15 August 2026, Volume 52 Issue 8
-
Select all|
-
Computer Engineering. 2026, 52(8): 1-15. https://doi.org/10.19678/j.issn.1000-3428.0260211
As the scale and complexity of complex intelligent systems represented by high-performance computing and embedded systems continue to grow, logs, as core operational data, have become critical for ensuring reliable system operation through automated anomaly detection. Traditional log anomaly detection methods driven by machine learning and deep learning primarily focus mostly on log sequence modeling but suffer from insufficient semantic understanding and limited generalization ability. Large Language Models (LLMs), with their superior semantic understanding and contextual reasoning capabilities, have effectively overcome this limitation. Since the advent of LLM technology, relevant research has emerged rapidly, achievements are scattered across multiple technical paths and lack a systematic review. This paper provides a comprehensive survey of log anomaly detection methods based on LLMs. We select 35 core papers and establish a unified technical classification framework, categorizing existing methods into five technical routes: prompt engineering, Retrieval-Augmented Generation (RAG), domain fine-tuning, Reinforcement Learning (RL), and large—small model collaboration. Our analysis reveals that Supervised Fine-Tuning (SFT) is currently the most widely adopted technical route, while the large—small model collaborative architecture, as an emerging paradigm, is shifting research focus from merely pursuing detection accuracy to balancing inference efficiency and industrial deployability. The current evaluation system remains heavily focused on detection performance metrics, with insufficient attention paid to efficiency overhead and interpretability. Finally, this paper identifies inference latency bottlenecks and data privacy challenges of LLMs when processing ultra-long and massive log streams, and provides insights into frontier directions such as lightweight deployment and online continual learning.
-
Computer Engineering. 2026, 52(8): 16-32. https://doi.org/10.19678/j.issn.1000-3428.0260397
Traffic flow prediction is a key core technology in Intelligent Transportation Systems (ITS) and has significant value in improving the efficiency of urban traffic management. Time series prediction and machine learning models are widely used in traditional methods; however, spatio-temporal Graph Neural Networks (GNN) have become a research hotspot owing to their outstanding advantages in traffic flow representation learning. However, existing methods have significant limitations. First, they rely on static graph structures, making it difficult to model long-range spatial dependencies and regional differences. Second, they fail to capture the time delay effect between road segments, resulting in insufficient modeling of spatio-temporal dependencies. To address these issues, this study proposes a Space Dynamic Graph Convolutional (SDGC) traffic flow prediction model based on the Time-Lag Aware Attention (TLAA) mechanism. The encoder—decoder architecture is adopted, and the mechanism is combined with spatial dynamic graph convolution to mitigate the adverse effects of the time-delay phenomenon on prediction accuracy. Specifically, gated causal convolution incorporates attention mechanisms to explicitly model the asynchronous temporal dependencies of spatial nodes by learning delay parameters. The dynamic adaptive spatial correlation matrix can perceive real-time changes in traffic status and accurately depict the dynamic propagation process of traffic flow in the road network. Meanwhile, the introduction of spatio-temporal embedding enables the model to accurately depict the short-term fluctuations, long-term periodic patterns, and spatio-temporal heterogeneity of traffic flow under network topology constraints. Experiments on public benchmark datasets demonstrate that this method effectively addresses challenges related to time lag and dynamic behavior. Compared with existing state-of-the-art baseline models, it achieves significant reductions in key prediction error metrics Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), ranging from 6.5% to 11.9%.
-
Computer Engineering. 2026, 52(8): 33-45. https://doi.org/10.19678/j.issn.1000-3428.0253402
Unstructured road segmentation is a crucial component of environmental perception for autonomous driving systems, and it suffers from challenges such as the integrity of global topological modeling, preservation of boundary details, and trade-off between model efficiency and accuracy. To address these challenges, this paper proposes a lightweight axial context network AXON-Net. The network employs an encoder—decoder architecture and introduces a Channel And Spatial Attention Block (CASAB) in the encoder, which adaptively recalibrates feature weights by aggregating multidimensional statistical information to effectively suppress environmental noise, thereby enhancing feature discriminability in complex backgrounds. A Lightweight Partial Context (LightPCT) model is integrated at the bottleneck, utilizing a partial channel interaction strategy to reduce computational redundancy and efficiently capture long-range dependencies for restoring road topological connectivity. In addition, the decoder integrates Dual-Path Channel Fusion (DPCF) and Thin Structure Enhancer (TSE) modules for bridging the feature semantic gap and explicitly enhancing axial geometric features for the refined recovery of blurred road edges. Experimental results on unstructured road datasets constructed from the India Driving Dataset (IDD) and Off-Road Freespace Detection (ORFD) dataset show that AXON-Net achieves road Intersection over Union (IoU_Road) scores of 95.3% and 88.1%, respectively. With only 8.49×103 parameters, it achieves a superior balance between segmentation accuracy and model efficiency. Ablation studies further validate the synergistic effectiveness of the proposed modules, demonstrating the potential application of the network for unstructured road perception tasks.
