15 September 2025, Volume 51 Issue 9
-
Select all|
-
Computer Engineering. 2025, 51(9): 1-13. https://doi.org/10.19678/j.issn.1000-3428.0069416
Traffic signal control plays an important role in alleviating traffic congestion and improving urban commuting efficiency. In recent years, breakthroughs have been made in traffic signal control algorithms based on deep reinforcement learning using real-time traffic data as input. However, traffic data in real-world scenarios often involve data distortion. Traditional solutions use reinforcement learning algorithms to control signal lights after repairing distorted data. However, on the one hand, the dynamic phases of traffic signal introduces additional uncertainty to distortion repair, and on the other hand, distortion repair is difficult to combine with deep reinforcement learning frameworks to improve performance. To address these issues, a distorted traffic signal control model based on hidden state prediction, HCRL, is proposed. The HCRL model comprises encoding, control, and encoding prediction sub-models. By introducing a hidden state representation mechanism for signalized intersections, the HCRL model can adapt better to deep reinforcement learning frameworks and effectively express the control state of signalized intersections. In addition, the HCRL model uses a special transfer training method to avoid data distortion interference in the control sub-model. Two real datasets are used to verify the impact of data distortion on the intelligent signal light control algorithms. The experimental results show that the HCRL model outperforms the distortion-completion-based traffic signal control models in all distortion scenarios and distortion rates; further, it demonstrates strong robustness against data distortion when compared with other baseline models.
-
Computer Engineering. 2025, 51(9): 14-24. https://doi.org/10.19678/j.issn.1000-3428.0069534
Detecting defects on unstructured roads is important for road traffic safety; however, annotated datasets required for detection is limited. This study proposes the Multi-Augmentation with Memory (MAM) semi-supervised object detection algorithm to address the lack of annotated datasets for unstructured roads and the inability of existing models to learn from unlabeled data. First, a cache mechanism is introduced to store the positions of the bounding box regression information for unannotated images and images with pseudo annotations, avoiding computational resource wastage caused by subsequent matching. Second, the study proposes a hybrid data augmentation strategy that mixes the cached pseudo-labeled images with unlabeled images inputted into the student model, to enhance the model′s generalizability to new data and balance the scale distribution of images. The MAM semi-supervised object detection algorithm is not limited by the object detection model and better maintains the consistency of object bounding boxes, thus avoiding the need to compute consistency loss. Experimental results show that the MAM algorithm is superior to other fully supervised and semi-supervised learning algorithms. On a self-built unstructured road defect dataset, called Defect, the MAM algorithm achieves improvements of 6.8, 11.1, and 6.0 percentage points in terms of mean Average Precision (mAP) compared to those of the Soft Teacher algorithm in scenarios with annotation ratios of 10%, 20%, and 30%, respectively. On a self-built unstructured road pothole dataset, called Pothole, the MAM algorithm achieves mAP improvements of 5.8 and 4.3 percentage points compared to those of the Soft Teacher algorithm in scenarios with annotation ratios of 15% and 30%, respectively.
-
Computer Engineering. 2025, 51(9): 25-37. https://doi.org/10.19678/j.issn.1000-3428.0069836
To optimize Quality of Service (QoS), Mobile Edge Computing (MEC) has been deeply integrated into the Internet of Vehicle (IoV) to provide geographically proximal computing resources for vehicles, thereby reducing task processing latency and energy consumption. However, traditional MEC server deployment relies primarily on terrestrial Base Stations (BSs), resulting in high deployment costs and limited coverage, making it difficult to ensure uninterrupted services for all vehicles. Air-ground collaborative IoV technology has emerged as a solution to these challenges. Unmanned Aerial Vehicles (UAVs) can dynamically assist Road-Side Units (RSUs) using their flexibility in line-of-sight links, providing more flexible computing resources for vehicular users, thereby ensuring the continuity and efficiency of in-vehicle services. Therefore, this study proposes a Dynamic Vehicular Edge Task Offloading Method (DVETOM) based on air-ground collaboration. This method adopts a vehicle-road-air architecture, establishing Vehicle-to-RSU (V2R) and Vehicle-to-UAV (V2U) links. Transmission and computation models are constructed for three modes: local execution of vehicular tasks, offloading tasks to the RSU, and offloading tasks to the UAV. An objective function is established with the joint optimization goal of minimizing system latency and energy consumption. DVETOM transforms the task offloading problem into a Markov Decision Process (MDP) and optimizes the task offloading strategy by using the Distributed Deep Deterministic Policy Gradient (D4PG) algorithm based on Deep Reinforcement Learning (DRL). Compared with 5 benchmark methods, experimental results show that DVETOM outperforms existing methods by 3.45%—23.7% in terms of reducing system latency and 5.8%—23.47% in terms of reducing system energy consumption while improving QoS for vehicular users. In conclusion, DVETOM enhances the offloading of vehicular edge computing tasks within the IoV effectively. It offers IoV users a more efficient and energy-conserving solution, showcasing its extensive potential for application in intelligent transportation systems.
-
Computer Engineering. 2025, 51(9): 38-48. https://doi.org/10.19678/j.issn.1000-3428.0069842
This study proposes a traffic strategy to improve the efficiency of vehicles passing through unsignalized intersections under vehicle-infrastructure cooperation, with the objectives of reducing vehicle acceleration change rate and target vehicle travel time. The study establishes a vehicle road collaboration scenario, divides dynamic conflict areas and static conflict areas, defines model input parameters, constructs a vehicle traffic sequence model and vehicle motion state control model, and verifies the effectiveness of the models through SIMULINK simulation. In common and special traffic scenarios, the strategy reduces the average maximum acceleration change rate during the vehicle deceleration phase by 17.27% and 45.95%, average amplitude of vehicle acceleration change by 37.06% and 38.89%, average maximum acceleration by 37.53% and 48.2%, and average travel time by 41.33% and 44.31%, respectively. In addition, compared to similar algorithms in the literature, this strategy optimizes the average travel time by 42.82% and average vehicle speed by 45.8%. The optimization effect is significant, and both indicators are more balanced. Simultaneously, the vehicle speed does not fluctuate frequently, and the ride comfort improves. Therefore, this strategy significantly improves overall traffic efficiency without much sacrifice to the comfort performance of partial vehicles.
