )   PDF (834KB) ( )  

References | Related Articles | Metrics

The large number of duplicate images in the database not only affects the performance of the learner, but also consumes a lot of storage space. For massive image deduplication, a duplicate detection algorithm for massive images was proposed based on pHash (perception Hashing). Firstly, the pHash values of all images were generated. Secondly, the pHash values were divided into several parts with the same length. If the values of one of the pHash parts of the two images were equal to each other, the two images might be duplicate. Finally, the transitivity of image duplicate was discussed, and corresponding algorithms for transitivity case and non-transitivity case were proposed. Experimental results show that the proposed algorithms are effective in processing massive images. When the similarity threshold is 13, detecting the duplicate of nearly 300000 images by the proposed transitive algorithm only takes about two minutes with the accuracy around 53%.