-
Computer Engineering. 2026, 52(8): 46-57. https://doi.org/10.19678/j.issn.1000-3428.0253462
Video analytics extracts high-value information from video streams and plays a crucial role in applications such as intelligent transportation and public safety. Although traditional cloud-based video analytics offers powerful computational capabilities, uploading massive amounts of video data results in high bandwidth consumption and network latency. Edge computing reduces network latency by processing video data near cameras; however, it still faces two major challenges. First, frame-by-frame analysis leads to redundant inference, and existing frame reuse methods cannot fully exploit local similarities in historical frames. Second, uneven core workload arises because task allocation across big LITTLE lacks real-time load awareness. To address these issues, this study proposes Vable, an efficient video analytics system for big.LITTLE edge devices. It employs a multihistorical frame, block-level frame reuse mechanism, which partitions video frames into fine-grained blocks and employs a tree-based storage structure combined with Locality-Sensitive Hashing (LSH) for similarity matching, enabling efficient cross-frame computation reuse and significantly reduces redundant inference overhead. Additionally, Vable introduces a core workload-aware list-based Directed Acyclic Graph (DAG) partitioning algorithm that dynamically allocates analysis tasks by monitoring the real-time load of big LITTLE, thus balancing computation and communication overhead while avoiding latency increases caused by load imbalance. A prototype of Vable is implemented and evaluated using two real-world datasets. Experimental results demonstrate that Vable reduces the end-to-end latency by 59.23% in UA_V1 compared to WiseFuse (the highest-latency baseline) and by 45.83% in UA_V2 compared to Faastlane, effectively lowering the end-to-end latency while maintaining high throughput.
-
Computer Engineering. 2026, 52(8): 58-70. https://doi.org/10.19678/j.issn.1000-3428.0253484
Image deraining aims to restore high-quality clean images from rain-degraded inputs and is a key technique for improving the robustness of outdoor vision systems. Existing Convolutional Neural Network (CNN)-based methods are limited by local receptive fields and thus struggle to effectively model long-range dependencies. Although some Transformer-based methods enhance global modeling capability, they show limited ability in handling multi-scale and directional rain streaks as well as recovering local high-frequency details. Moreover, existing frequency-domain enhancement methods often treat the frequency domain merely as an auxiliary representation, which lack fine-grained collaborative modeling of the complementary relationship between the spatial and frequency domains. To address these issues, this study proposes a Synergizing Frequency and Spatial Network (SynFSNet) for image deraining. Unlike existing methods that primarily use the frequency domain as a global enhancement tool, the proposed method uses the directional, density-related, and scale-varying characteristics of rain streaks in the frequency domain. Specifically, a Multi-Scale Fourier Fusion Module (MSFFM) is designed to progressively suppress rain streaks in a coarse-to-fine manner. Additionally, a Fourier Shaped Attention Module (FSAM) is introduced in which Rectangular Filter Attention (RFA) and Square Filter Attention (SFA) are used to enhance direction-sensitive and local-structure-sensitive frequency-domain representations, respectively. Furthermore, a dual-domain loss is employed to jointly constrain spatial structural restoration and frequency-domain consistency, thereby improving restoration quality in complex scenes. Experimental results show that SynFSNet achieves superior performance on multiple synthetic and real-world datasets, verifying the effectiveness of collaborative spatial—frequency domain modeling for complex image deraining tasks.