-
Computer Engineering. 2025, 51(9): 49-58. https://doi.org/10.19678/j.issn.1000-3428.0069784
The dynamic nature of tasks in Internet of Vehicles (IoV) environments increases the complexity of real-time computational offloading. To address the difficulty of completing real-time tasks in a timely manner owing to limited terrestrial network coverage in IoV scenarios, this study proposes a collaborative computational offloading approach for Satellite Vehicular Mobile Edge Computing Networks (SVMECN). First, a geometric relationship model between satellites and the ground is constructed to calculate the transmission rates between devices and satellites, as well as between terrestrial gateways and satellites. The task processing delay is computed based on this model. The model fully considers the real-time nature of tasks and dynamically adjusts for the impact of satellite movement on terrestrial data transmission. Through collaborative processing between satellites and terrestrial gateways, the latency requirements of in-vehicle applications are met. Second, the study proposes a collaborative computational offloading algorithm based on Pointer Attention Mechanism and Actor-Critic (ST-PART). This algorithm dynamically adjusts task priorities according to their real-time nature, offloads tasks for computation in order of priority, and dynamically selects and collaboratively processes tasks among different computing nodes to minimize task processing delays. The proposed algorithm is simulated in an SVMECN environment. Compared with traditional heuristic algorithms, the proposed algorithm improves operational efficiency. Experimental and analytical results indicate that the proposed algorithm can significantly reduce task processing delays while meeting the real-time requirements of tasks. Compared with algorithms without collaboration between terrestrial and satellite components, the proposed algorithm can reduce latency costs by 2.35%-68.68%.
-
Computer Engineering. 2025, 51(9): 59-70. https://doi.org/10.19678/j.issn.1000-3428.0069349
The identification of abnormal individuals in videos is an important research topic in the field of computer vision. Existing algorithms primarily focus on detecting the outbreak phase of abnormal behaviors but overlook their developmental stage. Moreover, they suffer from issues such as unclear definitions of abnormalities, poor interpretability, and weak generalizability across application scenarios. To address these problems, this study proposes a knowledge graph-based model for the early identification of abnormal individuals. The model performs pedestrian detection and tracking, pedestrian visual attention target detection, and pedestrian behavior recognition from videos to capture pedestrian attribute features related to abnormal behaviors. Moreover, the study establishes a knowledge graph network targeting abnormal individuals and proposes four node modeling algorithms, including those for age attributes and social distance. Nodes are modeled based on pedestrian attributes to better analyze the characteristics of abnormal individuals during the developmental stage of abnormal behaviors. Additionally, the study proposes three abnormal individual reasoning algorithms based on node state transitions for child abduction, theft, robbery, and fighting. These algorithms perform state reasoning on knowledge graph nodes to derive the probability of an individual engaging in abnormal behavior in the future, thereby enabling the early identification of abnormal individuals. The reasoning algorithms adopted enhance the interpretability of the model. An early abnormal individual detection dataset is created and annotated, defining four types of abnormal behaviors: theft, fighting, robbery, and child abduction. The samples in the dataset are sourced from various shooting scenarios. The effectiveness of the model is evaluated on this dataset, and the experimental results show that the model achieves a mean Average Precision (mAP) of 22.83%, outperforming other mainstream behavior recognition models. Specifically, it demonstrates an 18.96 percentage point improvement over the SlowFast model, indicating that the proposed model can effectively identify abnormal individuals before the outbreak of abnormal behaviors and is generalizable across application scenarios.
-
Computer Engineering. 2025, 51(9): 71-79. https://doi.org/10.19678/j.issn.1000-3428.0069598
In recent years, Artificial Intelligence (AI) has been widely used in fields such as large models, autonomous driving, and robotics. As the core of AI, neural networks process big data, learn, adapt complex patterns, and perform various tasks. For the implementation of neural networks, convolution algorithms are often used to extract local features of the input data to help them learn to understand the structure and pattern of data such as images and sounds. However, convolution computation involves intensive multiplication and accumulation operations and is a time intensive process, thus becoming a major obstacle for the real-time implementation of neural networks. In this study, to accelerate the convolution algorithm at the hardware level, a Winograd convolution acceleration operator based on a heterogeneous sampling window, which adopts a heterogeneous 4 × 2 sampling window to improve data utilization, is proposed. Additionally, a Winograd hardware acceleration module is designed using a pipeline and fixed-point structure, and a ReLU module based on pooling fusion is proposed. A prototype verification experiment is conducted on a Field Programmable Gate Array (FPGA). Finally, the acceleration ratio of the single-channel original sliding window convolutions is 86.4 and that of the three-channel sliding window convolutions is 28.8. The amount of read and write data is reduced to 11.07% of the original, and the resource consumption is lower than that of the Winograd convolution acceleration operators of the same type. It has the ability to integrate and build convolutional neural networks on a large scale, and compared with the Fast Fourier Transformation (FFT), it has distinct advantages.
-
Computer Engineering. 2025, 51(9): 80-90. https://doi.org/10.19678/j.issn.1000-3428.0069705
Multimodal Aspect-Based Sentiment Analysis (MABSA) aims to analyze the sentiment polarity of aspect words derived from text and image pairs. Existing methods primarily focus on extracting emotional features from both images and texts. However, the various features of images and texts may not necessarily be effective for the final sentiment analysis. Both images and text often contain a large amount of redundant and noisy information outside the areas related to aspect words, and different regions of images and text may be related to different aspect words. In the process of approximately establishing image text feature extraction, noise is introduced into multimodal aspect-level sentiment analysis tasks. In addition, the sentiment polarity related to aspect words in images and text may still be the opposite, implying an interactive information between the two. To address these issues, this paper proposes a multimodal aspect-level sentiment analysis model that combines local perception and multilevel attention. Specifically, the local perception module is designed to simultaneously select text content and image regions that are semantically relevant to aspect words. Subsequently, to improve the accuracy of sentiment aggregation, a multilevel attention module is introduced into the model, which uses a bottleneck attention mechanism to extract modal interaction information. The experimental results show that the model achieves State-Of-The-Art (SOTA) performance on the Twitter2015, Twitter2017, and Multi-ZOL datasets, significantly outperforms similar models.
