Computer Engineering & Science

Design and optimization of multiple parallel transmission lines in high-speed chip packaging

FAN Yu-qing, HU Jin, ZHENG Hao

2022, 44(05): 761-768. doi:

Abstract ( 120 )

PDF (1562KB) ( 197 ) 　　

With the development of packaging substrates in the direction of high-end and high-density, the signal integrity problem caused by the increase in signal density on the substrates have become increasingly serious. In order to study the relevance of reflection, crosstalk and other issues in high-speed interconnect structures with package substrate types, and design parameters and physical characteristics of transmission lines, this paper improves the simple two-wire parallel coupling model, uses three- dimensional electromagnetic simulation software to construct a new package-level three parallel transmission line model, analyzes the transmission line reflection and crosstalk characteristics of the organic substrate and the ceramic substrate, and studies the method of reducing the reflection coefficient and crosstalk noise under this structure. The simulation results show that the reflection coefficient S11 of the transmission line on the package substrate is related to the degree of impedance matching, and is greatly affected by the signal line width, thickness, and medium thickness. The optimal line width of the S11 minimum value is also different at different frequencies, and needs to be selected according to different signal frequencies. The near-end crosstalk coefficient is affected by the fringe field and is closely related to the line spacing. The far-end crosstalk coefficient is greatly affected by the thickness of the medium. Under the same conditions, the far-end crosstalk noise is generally less than the near-end crosstalk noise. The evaluation should be based on the signal desnsity on the substrate, substrate material properties, and media thickness.

A verification framework of network on chip for neuromorphic processors

CHEN Xiao-fan, YANG Zhi-jie, PENG Ling-hui, WANG Shi-ying, ZHOU Gan, LI Shi-ming, KANG Zi-yang, WANG Yao, SHI Wei, WANG Lei

2022, 44(05): 769-778. doi:

Abstract ( 140 )

PDF (1341KB) ( 140 ) 　　

In recent years, traditional computer architectures have gradually been faced with severe bottlenecks of “Memory Wall” and “Power Wall”, with the step-down Moores law. However, many other new forms of computing paradigms and computer architectures have been proposed, including neuromorphic computing. Given the characteristic of computing in memory, neuromorphic computing plays a vital role in breaking down the limitation caused by both “Memory Wall” and “Power Wall” constraints in Von Neumann architecture. Many neuromorphic applications on neuromorphic processors have already been demonstrated as high efficiency and accuracy. Currently, in the application scenarios of large-scale biological neural networks, it is necessary to improve the scalability of multi-core neuromorphic processors and maintain their high data throughput and low transmission delay. Today, most multi-core neuromorphic processors adopt a network-on-chip (NoC) as the interconnect structure. However, there are still relatively few verification studies on such NoC. Given the importance of NoC in designing a neuromorphic processors, it is quite necessary to set up a complete and robust NoC functional verification platform for neuromorphic processors. The purpose of this paper is to generate the stimulus files required for behavioral and FPGA hardware-level testing based on the randomization method, and to achieve a more comprehensive functional verification through efficient processing of log files.

A Flink load balancing strategy for cloud environment

XU Hao-tong, HUANG Shan, SUN Guo-zhang, HE Fei-li, DUAN Xiao-dong,

2022, 44(05): 779-787. doi:

Abstract ( 146 )

PDF (913KB) ( 143 ) 　　

As a new generation of big data computing engine, Flink has been widely used. When containers of Flink are deployed in cloud environment, its default task scheduling algorithm cannot perceive node resources information and adjust the load in time, and the capacity for independent equilibrium is poorer. Although mainstream container layout tools provide the possibility of container management, they fails to combine Flink characteristics to solve the problem of balancing the resource utilization while reducing the communication overhead in the container group. Aiming at the above problem, this paper proposes a Flink load balancing strategy for cloud environment, which comprehensively considers the distribution characteristics of operators in Flink cluster and the communication mechanism between containers, and takes the communication cost between nodes and load balancing as evaluation criteria. Experimental results show that, compared with Flink default scheduling algorithm, this algorithm can effectively improve the computing efficiency and system performance.