-
Computer Engineering. 2026, 52(8): 71-83. https://doi.org/10.19678/j.issn.1000-3428.0260594
The video stream of maritime multifunctional navigation mark monitoring presents the distinctive scene characteristics of a dominant static sea surface background and sparse dynamic ship targets. Constrained by the low-bandwidth transmission of navigation mark-borne communication networks, compressed video streams experience severe quality degradation, including noise superposition, blocking artifacts, ringing artifacts, frame blurring, and inter-frame motion estimation bias. Existing general-purpose video quality enhancement methods are unable to jointly balance single-frame spatial detail recovery and multi-frame temporal information utilization, and therefore fall short of the requirements of the aforementioned scenario. To address this issue, a Content Adaptive Fusion-based Video Stream Quality Enhancement of Maritime Multifunctional Navigation Mark Monitoring (CAF-VSMM) method is proposed, in which a dual-branch fusion enhancement model is constructed. The single-frame enhancement branch, built on multiscale features and a spatio-temporal network, exploits the latent temporal information of a single-frame via virtual frames and integrates the pixel shuffle with a residual network to achieve multiscale spatial feature fusion, effectively suppressing compression artifacts and spatial degradation induced by the maritime environment. The multi-frame enhancement branch, based on a dynamic shifted window and inter-frame motion compensation, efficiently models long-range spatio-temporal dependencies and employs hierarchical offset estimation together with deformable convolution to achieve accurate inter-frame alignment in multiship scenes, thereby alleviating temporal blurring and motion misalignment. The Dynamic Feature Perception-based Content-Adaptive Fusion (DFP-CAF) module, based on dynamic feature perception, dynamically adjusts the feature weights of the two branches according to the video content, realizing an adaptive enhancement strategy that emphasizes detailed recovery for static backgrounds and temporal consistency for dynamic targets. Experimental results show that, on the self-built Port Cluster Waterway Monitoring Video Dataset (PC-WMVD), the Peak Signal to Noise Ratio (PSNR), Structural Similarity (SSIM), and Video Multimethod Evaluation Fusion (VMAF) of the proposed method reach 42.71 dB, 99.33%, and 77.14%, outperforming the second-best method by 3.90 dB, 1.19 percentage points, and 6.50 percentage points, respectively; on the public dataset JCT-VC, they reach 28.66 dB, 88.60%, and 55.40%, exceeding the second-best method by 1.63 dB, 5.42 percentage points, and 3.19 percentage points, respectively. The model has 37.52×106 parameters and 51.74×109 Floating-Point Operations Per Second (FLOPs) and runs at 41.52 frames per second, meeting the real-time requirements of waterway monitoring. The proposed method is tested in an intelligent monitoring system for the waterways of port clusters, where it significantly improves the identifiability of key targets, such as ships and navigation marks, and effectively supports downstream tasks, such as ship tracking, verifying its engineering practicability and application value.
-
Computer Engineering. 2026, 52(8): 84-100. https://doi.org/10.19678/j.issn.1000-3428.0252971
Knowledge Graphs (KGs) are structured semantic knowledge representations with entities as nodes and relations as edges. They accurately depict various things in the real world and their complex associations and have become a core supporting technology across multiple domains, including artificial intelligence, natural language processing, recommendation systems, and intelligent question-answering. Hence, they form an important foundation for machines to understand semantics and achieve cognitive intelligence. This paper describes the basic concepts and system architecture of a KG, clarifies the knowledge representation unit with the ″entity—relation—attribute″ triple as the core, and analyzes the applicable scenarios and technical characteristics of both top-down and bottom-up construction approaches. The technical evolution of three core links in the KG construction process, namely, information extraction, knowledge fusion, and knowledge reasoning, are analyzed. Subsequently, the technical development context is systematically discussed, and the advantages and limitations of different methods are discussed. Further, through an in-depth analysis of the differences in technical route selection between DBpedia and Baidu as two typical KGs, the theoretical method is combined with an actual KG construction scenario. Finally, the challenges faced by current KG construction in terms of data quality, semantic consistency, and dynamic evolution are summarized, and future research directions are examined. This review provides comprehensive guidance for both theoretical research and practical applications in KG construction, thereby advancing technological development in this field.
-
Computer Engineering. 2026, 52(8): 101-113. https://doi.org/10.19678/j.issn.1000-3428.0252689
In recent years, deep learning has achieved tremendous success in application fields such as computer vision and natural language processing. This has led researchers in high energy physics to also turn their attention to deep learning technologies and explore their application in hadronic Jet Tagging tasks. Initially, researchers converted Jet data into image and sequence data, and used Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) to tag Jets. However, these approaches suffered from problems such as low computational efficiency and poor interpretability. To address these issues, researchers have made improvements to network architectures from multiple perspectives and conducted training on various constructed Jet Tagging datasets, thereby enhancing the classification performance of the models. This paper provides an in-depth review of the key modules of novel network models, including methods for representing Jets based on sets, the application of equivariant neural networks, and the exploration of Jet foundation models. Meanwhile, this paper analyzes and compares various tagging classifiers, evaluates the performance of different network architectures, summarizes the current status of relevant models, and discusses the application prospects of deep learning models in Jet Tagging tasks.