-
Computer Engineering. 2025, 51(9): 91-100. https://doi.org/10.19678/j.issn.1000-3428.0069347
Event extraction is an important information extraction task that aims to extract specific events or information from natural language texts. There are many overlapping event problems, where one word is used as a trigger for different event types or when event arguments for different roles in real-life event extraction scenarios. However, existing overlapping event extraction methods ignore the correlations and dependencies between event elements, such as event types and argument roles, resulting in a poor performance of overlapping event extraction. To solve this problem, this paper proposes an event extraction model via cascade decoding enhanced by dynamic heterogeneous graphs, named DHG-EE, which can effectively realize the structural representation of overlapping events and facilitates information transmission between event elements through a multi-granularity cascade decoding structure and a domain-event type-argument role heterogeneous graph network. First, the pre-trained model encodes the natural language text and constructs a multi-granularity heterogeneous graph network composed of domains, event types, and argument roles, which separates the overlapping event arguments from the corresponding multiple domain nodes and event-type nodes and efficiently represents the complex associations of overlapping events through the dynamic point-edge structure of the heterogeneous graph. Then, the multi-granularity cascading decoding structure decodes domain attributes, event types, event trigger words, and event arguments, in order from coarse to fine, according to semantic granularity and uses the information of the previous granularity as additional information to assist in the decoding of the next granularity. Experimental results show that the F1 value of the proposed model is better than that of the baseline models on the FewFC and DuEE1.0 benchmark event extraction datasets.
-
Computer Engineering. 2025, 51(9): 101-109. https://doi.org/10.19678/j.issn.1000-3428.0068953
To fully utilize the relationships between items, current recommendation algorithms introduce knowledge graphs to enrich the features of both items and users. However, most knowledge-graph-based recommendation algorithms often overlook the connection between the current hop triple set, the initial seed, and the previous hop triple set, leading to inaccurate feature representations of users and constructed items. To address these limitations, this study proposes a knowledge graph recommendation model that combines pre-existing triple sets. It generates initial representations of users and items based on heterogeneous propagation strategies and combines the relationship between the current hop triple set, the initial seed, and the previous hop triple set to control the representation of each hop triple set. Consequently, the final representations of users and items are generated, and the probability of interaction between users and items is predicted based on thees final representations. On datasets consisting of books and movies, the proposed model outperforms the current advanced knowledge-graph-based recommendation models in terms of Area Under the Curve (AUC), F1 score, and recall.
-
Computer Engineering. 2025, 51(9): 110-119. https://doi.org/10.19678/j.issn.1000-3428.0069711
Aiming at non-affine nonlinear multi-Agent systems with full-state constraints, this study investigates an event-triggered formation control strategy with prescribed performance. The study proposes a barrier function-based nonlinear mapping technique to transform full-state constraints into the boundedness of mapped variables, thereby eliminating feasibility conditions in the controller design. Then, it introduces a shift function and a prescribed time-convergent performance function to constrain the formation tracking error. Consequently, the restriction that the initial value of the formation tracking error must be within the performance constraint range is eliminated, thus improving formation performance. The study also designs an event-triggered prescribed performance formation controller to guarantee that Agents achieve the desired formation within a prescribed time and maintain it thereafter, while significantly reducing controller—actuator signal transmissions. Lyapunov stability analysis proves that all signals in the system are semi-globally, uniformly, and ultimately bounded. The theoretical analysis rules out the possibility of Zeno behavior occurring. Finally, numerical simulations verify the effectiveness of the proposed method.
-
Computer Engineering. 2025, 51(9): 120-128. https://doi.org/10.19678/j.issn.1000-3428.0069414
Event Aspect Category Detection (ACD) aims to identify the aspect categories present in event text. Data need to be collected from various fields and textual events, particularly when researching public opinion on social media. The first phase of social media opinion events lacks sufficient data for labeling event text. The pressing issue is precisely detecting event aspects using a limited amount of labeled data. This paper presents a novel method for event ACD with limited samples. This method utilizes a pre-trained model to construct soft prompt templates, performs hierarchical semantic characterization and interaction fusion, and adaptively combines multilayer prompt characterizations. The objective is to enhance the accuracy of event ACD using limited samples. Experiments on a self-constructed Chinese social media dataset and English dataset demonstrate that the proposed method is significantly superior to other baseline methods. Further ablation experiments and visualizations confirm the effectiveness of the proposed multilayer prompt interaction fusion module.
-
Computer Engineering. 2025, 51(9): 129-138. https://doi.org/10.19678/j.issn.1000-3428.0069410
Text-to-SQL semantic parsing task aims to transform natural language problems into executable SQL statements. In recent years, many researchers have applied methods such as pre-training models to this task and have made some progress. However, because existing pre-training models are not re-trained for Text-to-SQL tasks, they cannot adapt well to the scene semantic feature information of the task, which affects the parsing performance of the models. At the same time, many methods are prone to ignoring the relationship between natural language questions and database schemes, which results in semantic ambiguities in the parsing process. To solve these problems, this paper proposes a new RGA-T5 model for Text-to-SQL semantic parsing, which introduces a relation-aware Heterogeneous Graph Neural Network (HGNN) into the pre-training model T5, constructs the input entities and relations as nodes on the heterogeneous graph, and realizes semantic relation-awareness of the input sequences of the model by applying the Graph Neural Network (GNN). Simultaneously, the method also proposes a spatial gating adapter, the parameters of which are trained to realize fine-tuning of the model so that the model can be adapted to the semantic feature information in different scenarios for this task and reduce the introduction of irrelevant information. The experimental results show that the proposed method improves the performance over other advanced Text-to-SQL parsing methods on the Spider dataset, thereby verifying the model's effectiveness.
-
Computer Engineering. 2025, 51(9): 139-148. https://doi.org/10.19678/j.issn.1000-3428.0069439
Accurate traffic flow prediction is a key prerequisite for realizing intelligent transportation systems, and is of great significance for strengthening system simulation and control and improving the decision-making of managers. To address the problem of most existing Graph Convolutional Network (GCN) models ignoring the dynamic spatial and temporal variations in traffic data and insufficiently employing node information, which leads to insufficient extraction of spatial and temporal correlations, a traffic flow prediction model based on multiple spatio-temporal graph fusion and dynamic attention is proposed. First, the temporal characteristics of traffic flow data in multi-temporal states are extracted by different convolutional cells. The next step involves constructing a multiple spatio-temporal graph to capture the dynamic trend and heterogeneity of nodes in spatial distribution, followed by extracting spatial characteristics through the integration of GCN. Finally, the spatial and temporal characteristics are analyzed and fused using the multi-head self-attention mechanism to output prediction results. Experimental analyses are performed on two public datasets, PeMS04 and PeMS08, and compared with the Attention Based Spatial-Temporal Graph Convolutional Network (ASTGCN), Multiview Spatial-Temporal Transformer Network (MVSTT), Dynamic Spatial-Temporal Aware Graph Neural Network (DSTAGNN) and other benchmark models that utilize spatio-temporal graph convolution. The results show that the Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE) of the proposed model are reduced by 7.10%, 7.22%, and 6.47%, respectively, demonstrating the proposed model′s strong adaptability and robustness.