A Boosting classification algorithm for imbalanced drift data stream based on Hellinger distance

ZHANG Xi-long, HAN Meng, CHEN Zhi-qiang, WU Hong-xin, LI Mu-hang

2022, 44(05): 788-799. doi:

Abstract ( 86 )

PDF (1500KB) ( 109 ) 　　

Imbalanced data stream will seriously affect the classification performance of the algorithm and the emer-gence of concept drift is a difficult problem in the field of stream data mining. In order to improve the classification performance of such problem, a new Boosting Classification Algorithm for imbalanced drifted data stream based on Hellinger Distance (BCA-HD) is proposed. The algorithm innovatively uses the weighted combination of instance level and classifier level to dynamically update the classifier to adapt to the occurrence of concept drift. The integrated algorithm SMOTEBoost is used as the base classifier at the bottom layer, and the classifier uses resampling technology to deal with the imbalanced data. Finally, the proposed algorithm is compared with 9 different algorithms on 16 abrupt and gradual datasets. The results show that average value and average rankings of G-mean and AUC are both ranked first. Experiments show that the algorithm can better adapt to the simultaneous occurrence of concept drift and imbalance, which helps to improve the classification performance.

An attribute-based encryption scheme preventing irrelevant attributes interference

XU Cheng-zhou, ZHANG Wen-tao, LANG Jing-hong

2022, 44(05): 800-809. doi:

Abstract ( 86 )

PDF (616KB) ( 139 ) 　　

In order to improve the expressivity of the access structure in attribute-based encryption, and avoid the interference of irrelevant attributes in the access structure, a ciphertext-policy attribute-based encryption (CP-ABE) scheme based on reduced ordered binary decision diagram (ROBDD) access structure is proposed. The ROBDD access structure in this scheme can effectively express the access policy with complex access logic and prevent the interference of irrelevant attributes, which improves the encryption speed. RSA attribute authentication mechanism is introduced to achieve attribute authentication in non-leaf nodes of ROBDD, which realizes anti-collusion attack and protection of user attribute set. The effective path eigenvalues and encryption parameters in ROBDD are used to create polynomials, and any effective path eigenvalue can get encryption parameters through polynomial calculation, which reduces the cost of ciphertext storage overhead. The scheme implements user revocation, user attribute revocation, and system attribute revocation. Performance analysis and experimental simulation show that the proposed scheme has higher encryption and decryption efficiency, and lower ciphertext storage overhead.

A blockchain data sharing system based on ECC

LIN Jie-he, ZHANG Shao-hua, LI Chao, DAI Bing-rong

2022, 44(05): 810-818. doi:

Abstract ( 133 )

PDF (1023KB) ( 171 ) 　　

Blockchain technology can solve the problems of data leakage, data tampering, and difficult data traceability that exist in traditional data sharing by constructing an unforgeable, non-tamperable and traceable chain data structure model in a peer-to-peer network environment. However, the existing blockchain data sharing so-lutions also have problems such as high cost, low efficiency, and poor security. Aiming at these issues, a blockchain data sharing scheme based on elliptic curve cryptography (ECC) is proposed and the system is designed. The program relies on ECC to ensure the security of the data transmission process and improve the problems in the existing program. The corresponding smart contract was written by the Solidity programming language, and the data sharing system was simulated and tested based on the Ethereum platform.