-
Computer Engineering. 2026, 52(8): 114-125. https://doi.org/10.19678/j.issn.1000-3428.0260116
Recently, generative models have advanced rapidly and demonstrated remarkable capabilities in image synthesis, artistic creation, and other fields. However, although these models improve generation performance, they introduce significant privacy risks: the private information contained in the training data may be leaked through the generated content. To address this issue, machine unlearning techniques have been proposed to weaken the memorization of specific data by the model, thereby preventing the leakage of private information during generation tasks. Among these, Identity (ID) unlearning methods for face generation aim to prevent a model from generating images with specific identities, thereby protecting personal privacy. However, the existing ID unlearning methods suffer from insufficient robustness. Attackers can exploit a small amount of data from the forgotten identity to fine-tune the model in a limited number of steps, thereby recovering the forgotten identity, a threat known as the relearning attack. Although traditional ID unlearning methods can effectively reduce the similarity between the target identity and generated images, the unlearning effect is easily reversed after a relearning attack, compromising their robustness. To address this problem, a relearning attack method is constructed to expose the robustness deficiency of existing ID unlearning methods. On this basis, a robust ID unlearning method is proposed that enables the model to maintain a stable and effective unlearning performance even under relearning attacks. First, a relearning attack method targeting Generative Adversarial Networks (GAN) is proposed. Building on an already unlearned model, this method performs a limited number of gradient updates using a small amount of data from the forgotten identity, driving the generated images back toward the forgotten identity representation, thereby evaluating the robustness of the ID unlearning methods. Second, a Robust ID Unlearning (RIDU) method based on smoothness optimization is proposed. During the unlearning training process, this method applies random perturbations to the model's weight parameters; therefore, the optimization not only pursues the unlearning effect at the current parameter point but also considers the overall stability within the parameter neighborhood, thus forming a smooth and stable unlearning region. Thus, the unlearning effect no longer relies on a single fragile local optimum, making it difficult for relearning attacks to recover forgotten identities. Moreover, the method employs appropriate loss functions to ensure a balance between the unlearning objective and the generation quality of the model. Experimental results on the CelebAHQ dataset demonstrate that under nonattack conditions, RIDU significantly reduces the similarity between the target identity and generated images, outperforming traditional methods. Under relearning attacks, although the unlearning effect of traditional methods is easily reversed, RIDU maintains strong unlearning capability and effectively suppresses the recovery of the target identity. This method enables the model to forget specific identities while preserving its generation quality with minimal degradation.
-
Computer Engineering. 2026, 52(8): 126-137. https://doi.org/10.19678/j.issn.1000-3428.0070466
The number of malicious websites accessed by in-vehicle third-party services has been rapidly increasing, posing a significant threat to the security of the Internet of Vehicles (IoV). Currently, three major challenges exist for IoV malicious website detection: traditional tools exhibit high detection latency when processing large-scale website data, the presence of obfuscated malicious Uniform Resource Locators (URLs) reduces detection accuracy, and the difficulty in obtaining malicious website datasets further hinders effective detection. Collectively, these factors limit the efficiency and accuracy of detection. To address these issues, this study proposes a fast multi-stage malicious website detection method based on logistic regression. This method uses search engines for the preliminary filtering of legitimate websites to reduce computational resource wastage. It designs matching rules through the analysis and summarization of malicious obfuscation techniques and introduces a heuristic rule-based malicious website filtering method to effectively filter obfuscated website URLs, overcoming the limitations of traditional tools in detecting URLs with malicious obfuscation. To further enhance the detection accuracy, it constructs a comprehensive and lightweight set of malicious website detection features and employs logistic regression classification for feature extraction and analysis. Experimental results demonstrate that the proposed method significantly outperforms traditional methods in terms of accuracy and efficiency in malicious website detection, achieving an accuracy of 98.1% on public datasets and reducing the detection time by approximately 75%.
-
Computer Engineering. 2026, 52(8): 138-148. https://doi.org/10.19678/j.issn.1000-3428.0252067
With the continuous evolution of network environments and the constant emergence of Internet applications, the applicability of machine learning classifiers trained on previous traffic data to new sample spaces gradually weakens. This leads to a decline in the recognition capability of classification models; thus, these models are ineffective in meeting the ever-increasing demands of network services and security. If the classifier is manually updated based on experience, it consumes a significant amount of effort, and it is difficult to ensure the generalization performance of the new classifier. Moreover, the continuous influx of new data poses severe challenges in balancing model training accuracy with the storage of computational resources. To address this issue, an incremental learning strategy employing spatial optimization techniques is proposed to achieve efficient network traffic classification. First, by optimizing the spatial distribution of old and new traffic samples, clusters of different categories (both old and new) are maintained with minimum intervals to avoid distribution conflicts arising from old and new tasks shared feature space. Within the optimized feature space, a small number of old data samples are replayed, and knowledge distillation techniques were utilized to maintain the stability of the original model parameters. Only the extended parts of the model are adjusted to update the classifier at minimal cost. Experimental results on the USTC-TFC2016 dataset demonstrate that, compared with other methods, this approach exhibits higher stability and effectiveness in terms of model accuracy, resource consumption, and performance.