-
Computer Engineering. 2025, 51(9): 149-157. https://doi.org/10.19678/j.issn.1000-3428.0069306
Currently, in deep learning-based smart contract vulnerability detection solutions, the direct use of bytecode or source code for textual sequence feature representation lacks a comprehensive understanding of program semantics. The smart contract vulnerability detection technology based on Abstract Syntax Tree (AST) embedding fully considers the syntax and semantic features needed for contract vectorization and appropriate processing granularity, enabling more accurate capturing of smart contract vulnerability features. First, it employs Solidity syntax tree parsing to design a smart-contract vectorization method based on AST embedding. It partitions node types recursively at the statement level to generate sequences of statement trees. Subsequently, a recursive neural network is employed to encode each statement tree from the bottom up, transforming the intricate AST structure into statement-level feature vectors. Building on this foundation, a Bidirectional Gated Recurrent neural network model with an Attention mechanism (BiGRU-ATT) is constructed. This facilitates the learning of features from the sequences of statement trees and accomplishes the detection and categorization of five typical vulnerabilities: re-entrancy, unchecked return values, timestamp dependency, access control, and denial-of-service attacks. Experimental results demonstrate that the proposed method improves the micro-F1 and macro-F1 metrics by 13 and 10 percentage points, respectively, compared to the direct vectorization of source code as a text sequence. In tasks related to timestamp dependence, access control, and denial-of-service attack vulnerability classification, the BiGRU-ATT model with the attention mechanism achieves an F1 value of over 88%.
-
Computer Engineering. 2025, 51(9): 158-165. https://doi.org/10.19678/j.issn.1000-3428.0069537
Messages transmitted through the Internet of Vehicle (IoV) are in an open network environment; therefore, they are vulnerable to attacks and privacy leakage. To address these problems, this paper proposes an anonymous traceable message authentication scheme based on an identity-based cryptographic mechanism. First, an Identity-Based Signature (IBS) algorithm is designed to ensure the integrity of the transmitted message, so that the legitimacy and non-repudiation of the message sender′s identity can be verified. Second, the scheme adopts an anonymous mechanism to provide privacy preservation, which is generated by the vehicle and is used to obtain the private key; because the real identity is not transmitted in the network, no one can obtain the real identity of the vehicle except the vehicle and the authority. In addition, the generated private key is encrypted to ensure the confidentiality of the private key transmission. Finally, this paper proves that this scheme has existential unforgeability against chosen-message attacks in the random oracle model, which can meet the security requirements of the IoV. The simulation results show that the scheme has lower computational and communication overheads than similar schemes.
-
Computer Engineering. 2025, 51(9): 166-176. https://doi.org/10.19678/j.issn.1000-3428.0069162
For improving system security and stability, multivariate time series anomaly detection methods are generally used to detect and identify abnormal patterns and behaviors in systems. To mitigate reduced anomaly detection accuracy caused by complex dependencies within multivariate time series, a multivariate time series anomaly detection model, HGAT, is proposed. This model is based on a graph attention network and is optimized by combining prediction and reconstruction methods. First, graph attention networks are used to capture the temporal and spatial dependencies of multivariate time series. Second, a Transformer that integrates Variational Autoencoders (VAE) is used as the reconstruction module and a Time Convolutional Network (TCN) is used as the prediction module to jointly detect abnormal sequences. The self-attention mechanism of the Transformer enables the reconstruction module to model the entire time series by directly considering the relationship between any two positions in the sequence to capture the global dependency relationship of the sequence. TCN can effectively extract local features of the time series by stacking convolutional layers and increasing the receptive field. Finally, by comprehensively considering the reconstruction and prediction modules via abnormal scores, the overall distribution of the sequence is analyzed from both global and local perspectives based on spatiotemporal joint representation. Experiments are conducted on the SMAP, MSL, and SMD datasets, and the results show that the F1 values for the HGAT are 96.20%, 97.50%, and 92.85%, respectively. These values are superior to those for the baseline method.
-
Computer Engineering. 2025, 51(9): 177-191. https://doi.org/10.19678/j.issn.1000-3428.0068904
Unfair issues arise in secret sharing owing to the insufficient consideration given to participants′self-interested behaviors, such as honest participants being unable to reconstruct the shared secret while dishonest participants are able to. To address this issue, this study proposes a secret sharing scheme called FVSS by integrating secret sharing with blockchain technology and leveraging smart contracts. First, the study constructs a secret share obfuscation mechanism that binds user passwords with secret values to distribute false shadow secret shares to participants, thereby protecting the sharing of real shares from guessing attacks. Second, it designs a method for verifying the integrity of shadow secret shares based on polynomial commitments, enabling the bidirectional verifiability of shadow secret shares among participants and ensuring the effectiveness of mutual supervision. Subsequently, to achieve targeted fairness guarantees in secret sharing, the study establishes a fairness incentive-penalty strategy based on smart contracts to motivate participants to reconstruct secrets efficiently and monitor malicious third-party distribution behaviors effectively. Finally, the study conducts theoretical analyses and experimental validation of the verifiability, fairness, security, and resource overhead of the proposed scheme. Analysis and experimental results demonstrate that the scheme can effectively constrain the self-interested behaviors of malicious participants in secret sharing, resist known attacks, and provide greater security by supporting user passwords and commitment values. Moreover, the scheme′s average time overhead is at the millisecond level, indicating good practicality and scalability.