Construction of a traffic blocking model for wireless sensor network based on random forest algorithm

XU Li-jin, HE Yan-fang

2022, 44(05): 819-825. doi:

Abstract ( 102 )

PDF (665KB) ( 106 ) 　　

Aiming at the problems of low detection accuracy and poor blocking effect of attack traffic in wireless sensor networks, a wireless sensor network attack traffic blocking model based on random forest algorithm is constructed. Through TF-IDF algorithm, the feature of payload is automatically extracted based on the word frequency matrix of characters (words). According to the characteristic results, the random forest algorithm is used to classify the network traffic through the word frequency matrix, and the traffic attack in the network can be traced based on the classification results to complete the detection of abnormal wireless sensor networks. The packet filtering of the flow table is used to block the traffic of wireless sensor attack. Experiments show that, when detecting attack traffic, the detection accuracy of the model can reach 100%, the highest harmonic mean is 99.18%, and the highest error rate is only 7.3%, and the false positive rate is only 5.5%. At the same time, it can effectively block the network attack traffic and restore the network to normal in a short time. It has good attack traffic detection effect and attack traffic blocking effect.

Realization of malicious code family classification based on semi-supervised generative adversarial network

WANG Dong, YANG Ke, XUAN Jia-xing, HAN Yu-tong, ZHAO Li-hua, WANG Xu-ren

2022, 44(05): 826-833. doi:

Abstract ( 145 )

PDF (825KB) ( 150 ) 　　

With the development of Internet, malicious code tend to be massive and polymorphic. The classification of malicious code family is one of the challenges of cyber security. Combining the semi supervised generation network with the deep convolutional neural network, a multi-family malicious code classification model is proposed. Taking the gray image of malicious codes as the feature, based on the efficient one-dimensional convolutional neural network (1D-CNN), using the semi-supervised generative adversarial network (SGAN), an efficient and accurate malicious code family classification model is constructed as SGAN-CNN, which can improve the malicious code classification ability from aspects of efficient feature extraction and SGAN optimization. In order to verify the classification ability of the model, experiments are carried out on the Microsoft malware classification challenge data set. 5-fold cross-validation shows that the proposed model achieves 98.81% of the average accuracy of the test set with 80% of the tag rate, 98.01% of the average accuracy of the test set with 20% of the tag rate, and achieves better experimental results. In the case of small samples, it can also achieve good classification accuracy.

Overlapping community detection based on colored random walk

CHANG Yang, MA Hui-fang,

2022, 44(05): 834-844. doi:

Abstract ( 97 )

PDF (1024KB) ( 98 ) 　　

Community detection is a fundamental and widely-studied problem. Most existing community detection methods focus on network topology. However, with the proliferation of rich information available for entities in real-world networks, it is indispensable to capture the rich interaction between the structure and attributes in the graph for community detection. This paper proposes an Overlapping Community Detection method based on Colored random walk (OCDC). The algorithm is able to conquer the limitation of random walk-based community detection methods that directly utilize the original network topology. Specifically, initial seed nodes in the network is firstly selected. Secondly, a seed replacement strategy is developed to capture a better-quality seed replacement path set. Thirdly, the structure-attribute interaction node transition matrix is generated to perform the colored random walk in order to obtain the colored distribution vector. Finally, based on the combination of structure and attri- bute, the parallel conductance is captured to expand the community. Experiments on synthetic networks and real-world network show that our proposed algorithm can accurately identify attributed communities and significantly outperform the state-of-the-art methods.

Crack extraction from single tunnel image based on fully convolutional neural network

QIU Jing-bo, YAN Xue-feng, WANG Jun, GUO Yan-wen, WEI Ming-qiang,

2022, 44(05): 845-854. doi:

Abstract ( 187 )

PDF (1905KB) ( 161 ) 　　

A single tunnel image crack extraction algorithm based on fully convolutional neural network is proposed, which can effectively avoid the interference of pseudo-crack noise points in complex backgrounds and achieve accurate segmentation of tunnel cracks. First, a deep residual network model is constructed to extract crack features. Second, in order to recover the size of the crack feature map as well as the crack details, the feature map size is recovered using a deconvolution operation in an improved fully convolutional neural network. In order to improve the fineness of crack extraction, a detail repair module is proposed, which can maintain the integrity and edge details of cracks. Finally, a crack dataset NUAACrack-2000 is published, which contains 2000 tunnel crack images and accurately labeled labels. Extensive experiments show that the proposed algorithm outperforms traditional image segmentation algorithms in avoiding noise point interference. It is superior to the mainstream crack extraction algorithms based on machine learning in preserving the integrity of the extracted cracks and processing edge details.