-
Computer Engineering. 2026, 52(8): 149-162. https://doi.org/10.19678/j.issn.1000-3428.0252018
Federated Learning (FL), a distributed Machine Learning (ML) framework, can achieve collaborative model training while protecting data privacy. However, challenges remain in ensuring privacy protection, establishment of mutual trust among participants, and defending against malicious attacks. In Hierarchical Federated Learning (HFL), the potential untrustworthiness of central servers, intermediate layers, and edge devices poses risks of privacy leakage and malicious manipulation. Additionally, malicious users may upload abnormal parameters, disrupting the training process and affecting model performance. Therefore, efficient security verification and malicious detection are critical issues in HFL. To address the challenges of mutual distrust among participants and Byzantine attacks in HFL, this paper proposes a secure aggregation scheme with non-interactive verification under a hierarchical architecture. First, a mutual verification mechanism for HFL is designed based on a commitment scheme, allowing participants to perform mutual verification. Second, a constraint and detection scheme for malicious updates is constructed using non-interactive zero-knowledge range proofs, enabling the server to detect and filter out malicious users. Third, a noise masking scheme is designed based on the Chinese Remainder Theorem (CRT), supporting user exit and reconnection while ensuring users' local privacy. Finally, security analysis and experimental evaluation demonstrate that the proposed scheme achieves secure mutual verification and malicious detection with high efficiency.
-
Computer Engineering. 2026, 52(8): 163-175. https://doi.org/10.19678/j.issn.1000-3428.0252005
As an efficient information storage and transmission method, QR codes are widely used in payment, advertising, and logistics. However, the robustness and steganography of existing QR code steganography techniques in noisy and disturbed environments are insufficient to meet the needs of high-security information transmission. Therefore, a robust QR code steganography algorithm based on path planning and pixel flipping is proposed. The algorithm ensures that the anti-jamming ability of the secret information is enhanced without affecting the normal recognition of the QR code by treating the QR code as a labyrinth and selecting pixels to be flipped in combination with the path planning algorithm. In terms of technical implementation, first, a path planning algorithm is designed for selecting the optimal pixel points to reduce the impact of information embedding on the image quality of the QR code; second, the embedding of the secret information is realized by applying the pixel flipping technique and analyzing its performance under different noise conditions. The algorithm is tested via experiments using typical interference scenarios such as noise attack, image perturbation, and physical distortion; the embedding capacity, image quality, and information recovery rate are used as evaluation indexes. The results show that the algorithm has significant advantages in improving the robustness and steganography of QR code steganography and is suitable for scenarios with high information security requirements. Moreover, the proposed algorithm opens avenues for further development of QR code steganography techniques.
-
Computer Engineering. 2026, 52(8): 176-183. https://doi.org/10.19678/j.issn.1000-3428.0070593
Chiplets have emerged as a new trend in chip design. With the evolution of microprocessor architectures toward multi-core and many-core implementations, the complexity of Cache coherence protocols has increased substantially. For microprocessors integrating multiple chiplets, the verification of Cache coherence protocols is a technically critical issue with significant practical value. The directory-based Cache coherence protocol is the most widely adopted hardware-implementation approach. Focusing on a self-developed distributed Directory Controller Unit (DCU), this study investigates agile verification technologies for DCU in multi-chiplet integrated processors. An optimized mechanism based on a Negative Selection Algorithm (NSA) is employed in random testing, where functional coverage was improved by 28 percentage points under a 40% reduction in random test stimuli. By combining random and directed test stimuli, the functional coverage ultimately reaches 100%. Additionally, a coherence checker is designed to monitor the entire lifecycle of the coherence protocol transactions. This enables the rapid localization of design errors and accurate tracing of problematic scenarios, with 90% of the design bugs successfully identified. This approach effectively enhances verification efficiency and quality.
-
Computer Engineering. 2026, 52(8): 184-192. https://doi.org/10.19678/j.issn.1000-3428.0070725
As software scale and complexity grow exponentially, monitoring and analyzing program runtime behavior has become increasingly challenging. Dynamic Binary Instrumentation (DBI) is an effective solution to this problem, with mature tools such as Pin and Valgrind supporting mainstream architectures such as x86 and ARM. However, these tools lack support for emerging domestic instruction set architectures such as LoongArch. LoongArch, a self-developed instruction-set architecture in China, exhibits high levels of autonomy, advancement, and compatibility. However, owing to its relatively short developmental history, the ecosystem remains incomplete, particularly in the debugging toolchain. Developing a dynamic binary instrumentation tool is important to address this gap and promote the maturation of the LoongArch ecosystem. This study aims to design and implement a dynamic binary instrumentation tool based on the Quick Emulator (QEMU) framework to support program monitoring and analysis in LoongArch. The tool, modeled after Pin, implements five fundamental instrumentation granularities and related Application Programming Interfaces (APIs), along with over 20 instrumentation tools for direct use or as learning resources for tool development. To enhance performance, the framework is optimized through improvements in conditional jump instruction translation, basic block linking, and instrumentation inlining. Performance tests demonstrate that the optimized framework achieves improvement of over 100 times in instruction-level instrumentation efficiency and nearly 33 times in basic block-level instrumentation efficiency.