-
Computer Engineering. 2025, 51(9): 192-200. https://doi.org/10.19678/j.issn.1000-3428.0070021
As user awareness of privacy protection increases, an increasing number of websites and services are employing the Transport Layer Security (TLS) protocol to safeguard user data. Consequently, the proportion of TLS-encrypted traffic within overall network traffic is steadily increasing. However, most current abnormal traffic detection methods are general-purpose models that target all traffic or all encrypted traffic. Methods that specifically focus on TLS-encrypted traffic are few. Therefore, this study proposes a supervised autoencoder-based method for detecting abnormal TLS-encrypted traffic. This method focuses on training a supervised autoencoder that uses network traffic as the input and generates reconstructed traffic with the same dimensionality as that of the input. The model requires extremely high similarity between normal traffic and its corresponding reconstructed traffic, whereas the similarity between abnormal traffic and its reconstructed counterpart should be extremely low. To achieve these reconstruction requirements, a reconstruction loss function is designed to supervise and optimize the internal parameters of the autoencoder. During the detection phase, the reconstruction capability of the autoencoder is utilized to determine whether the input traffic is abnormal, by measuring the cosine similarity between the input and reconstructed traffic. Furthermore, a specialized dataset tailored for TLS-encrypted abnormal traffic detection is constructed by integrating relevant data. Experimental results on this dataset demonstrate that the proposed method achieves an accuracy of 99.52% in the binary classification task of TLS-encrypted abnormal traffic detection, outperforming other comparative models. In addition, various visualization strategies are employed to demonstrate the effectiveness of the proposed method.
-
Computer Engineering. 2025, 51(9): 201-212. https://doi.org/10.19678/j.issn.1000-3428.0069365
Ultra-dense Wireless Body Area Networks (WBAN) integrated into 6G networks face the issues of scarce computational resources and co-channel interference. This study proposes an interference-aware efficient task offloading strategy to address these challenges. First, a Software Defined Network (SDN)-based edge computing architecture for WBAN medical information is designed, and a priority scoring mechanism and priority queuing model that consider historical states are established. Second, a Neighbor Node Aware Algorithm (NNAA) is proposed to generate a neighbor node matrix for the current superframe node. Subsequently, an Interference-Aware Offloading Strategy (IAOS) is introduced, which defines a system benefit model that considers offloading gains, offloading overheads, and offloading states. Next, an objective function that considers both the system benefits and the number of concurrent nodes is designed. Crossover and mutation strategies from genetic algorithms are incorporated to escape local optima, and corrections are made for infeasible solutions. Finally, an Improved Binary Kepler Optimization Algorithm (IBKOA) is used to derive the offloading decision that maximizes the objective function. Experimental results demonstrate that in environments with varying data volumes, the IAOS strategy reduces the latency by an average of 74.5% compared with other algorithms. In environments with varying numbers of patients, the IAOS strategy achieves average reductions of 61.43%, 59.28%, and 58% in node interference rate, latency, and energy consumption, respectively, compared with the comparison algorithms, while increasing the throughput and system benefits by average values of 149.5% and 74.38%, respectively.
-
Computer Engineering. 2025, 51(9): 213-219. https://doi.org/10.19678/j.issn.1000-3428.0069171
A high-speed railway wireless channel is a highly time-varying wireless channel, posing challenges to the enforcement of delay constraints for passengers in practical scenarios. Within the framework of the Cloud Radio Access Networks (C-RAN) architecture for high-speed railways, this study constructs an energy efficiency maximization problem that simultaneously satisfies the statistical delay constraints of user services and forward link capacity limitations. Considering the characteristics of high-speed trains, such as their deterministic movement direction, fixed trajectory, and predictable path loss, the model is established by utilizing the path loss information instead of traditional channel state information, thereby effectively reducing the complexity of the energy efficiency maximization problem. As the original problem is a non-convex optimization problem, it is first transformed into an equivalent convex optimization problem. Subsequently, a two-layer iterative power allocation algorithm is designed, and the Lagrangian duality method is employed to solve the inner subproblem. Simulation validations are conducted under varying train speeds, Quality of Service (QoS) indices, and numbers of users. The results demonstrate that the proposed algorithm not only strictly adheres to the statistical delay constraints but also significantly enhances the energy efficiency of the system. Thus, this study provides feasible solutions and theoretical support for green and efficient resource allocation tasks in high-speed railway C-RAN systems.
-
Computer Engineering. 2025, 51(9): 220-230. https://doi.org/10.19678/j.issn.1000-3428.0069092
Existing Hyper-Spectral Image (HSI) super-resolution tasks suffer from issues such as inadequate information representation, limited utilization of prior knowledge, and low reconstruction accuracy. These problems significantly affect the accuracy of subsequent image-processing tasks. To address this challenge, this study proposes a HSI super-resolution reconstruction algorithm based on spatial-spectral prior and high-order tensor representation. The algorithm divides image into non-local similar blocks using the k-means algorithm and constructs them into high-order tensors. The balanced decomposition strategy of the tensor-train is used to exploit the low-rank redundancy in these non-local high-order tensors. In addition, considering the local smoothness property in the image spatial domain, a weighted group sparse regularization term is employed. Furthermore, a weighted spectral unmixing constraint term is used to address fusion distortion issues in spectral data during processes, and the calculation process for each variable is presented using the alternating direction method of multipliers. In experimental evaluations on three publicly available real datasets—CAVE, Pavia University, and Indian Pines, the proposed algorithm improves the average Peak-Signal-to-Noise-Ratio (PSNR), Structural Similarity Index (SSIM) by 0.290 8 dB, 0.002, Spectral Angle Mapper (SAM) and global relative error index are reduced by 0.116° and 3.1%, respectively.
-
Computer Engineering. 2025, 51(9): 231-241. https://doi.org/10.19678/j.issn.1000-3428.0069154
Traffic scene graph plays an important role in structurally representing traffic scenes. Current methods for scene graph generation predict relationships between entities to generate unbiased scene graphs. However, with existing methods, the long-tailed distribution of datasets and ambiguous feature representation of entity relationships result in traffic scene graphs that fail to provide accurate and meaningful traffic scene information for downstream tasks. To address these issues, this study proposes a Contextual Semantic Embedding (CSE) and Coarse-Fine-Grained Blending (CFGB) traffic scene graph generation network CSE-CFGB. Specifically, the CSE module is used to establish the unique semantic representations of entities and predicates. Subsequently, the CFGB network is employed to robustly predict relationships between entities. The Main Branch (MB) utilizes CSE to directly predict relationships between entities; the Coarse-grained Branch (CB) is responsible for learning robust features of head predicates using a reweighting mechanism; and the Fine-grained Branch (FB) refines the learning of tail predicates using a Logit adjustment method. Additionally, a branch weights table is incorporated to facilitate cooperation between the two auxiliary branches and help balance the prediction performance of the head and tail predicates by the MB. In experimental evaluations conducted on the Visual Genome dataset, the proposed scene graph generation network achieved excellent performance in the PredCls task, with average performance metrics Mean@50和Mean@100 reaching 49.5% and 51.7%, respectively. The experimental results indicate that the proposed method addresses the issues of ambiguous entity relationship representation and long-tailed distributions in a dataset during model training.