Discrimination-enhanced generative adversarial network in text-to-image generation

TAN Hong-chen, HUANG Shi-hua, XIAO He-wen, YU Bing-bing, LIU Xiu-ping

2022, 44(05): 855-861. doi:

Abstract ( 105 )

PDF (884KB) ( 145 ) 　　

Based on Generative Adversarial Networks (GANs), most current text-to-image generation algorithms focus on designing different attention generation models to improve the characterization and expression of image details. However, they ignore the discriminators perception of key local semantics, so the generation models can easily generate poor image details to “fool” the discriminators. This paper designs a vocabulary-image discriminative attention module in the discriminators to enhance the discriminators ability to perceive and capture key semantics, and drive the generation model to generate high-quality image details. Therefore, a discrimination-enhanced generative adversarial model (DE-GAN) is proposed. The experimental results show that， on the CUB-Bird dataset, DE-GAN achieves 4.70 on the IS index, which is 4.2% higher than the baseline model and achieves high performance.

Semantic segmentation of 3D point cloud based on all fusion network

LIU Li-man, TAN Long-yu, PENG Yuan, LIU Jia

2022, 44(05): 862-869. doi:

Abstract ( 100 )

PDF (1213KB) ( 115 ) 　　

In order to improve the accuracy of point cloud semantic segmentation in indoor scenes, an all fusion network for semantic segmentation of 3D point clound is proposed. The network consists of a feature encoding module, a progressive feature decoding module, a multi-scale feature decoding module, a feature fusion module, and a semantic segmentation header. The feature encoding module uses inverse density weighted convolution as the feature encoder to perform hierarchical feature encoding on point cloud, so as to extract multi-scale features of the point cloud. Then, the progressive feature decoder is used to decode high-level semantic features layer by layer to obtain the point cloud progressive decoding feature. In the same pair, the multi-scale feature decoder performs feature decoding on the extracted multi-scale features to obtain multi-scale decoding features of the point cloud. Finally, the progressive decoding feature is fused with the multi-scale decoding feature, then semantic segmentation header is introduced to realize the point cloud semantic segmentation. The all fusion network robustly enhances the feature extraction ability of the network, and the experimental results also verify the effectiveness of the method.

Cell annotation refinement and adaptive weighted loss for CD56 image segmentation

LIU Rong, WU Xin, AO Bin, WEN Qing, LI Kuan

2022, 44(05): 870-878. doi:

Abstract ( 93 )

PDF (1449KB) ( 123 ) 　　

CD56 is a nerve cell adhesion molecule that can be used in the diagnosis and study on a variety of tumor cells. CD56 is one of the latest tumor molecular markers, and the research on CD56 images in the field of digital medical image processing has just started. With the development of deep learning technologies such as semantic segmentation, more and more researchers are applying semantic segmentation technology to medical image processing in order to assist doctors in diagnosis. In CD56 images, the proportion of the number of pixel points of background, negative cells and positive cells is very unbalanced, which is roughly 70∶10∶1 and affects the segmentation effect of semantic segmentation technology on CD56 images. In this paper, the loss function of the semantic segmentation model is improved by adding loss weight to different class and adding adaptive weight to each pixel, so that the model can pay more attention to cells, especially positive cells. At the same time, the clustering method is used to refine the annotations of CD56 images before model training, which further improves the segmentation accuracy of the model. The experimental results on CD56 image dataset show that, the refinement of the image annotations and the improvement of the loss functions in the relevant semantic segmentation models can improve the segmentation accuracy of the CD56 images.