-
Computer Engineering. 2026, 52(8): 193-202. https://doi.org/10.19678/j.issn.1000-3428.0252315
Dataflow programming is an emerging parallel-programming paradigm. By splitting tasks into fine granularities, high-performance parallel computing can be achieved at the expense of increased programming complexity. This study proposes a new program transpilation method and compilation procedure that leverages two levels of intermediate representation and converts control flow computational programs into a dataflow architecture in a fully automatic manner. Owing to the substantial disparities between these programming models, many challenges exist in automatic conversion schemes, which are split into four categories and addressed separately. Experimental results on the PolyBench/C dataset indicate that the proposed conversion scheme can produce dataflow parallel programs on par with those produced by the Stateful DataFlow multiGraph (SDFG)-based OpenMP parallelizing compiler while surpassing it in approximately half of the cases. This study also conducts case studies on multiple program samples and discusses the causes of performance distinctions in different compilation schemes, revealing the superiority of the Codelet model with a massive number of threads on irregular parallelism.
-
Computer Engineering. 2026, 52(8): 203-209. https://doi.org/10.19678/j.issn.1000-3428.0070145
Matrix transposition is a fundamental and crucial operation in digital signal processing and has extensive applications in various fields, such as array signal processing, spectral analysis, signal filtering, and signal recognition. The traditional matrix transposition method, owing to its complex algorithm, long processing time, and low efficiency, can no longer satisfy the high-performance requirements of modern signal processing systems. To solve these problems, this paper proposes a block-based matrix transposition method based on a Field Programmable Gate Array (FPGA) and Double Data Rate 3 Synchronous Dynamic Random Access Memory (DDR3 SDRAM). The core of this method is the optimization of data mapping and access patterns. Distributing the efficiency bottlenecks in the transposition process to both the read and write stages effectively improves the transposition efficiency. Specifically, this method utilizes the high-speed parallel processing capability of the FPGA combined with the high-speed storage characteristics of the DDR3 SDRAM to achieve the rapid transposition of large matrices. Two DDR3 SDRAM chips are employed to construct a ping-pong buffering architecture. One is used for reading the original matrix data, and the other is used for writing the transposed data, thereby achieving a real-time pipelined transposition of the matrix. The test results on an actual hardware platform indicate that this method has a significant effect on handling the transposition of large M×N matrices with dimensions that are not powers of 2. Compared with distributed storage matrix transposition algorithms, this method is not limited by the size of the matrix and can be applied to the transposition of any large matrix. It is also easy to implement and has a transposition efficiency of 90%.
-
Computer Engineering. 2026, 52(8): 210-224. https://doi.org/10.19678/j.issn.1000-3428.0070632
Domestic deep-learning accelerators have developed rapidly in recent years.Hardware resources have been changing continuously, and a series of tensor core instructions have been introduced. Therefore, manually adapting and optimizing the convolution operator in the accelerator is a considerable challenge for developers. To this end, this study proposes a convolution code generator for a domestic accelerator to simplify the adaptation and optimization process of the convolution operator. The generator provides configuration parameters as an external interface, and users only need to configure these parameters to generate specific convolution operators. The generator itself consists of a three-layer architecture: the instruction layer encapsulates the underlying instructions and categorizes them according to the hardware architecture; the component layer organizes the corresponding instructions according to the preset hardware architecture information and provides highly abstract and reusable functional components from the perspective of thread blocks and thread bundles; and the operator construction layer splices the functional components according to an implicit convolution algorithm and generates a convolution operator. To ensure the computing performance of the convolution operator, the generator is optimized from two aspects: using the vectorization and thread partitioning algorithms to optimize the global memory access performance and using the transposition algorithm to transform the thread structure of the multiply-accumulate instruction to optimize the write-back performance. The test results show that the optimization algorithm of the generator can significantly improve the operator performance; under two hardware versions, the convolution operator performance of the NHWC storage layout reaches 95% and 90% of the official operator performance. The generator provides a new solution for adaptation and optimization of the convolution operators of domestic accelerators.