-
Computer Engineering. 2025, 51(9): 242-251. https://doi.org/10.19678/j.issn.1000-3428.0069754
Few-shot learning aims to classify new categories based on only one or a few examples. To address this problem, data augmentation is often used as a direct and effective approach. Further, the augmented data should be diverse and discriminable. This paper proposes a new two-stage data augmentation method based on the transfer of intra-class variations in the base classes. The learning process is divided into a representation learning stage and a few-shot learning stage. In the representation learning stage, representations of instance-specific features of the base-class data are obtained through self-supervised tasks and class-specific features are obtained through supervised tasks. Intra-class variations of the base-class data are calculated using these two features, and the distributions of intra-class variations for each base class are modeled. In the few-shot learning stage, the model samples task-related intra-class variation information from the intra-class variation distributions of the base class and adds it to the few-shot features to enhance the few-shot data. The experimental results show that the 5-way 1-shot classification performance of the proposed method on the miniImageNet, tieredImageNet, and CUB datasets is improved by 4 to 7 percentage points compared to the baseline model, and by 3 to 7 percentage points in 5-way 5-shot classification. Its performance is competitive when compared with other existing data augmentation methods. This indicates that the generated enhanced data can improve the diversity of few-shot data while maintaining discriminability, verifying our method's feasibility and effectiveness.
-
Computer Engineering. 2025, 51(9): 252-267. https://doi.org/10.19678/j.issn.1000-3428.0069458
Automatic and accurate segmentation of renal arterial vessels from pathologic whole section images is crucial and a prerequisite for the diagnosis of renal diseases. Most existing methods focus on detecting and segmenting prominent glomeruli. Studies focusing on segmenting arterial vessels are scarce because of the highly variable morphological appearance and unclear boundaries of such vessels. To address these issues, a cascade detection and segmentation framework is proposed for accurate segmentation and quantitative analysis of renal arterial vessels. In the first stage, a multi-window adaptively calibrated Renal Artery Detection Network (RADNet) is constructed to locate the renal artery region. In the second stage, a segmentation network is designed to accurately segment the renal artery walls and lumens by combining efficient channel spatial attention and a visual Transformer into a convolutional network (U-Net). Finally, a quantitative analysis is performed by calculating the correlation between the quantitative results and clinical information. The detection network adopts a multiscale adaptive calibration method, which can localize the arterial region; the segmentation network utilizes the Transformer and efficient channel and spatial attention mechanisms to better extract arterial walls and lumens with complex morphological appearances and ill-defined boundaries. The experimental results show that the proposed framework achieves significant improvements in the detection and segmentation of small arterial vessels compared to previous models. In addition, this method has great potential for clinical applications in the segmentation and quantification of small lesions in medical images.
-
Computer Engineering. 2025, 51(9): 268-279. https://doi.org/10.19678/j.issn.1000-3428.0069450
When processing fog scenes, current image dehazing methods can remove fogs from an image but they tend to cause the color distribution of the image to shift significantly during the restoration process. This study proposes a polarization image dehazing algorithm based on target scene color restoration to address the problem of color distortion in traditional polarization dehazing algorithms. By leveraging the characteristics of low saturation and high brightness of atmospheric light and combining them with the total light intensity and polarization difference diagrams, atmospheric light value at infinity is estimated to reduce the influence of white-highlighted objects. The estimated position of atmospheric light at infinity is used to extract the polarization angle of atmospheric light, and the polarization matrix of atmospheric light varying with depth is automatically estimated using the Stokes vector. Based on the spatial correlation at the edge jumps of atmospheric scattered light, regularization constraints are constructed using atmospheric polarization and image chromaticity to correct the atmospheric scattered light. In addition, to improve the color fidelity of the target scene, this study proposes an adaptive color equalization method to improve the color distribution of the restored results. Compared with classical dehazing algorithms such as DCP, BCCR, GPLPF, PLF, and POBS, the proposed algorithm improves the NIQE, BRISQUE, and color deviation factor K by 9.62%, 13.49%, and 40.13%, respectively. Moreover, it has good restoration effects for scenes with different haze concentrations, effectively improving the visibility of targets at different depths and avoiding color distortion, particularly for dense fog scenes.
-
Computer Engineering. 2025, 51(9): 280-293. https://doi.org/10.19678/j.issn.1000-3428.0069353
In Unmanned Aerial Vehicle (UAV) aerial photography, targets are usually small targets with dense distribution and unobvious features, and the object scale varies greatly. Therefore, the problems of missing detection and false detection are easy to occur in object detection. In order to solve these problems, a lightweight small object detection algorithm based on improved YOLOv8n, namely PECS-YOLO, is proposed for aerial photography. By adding P2 small object detection layer in the Neck part, the algorithm combines shallow and deep feature maps to better capture details of small targets. A lightweight convolution, namely PartialConv, is introduced to a new structure of Cross Stage Partial PartialConv (CSPPC), to replace Concatenation with Fusion (C2f) in the Neck network to realized lightweight of the model. By using a model of Spatial Pyramid Pooling with Efficient Layer Aggregation Network (SPPELAN), small object features can be captured effectively. By adding Squeeze-and-Excitation (SE)attention mechanism in front of each detection head in the Neck part, the network can better focus on useful channels and reduce the interference of background noise on small object detection tasks in complex environments. Finally, EfficiCIoU is used as the boundary frame loss function, and the shape difference of the boundary frame is also taken into account, which enhances the detection ability of the model for small targets. Experimental results show that, compared YOLOv8n, the mean Average Precision at Intersection over Union (IoU) of 0.5 (mAP@0.5) and the mean Average Precision at IoU of 0.5∶0.95 (mAP@0.5∶0.95) of PECS-YOLO object detection algorithm on VisDrone2019-DET dataset are increased by 3.5% and 3.7% respectively, the number of parameters is reduced by about 25.7%, and detection speed is increased by about 65.2%. In summary, PECS-YOLO model is suitable for small object detection in UAV aerial photography.