Review of early classification of Alzheimers disease based on computer-aided diagnosis technology

CHU Yang, XU Wen-long

2022, 44(05): 879-893. doi:

Abstract ( 287 )

PDF (815KB) ( 270 ) 　　

As one of the major neurodegenerative diseases, Alzheimers disease (AD) has become the most common cause of dementia. Up to now, there is still a lack of effective targeted therapeutic drugs and treatments to prevent the progression of this disease. With the continuous development of computer technology, the use of computer-aided diagnosis technology tools for early AD classification research will provide clinicians with important help. We review the early diagnosis and classification of AD using traditional machine learning and deep learning techniques in recent years. Brain neuroimaging data (such as MRI, PET), electroencephalogram (EEG) and other biomarkers are mainly focused on. By combining machine learning, the classification of early diagnosis of AD is studied. Firstly, the application of machine learning methods to the early AD classification is analyzed, and the classification using different classification algorithms is compared. Secondly, the different biomarkers of subjects using single mode or multi-mode are compared to analyze early AD classification research. Finally, the challenges faced by AD classification are introduced, and future research directions are proposed.

Sparse autoencoder based on earth mover distance

FAN Yun

2022, 44(05): 894-900. doi:

Abstract ( 88 )

PDF (1403KB) ( 111 ) 　　

KL divergence is adopted widely in the field of machine learning to measure distances between distributions in model loss function. In the sparse autoencoder, the KL divergence is used as the penalty term of the loss function to measure the distance between the neuron output and the sparse parameter, so that the neuron output approaches the sparse parameter, thereby suppressing the activation of the neuron to obtain sparse coding. In WGAN, Wasserstein distance is used to solve the gradient va- nishing and mode collapse problems of GAN, making the training of GAN more stable. The experimental results show that, compared with the sparse autoencoder using KL divergence and JS divergence, the sparse autoencoder using EMD distance as a penalty term can reduce the reconstruction error between real samples and reconstructed samples. As the penalty parameter increases, the encoding becomes more sparse.

An address matching algorithm based on hybrid neural network model and attention mechanism

CHEN Jian-peng, CHEN Jian, SHE Xiang-rong, SHUI Xin-ying, CHEN Gang

2022, 44(05): 901-909. doi:

Abstract ( 129 )

PDF (1186KB) ( 214 ) 　　

The standardization of Chinese geographic addresses plays a crucial role in the current construction of smart cities. The traditional geographic address standardization technology usually uses the methods of similarity calculation or rule base matching based on the text character level, and the processing effect of complex, special or redundant addresses is poor. This paper proposes an address match- ing algorithm that combines attention mechanism and multi-level representation by converting the address standardization task into a matching degree calculation task for similar addresses. Firstly, according to the special grammatical structure of the address text, a standard address tree is constructed by using the Trie grammatical tree. Secondly, based on the attention mechanism, the Bi-LSTM network and the CNN network are used to generate multi-level semantic representations of address pairs. Finally, the similarity is calculated by Manhattan distance. On the self-built dataset, the proposed SGAM (Symmetrical Geographic Address Matching) model improves the matching accuracy (91.22%) by 4%~10% in comparison to TextRCNN, FastText, attention-based convolutional neural network (ABCNN) and other models, proving that the SGAM model has better performance on the address matching task.

Farm robot path planning based on improved ant colony algorithm

ZHAO Guang-yuan, ZHAO Ying

2022, 44(05): 910-915. doi:

Abstract ( 93 )

PDF (1155KB) ( 151 ) 　　

The path planning of farm inspection robots is the key to realize intelligent monitoring of large-scale farms. Aiming at the problem of finding the optimal charging route during robot inspections, an improved ant colony algorithm is proposed. This algorithm uses the global information of the working environment to establish a target attraction function, guides the ant colony to choose the best path to reach the target point, and reduces the iteration time of the algorithm. By adding additional pheromone update items and improving the pheromone volatilization coefficient, the global search capability of the algorithm is enhanced to avoid the premature convergence in the later stage of the algorithm search and falling into the local optimum. Simulation experiments in simple and complex environments show that, compared with the classic ant colony algorithm, the algorithm has faster convergence speed and good stability, and can quickly converge to the best path.