-
Computer Engineering. 2026, 52(8): 225-237. https://doi.org/10.19678/j.issn.1000-3428.0070570
General-Purpose Graphics Processing Units (GPGPUs) are widely utilized for various computational tasks owing to their robust parallel processing capabilities. However, GPGPUs employing the Single Instruction Multiple Threads (SIMT) execution model often encounter non-consistent control flows during kernel execution, leading to warp divergence and a subsequent decline in the overall accelerator performance. To address the performance degradation caused by non-consistent control flows in kernel execution, this study introduces MergeCFG, a branch compilation optimization technique tailored for specific scenarios. During the intermediate code optimization phase in the compiler, MergeCFG conducts a control flow analysis to identify consecutive branch structures in the control flow graph that share identical conditional branches, thereby identifying potential optimization opportunities. Subsequently, based on instruction analysis, it assesses the feasibility of optimization to determine whether opportunities for reducing branch operations exist. Finally, by employing basic block duplication and merging techniques, it optimizes the control flow structure to minimize branch operations, thereby simplifying the control flow and enhancing the program execution efficiency. Experiments are conducted using a domestic GPGPU system with seven suitable benchmark test suites to validate the feasibility of this method. The results demonstrate that this method effectively reduces branch operations within programs, leading to significant performance improvements in optimized test cases. The average speedup across the evaluated cases ranges from 2% to 12%, with certain test cases exhibiting performance enhancements exceeding fivefold.
-
Computer Engineering. 2026, 52(8): 238-246. https://doi.org/10.19678/j.issn.1000-3428.0070349
In recent years, with the rapid development of Artificial Intelligence (AI), neural networks have been widely used in various artificial intelligence fields. However, the demand for computing power is increasing with the complexity of logic and algorithms. This issue is particularly prominent in resource-constrained embedded heterogeneous platforms. Therefore, it has become a research hotspot. This paper analyzes the development of parallel computing engines and software-defined architecture globally, explores the characteristics of software-defined architecture, and proposes a design scheme for a Parallel Computing Architecture based on Software Definition (SDPCA). This computing architecture adopts a modular structure of ″data-control-capability″ and comprehensively utilizes key technologies of parallel computing architecture such as software-defined resources, dynamic scheduling strategies, virtualized memory, and operator optimization. These technologies effectively improve the runtime computational efficiency of SDPCA, providing a lightweight framework with dynamic reconstruction features. Under the typical hardware conditions of an embedded heterogeneous platform, SDPCA is tested using neural network convolutional layer computation and model inference. The experimental results show that the SDPCA function is stable and runs accurately, and that the computational performance of the convolutional layer is more than four times higher than that of OpenCL. The image inference tests of the four common neural network models reveal that the recognition efficiency of SDPCA increases by more than 2.6 times compared to OpenCL and it is more stable in different model calculations. Additionally, SDPCA achieves a recognition accuracy of over 91.7% for all four models, with an average inference accuracy improvement of 3.9 percentage points compared with OpenCL. The functional and performance indicators meet the design expectations and have certain reference values.
-
Computer Engineering. 2026, 52(8): 247-259. https://doi.org/10.19678/j.issn.1000-3428.0070473
In the era of big data, information sources are diverse and complex, making it difficult to ensure the integrity of the information systems formed from the collected data. In increasingly Incomplete Information Systems (IIS), using rough set theory based on asymmetric tolerance relations for knowledge distillation and improving the calculation speed of approximation sets are critical issues that must be addressed before their application. An improved scheme is proposed for the redundant tolerance class problem of asymmetric tolerance relations. A Boolean matrix representation method for the upper and lower approximation sets in incomplete information systems is designed, and a matrix-partitioning algorithm for calculating rough approximation sets of asymmetric tolerance relations is proposed. Furthermore, the acceleration of the approximation set calculation process is implemented on a Graphics Processing Unit (GPU) to improve the calculation efficiency of the approximation sets. Furthermore, given the limited storage space of GPUs, a hierarchical structure and its algorithm for objects in incomplete information systems are constructed, converting the global relationship matrix into multiple local relationship matrices to alleviate the storage pressure on GPUs during the calculation of the nearest tolerance classes. Experimental results on the UCI and generated datasets show that the number of tolerance classes based on the nearest tolerance relations is significantly reduced compared to the baseline case. Additionally, the matrix-partitioning algorithm effectively accelerates the approximate set calculation process for GPUs. Compared with CPU serial computation and distributed parallel computation, the GPU partitioning algorithm executes at an average speed of 16.69 times and 3.89 times faster.