-
Computer Engineering. 2025, 51(9): 294-305. https://doi.org/10.19678/j.issn.1000-3428.0069459
To address the challenges of small object blurring, large object scale difference, and category imbalance in ship detection, this paper designs a dynamic ″copy-paste″ data augmentation method, embeds it into the YOLOv5 model, and proposes an improved YOLOv5s algorithm for sea surface object detection. In the backbone network, a shallow local perception module is introduced to improve the receptive field by combining a hybrid dilated convolution, depthwise separable convolution, and residual connection branch in parallel. This enhances the extraction of detailed local information. In the neck network, an attention fusion module is designed to aggregate shallow spatial information and deep semantic information using spatial and channel attention mechanisms, respectively. This improves the feature expression capability of the network. For the detection head, a hierarchical fusion decoupling head is designed by downsampling and fusing features from the adjacent shallow detection head to enhance object classification and positioning accuracy. The dynamic ″copy-paste″ data augmentation strategy involves extracting objects from training set images and storing them in a target sample library. During each training epoch, targets are randomly selected from this library based on their probability distribution values. After applying geometric and photometric transformations in certain proportions, these targets are pasted into the training images to increase the foreground target density. The SMD-Plus dataset is used for experimental verification. The experimental results show that mAP@0.5 and mAP@0.5 ∶95 values for the proposed algorithm are improved by 6.7 and 5.2 percentage points, respectively, compared with the YOLOv5s model. Migration experiments are conducted on the WSODD dataset, and mAP@0.5 and mAP@0.5 ∶95 values are improved by 3.7 and 3.3 percentage points, respectively. Additionally, the improved algorithm and the proposed dynamic data augmentation method alleviate the problems of class and size imbalance, improve the detection accuracy of small targets, and are suitable for ship detection tasks.
-
Computer Engineering. 2025, 51(9): 306-316. https://doi.org/10.19678/j.issn.1000-3428.0069426
Real-time simulation and evaluation of carbon reduction effects of signal control strategies is a necessary prerequisite for strategy optimization. In existing studies, carbon emission calculation models that do not couple with traffic simulation or only couple with macroscopic and mesoscopic traffic simulations are hindered by low-resolution or insufficient controllability for strategy evaluation. Models that couple with microscopic traffic simulation have the problems such as significant differences between the carbon emission module and the characteristics of Chinese car models, unreasonable division of emission calculation segments, and failure to consider actual vehicular operating conditions. Therefore, this article proposes a microscopic co-simulation model of traffic flow and carbon emission for real-time control strategy evaluation. Based on the data transmission mechanism and vehicle model matching relationship, a microscopic traffic flow simulation model and a modified International Vehicle Emission (IVE) model are coupled, and the interface is used to achieve synchronous deduction and calculation of vehicle motion and carbon emissions. In terms of calibration, a two-stage parameter calibration method is designed. In the first stage, the calibration of emission model is carried out with consideration of vehicular motion characteristics. In the second stage, the calibration of traffic simulation is carried out with consideration of actual vehicular travelling conditions. The experimental results show that compared with the measured values, the simulated emission factors of minibuses and buses have errors of 10.1% and 15.6%, respectively. After implementing the ″green wave″ signal control optimization, the carbon emissions of vehicles significantly decrease. The coupled model can accurately measure transportation carbon emissions with high resolution and effectively evaluate transportation emission reduction measures.
-
Computer Engineering. 2025, 51(9): 317-327. https://doi.org/10.19678/j.issn.1000-3428.0069366
In the surrogate-assisted optimization of expensive objective functions, obtaining enough samples is time-consuming in fluid simulations. This paper proposes a Transfer Learning based Dual-Surrogate-Assisted hull form Optimization (TLDSAO) algorithm to reduce the required number of samples and improve the solving performance of the problem. Firstly, transfer learning is used to assist modeling during the surrogate construction phase, utilizing transfer of source domain knowledge to the target domain to reduce the demand for hull form samples. Secondly, the coarse surrogate and fine surrogate models are constructed for the sample data to perform the dual-surrogate-assisted optimization. For this purpose, an external public pool is introduced to exchange population information between the two surrogates to improve the search performance of the optimization algorithm. Finally, the proposed method is applied to a hull form optimization experiment of KCS. Experimental results show that under the same sample size, the addition of transfer learning significantly improves the accuracy of the surrogate model and can reduce the sample size by approximately half compared to the algorithm without transfer learning with the same accuracy. The TLDSAO algorithm reduces the total drag coefficient of KCS by 10.85%, with a prediction error of 2.93%. Compared with three comparison methods under the same conditions, the optimization results are further reduced by 0.61%, 6.11%, and 1.56%, respectively. Comparison methods achieve a performance comparable to that of the TLDSAO algorithm only when the sample size is increased by 40. Despite fewer samples, the TLDSAO algorithm achieves better solutions and lower prediction errors.
-
Computer Engineering. 2025, 51(9): 328-339. https://doi.org/10.19678/j.issn.1000-3428.0069559
To address the challenges of tight capacity and high delayed rate of meal delivery tasks during peak dining period, a real-time optimization policy based on Deep Reinforcement Learning (DRL) for instant meal delivery is proposed to improve the long-term customer service level of platforms. First, considering the constraints of meal preparation time, pickup and delivery sequence, and time window in meal delivery, the instant meal delivery problem with stochastic requests is modeled as a Markov Decision Process (MDP) to maximize the expected average customer service level. Second, the Proximity Policy Optimization (PPO) algorithm is combined with the Insertion Heuristic (IH) algorithm to design an instant meal delivery optimization policy, PPO-IH. A policy network with an integrated attention mechanism is employed by PPO-IH for matching orders to couriers, and the network is trained by the PPO algorithm. The courier routes are updated with an IH algorithm. Finally, through comparative experiments with the Greedy, minimum difference strategy, allocation heuristic, and two deep reinforcement learning algorithms, PPO-IH is shown to perform better in 71.5%, 95.5%, 87.5%, 79.5%, and 70.0% days with the given data, respectively. Additionally, PPO-IH achieves a higher average level of customer service, shorter average delivery times per order, and a lower rate of delayed deliveries. Furthermore, PPO-IH demonstrates certain effectiveness and generalization under different rider numbers, order densities, and order time window scenarios.