A BN parameter learning algorithm based on variable weight fusion for small datasets

GUO Wen-qiang, KOU Xin, LI Meng-ran, HOU Yong-yan, XIAO Qin-kun

2022, 44(05): 916-923. doi:

Abstract ( 84 )

PDF (864KB) ( 99 ) 　　

Aiming at the problem of low accuracy of Bayesian network (BN) parameter learning results under the condition of small datasets, the necessity of variable weight design of BN parameters under the condition of small datasets is analyzed, and a BN parameter learning algorithm based on variable weight fusion, named VWPL, is proposed. Firstly, the inequality constraints are determined according to the expert experience, the minimum sample data set threshold is calculated and the variable weight factor function that changes with the sample size is designed. Then, the initial parameter set is calculated according to the sample, and the parameter expansion is carried out by the Bootstrap method to obtain the candidate parameter set that meets the constraints, and the final BN parameters can be obtained by substituting them into the BN variable weight parameter calculation model. The experimental results show that when the amount of learning data is small, the learning accuracy of the VWPL algorithm is higher than that of the MLE algorithm, the QMAP algorithm and the fixed-weight learning algorithm. In addition, the VWPL algorithm is successfully applied to the bearing fault diagnosis experiment, which provides a method of BN parameter estimation for small datasets.

An academic rising star prediction method based on multi-graph convolutional neural network and attention mechanism

SHAN Hui, DING Cheng-xin, ZHAO Zhong-ying, ZHOU Ming-cheng, JIA Xiao-sheng, LI Chao,

2022, 44(05): 924-932. doi:

Abstract ( 139 )

PDF (900KB) ( 143 ) 　　

Identifying potential academic rising stars from academic newcomers can provide decision support for tasks such as talent introduction, project review, and expert database construction, which has important research significance and application value and has received extensive attention from the academic community. However, the existing academic rising star prediction methods do not organically combine the academic cooperation relationship and individual attribute information, resulting in low accuracy. To solve the above problem, this paper proposes an academic rising star prediction method MGCNA based on multi-graph convolutional neural network and attention mechanism.It comprehensively considers cooperative networks and similar networks. Based on the two networks, the graph convolutional neural network is used to learn the authors feature representation, and then the attention mechanism is used for information fusion, so as to predict the academic rising stars with high potential. Finally, experiments are carried out on real datasets from the ArnetMiner platform, and the experimental results demonstrate the effectiveness of MGCNA in predicting academic rsing star tasks.

A group recommendation algorithm based on non-negative matrix factorization

JIA Jun-jie, YAO Ye-wang, CHEN Wang-hu

2022, 44(05): 933-943. doi:

Abstract ( 95 )

PDF (840KB) ( 131 ) 　　

In recent years, with the rapid development of media technology, the phenomenon of peoples group activities has gradually increased, and the group recommendation system has gradually attracted attention. Existing group recommendation systems often treat different members as homogeneous objects, ignoring the relationship between members professional backgrounds and inherent attri- butes of items, and cannot really solve the problem of preference conflicts in the fusion process. Therefore, a group recommendation algorithm based on non-negative matrix factorization is proposed. The algorithm decomposes the group rating information into the user matrix and the item matrix by non- negative matrix factorization. According to the two matrices, the item membership matrix and member professionalism matrix are calculated by using membership and professionalism weights respectively, and the contribution degree of each member on different items is obtained to construct the group prefe- rence model. The experimental results show that the proposed algorithm still has high recommendation accuracy in the case of different group size and intra-group similarity.

Current Issue

Author center

Review center

Online journal