-
Computer Engineering. 2025, 51(9): 340-349. https://doi.org/10.19678/j.issn.1000-3428.0069139
Vehicle detection and identification is a key technology in the field of intelligent transportation and autonomous driving. It plays a vital role in road safety and autonomous driving and has always been a research hotspot. Target detection models based on deep learning have greatly improved vehicle detection accuracy, but challenges related to the accuracy and reliability of existing vehicle detection technology under adverse conditions, such as low light at night and bad weather, remain. To address these challenges, a vehicle detection method and model specifically targeted at nighttime dark conditions-MMD-YOLOv7-is proposed based on the YOLOv7 model. First, a new Multi-Channel Coordinate Attention (MCCA) module is constructed based on the Coordinate Attention (CA) mechanism, which significantly improves the model's ability to capture global and local feature information. Second, a Multi-Scale Convolution (MSC) module is innovatively designed and constructed to achieve targeted improvements to the Efficient Layer Aggregation Networks (ELAN) structure, allowing the model to better adapt to noise interference in the night visual environment, while improving the capability and accuracy of feature extraction. Finally, the Diverse Branch Block (DBB) is introduced to further enhance the model's ability to capture complex features. To verify the effectiveness of the proposed model, 6 000 night scene images in the BDD100K dataset are selected for training and testing. The experimental results show that the proposed model's vehicle detection accuracy is improved by 5.3 percentage points compared to the original YOLOv7 model, indicating its strong ability to handle low light conditions. In addition, it shows good performance on multiple public vehicle detection datasets, verifying its strong robustness and generalization capabilities.
-
Computer Engineering. 2025, 51(9): 350-361. https://doi.org/10.19678/j.issn.1000-3428.0069454
In the field of forest fire detection, existing target detection methods based on deep learning find it difficult to achieve good results because of the complex background environment, small initial targets of the flame, irregular flame shape and distribution, and complex requirements for the deployment of a detection model. To better prevent the occurrence and spread of forest fires, protect the forest environment, and ensure the safety of people and property, a lightweight real-time forest fire detection method called RF-YOLOv8s is proposed. Based on the YOLOv8 model, this method first introduces a convolution module RFCBAMConv based on receptive field attention mechanism, which emphasizes the spatial characteristics of the receptive field to strengthen the learning ability of flame characteristics. Second, in the feature fusion part, a lightweight Cross-Scale Feature Fusion Module (CCFM) is used to reduce model parameters and computation while enhancing the model's adaptability to scale changes and detection ability for small-scale objects. Simultaneously, a Dynamic detection Head (DyHead) is used to unify and strengthen multi-scale information with the help of an attention mechanism, improving the model's detection effect. In addition, Inner-CIoU is used as a new bounding box loss function to overcome the generalization deficit of bounding box regression and accelerate the model's convergence speed by controlling the auxiliary bounding box. The experimental results show that on forest fire image data, the average accuracy of RF-YOLOv8s forest fire detection algorithm is 90.2% and the number of model parameters is only 8.88×106; these are 2.5 percentage points higher than the average accuracy and 20.2% smaller than the number of model parameters of the YOLOv8 algorithm. While having stronger detection ability, it meets the requirements of lightweight models in the field of forest fire detection, ensuring the algorithm's practicability in forest fire detection scenarios.
-
Computer Engineering. 2025, 51(9): 362-372. https://doi.org/10.19678/j.issn.1000-3428.0069376
With the rapid development of port logistics and the increasing prevalence of automated container terminals, the issue of relocating export containers at specific locations in yard bays has garnered significant attention. This study addresses the container bay relocation problem by designing heuristic relocation rules under various container distributions and meticulously optimizes the handling of empty and special stacks. Moreover, it proposes a policy-based fast solution algorithm. Building on this algorithm, this study introduces a rule scoring system to develop a rule-based ordered branch and bound algorithm and a directed search algorithm. The branch and bound algorithm can obtain the optimal solution to a problem, whereas the directed search algorithm can achieve a superior feasible solution in a shorter time. Case study results demonstrate that both the rule-based ordered branch and bound algorithm and the directed search algorithm efficiently solve small-scale cases and achieve 47.78% and 56.59% improvements in solving efficiency for large-scale cases, respectively, compared with existing algorithms.
-
Computer Engineering. 2025, 51(9): 373-378. https://doi.org/10.19678/j.issn.1000-3428.0069528
An improved particle swarm-based multi-objective optimization scheduling algorithm for elevator group control is proposed to address unsatisfactory user experience and system energy consumption in Elevator Group Control Systems(EGCS). First, considering the complexity of the system control objectives, a multi-objective optimization model is established using indicators such as passenger waiting time, riding time, long waiting time, and system energy consumption. Using the linear weighted summation method to design the system's comprehensive evaluation function, changing the weights can adapt to different traffic patterns. Second, the Grey Wolf Optimization (GWO) algorithm is introduced to address the issue of the Particle Swarm Optimization (PSO) algorithm being prone to falling into local optimal solutions. The grey wolf-particle swarm hybrid optimization algorithm is applied to a multi-objective scheduling system. Simulation results show that this hybrid algorithm can effectively reduce the average riding and waiting times for users and the number of elevator starts and stops, thereby enhancing the elevator group control system's overall performance.
-
Computer Engineering. 2025, 51(9): 379-386. https://doi.org/10.19678/j.issn.1000-3428.0069412
This paper proposes an off-grid near-field channel estimation algorithm based on the simplex method to address the estimation error problem of sampling points on the grid caused by grid channel estimation using polar-domain sparsity in eXtremely Large-scale Multiple-Input-Multiple-Output (XL-MIMO) systems. First, a polar-domain transformation matrix is used to replace the traditional Discrete Fourier Transform (DFT) matrix for the sparse representation of near-field channels. The Polar-domain Simultaneous Orthogonal Matching Pursuit (P-SOMP) algorithm is then employed for an efficient initial estimation. However, because of the large dimensions and poor column orthogonality of the polar-domain transformation matrix generated jointly from the angle and distance, energy leakage occurs in the polar far-field channels. Therefore, the simplex method is further utilized to achieve precise off-grid estimation based on the maximum likelihood principle, thereby improving the estimation accuracy. Simulation results demonstrate that the off-grid near-field channel estimation algorithm based on the simplex method enhances the estimation accuracy compared to the traditional P-SOMP algorithm.
Volume 51 No 9
Superintended by:
China Electronics Technology Group Corporation
Sponsored by：
East-China Institute of Computer Technology
Shanghai Computer Society
Edited & Published by：
Editorial Office of Computer Engineering
Chief Editor：Jiang Bo
Executive Chief Editor:
Hu Bing
Tel：021-67092217
E-mail: ecice06@ecict.com.cn
CN: 31-1289/TP
ISSN: 1000-3428