-
Computer Engineering. 2026, 52(8): 260-271. https://doi.org/10.19678/j.issn.1000-3428.0070553
To address the limitations and deficiencies of traditional ″human-centered″ Mobile Crowd Sensing(MCS) in terms of timeliness and global sensing coverage, this study proposes a human-Unmanned Aerial Vehicle (UAV) hybrid air—ground collaborative MCS and computing framework. In this framework, low-altitude UAVs and ground participants fully leverage their respective sensing and computing resources through collaborative interactions to achieve localized data sensing and computation. Considering the stringent requirements for data timeliness in delay-sensitive application scenarios, the Age of Information (AoI) metric is introduced into this framework. By jointly optimizing UAV flight trajectories, selection variables for offloading computations from mobile devices to UAVs, and data offloading ratios, the framework aims to minimize the AoI of all sensed data. To achieve this objective, a multi-agent Deep Reinforcement Learning (DRL) algorithm CAMATD3 is designed, which combines the dual Q-learning mechanism of multi-agent twin delayed deep deterministic policy gradient MATD3 with the training mechanism of Deep Q-Networks (DQN). Additionally, a phased decision-making mechanism and comprehensive experience replay strategy are proposed to address the challenges posed by the hybrid action space during the optimization process. The simulation results demonstrate that the proposed CAMATD3 algorithm significantly outperforms existing methods in terms of agent training effectiveness and AoI minimization, thereby substantially improving the timeliness of data sensing.
-
Computer Engineering. 2026, 52(8): 272-282. https://doi.org/10.19678/j.issn.1000-3428.0070560
The 5G network is a new generation of mobile communication technology characterized by high speed, low latency, large bandwidth, and extensive connectivity, providing an infrastructure for numerous innovative and intelligent applications. However, its high frequency results in limited coverage. Existing base site selection methods fall into local optima, and it is difficult to achieve high coverage at low cost. In this study, an approach for determining 5G base station locations is proposed based on network coverage and user demand, utilizing historical user terminal communication data. First, based on historical user terminal communication data, the concept of mutual exclusion probability between base stations is proposed by considering both the overlap of base station coverage and level of user demand satisfaction. Thus, the likelihood that specific base stations cannot coexist in a given area while meeting user demands is described. Accordingly, a method is presented for calculating mutual exclusion probabilities based on a Graph Convolutional Network (GCN) for nonadjacent stations. Furthermore, the concept of base station separation degree is proposed to describe the likelihood of effectively selecting multiple base stations. For the optimization problem of the 5G base station location, the target function is constructed and proven based on the separation degree while satisfying the submodularity property, and a greedy algorithm is provided for the optimal base station location. Experimental results on various scaled datasets show that the proposed method for 5G base station location outperforms other methods in terms of coverage and average user demand satisfaction by 1%—20% and 1%—6%, respectively.
-
Computer Engineering. 2026, 52(8): 283-292. https://doi.org/10.19678/j.issn.1000-3428.0070270
To support real-time data transmission and maintain stable connectivity for diverse mobile users, an air-ground cooperative Ultra-Dense Network (UDN) is adopted to enhance network service capabilities. In this novel network architecture, airborne base stations can better support user mobility, while terrestrial UDNs enable more efficient spectrum reuse. However, differentiated channel conditions and coverage disparities introduce new challenges in interference and spectrum allocation. Therefore, this paper focuses on a novel air-ground cooperative UDN and addresses the challenges of interference coordination and spectrum efficiency balancing in mobile scenarios within this network model. A Graph Neural Network (GNN)-based Deep Reinforcement Learning (DRL) algorithm, Graph-DRL, is proposed to dynamically schedule spectrum resources and improve the real-time performance of the algorithm. First, a GNN is utilized to model the user movement correlations among neighboring base station cells. By sharing the demand information of the local cell, cooperation is enhanced to determine the bandwidth reuse factor. Second, a value-based learning network is designed, with the average throughput of the network as the optimization objective. Considering factors such as user mobility, user density, and spectrum reuse degree, a trained Q value network is employed for resource allocation. Simulation results demonstrate that, compared to benchmark algorithms, the proposed algorithm significantly improves the transmission rates for highly mobile users and cell-edge users.
-
Computer Engineering. 2026, 52(8): 293-301. https://doi.org/10.19678/j.issn.1000-3428.0070720
Reconfigurable Intelligence Surface (RIS) is a promising technology for future wireless communication. Unmanned Aerial Vehicles (UAVs) are gaining increasing attention in the field of wireless communications because of their unique advantages. However, with the increasing complexity of communication environments, information transmission faces increasingly severe eavesdropping risks, and the integration of RIS with UAV opens new solutions and possibilities for the future of wireless communications. Motivated by this opportunity, this paper proposes a secure wireless communication system based on RIS-UAV. The system leverages the flexibility of UAVs and the dynamic adjustment capabilities of RIS to effectively counter potential eavesdropping behavior. From the perspective of physical layer security, the system considers the phase shift of RIS and flight trajectory of UAVs, aiming to maximize the average security rate of the system. Because the problem is inherently nonconvex, the solution decomposes it into two sub-problems. First, a closed-form solution of the phase shift is derived from the unique structure of the system. Subsequently, the UAV flight trajectory is converted into a convex optimization problem via first-order Taylor expansion, and the iterative convex approximation method is used for the iterative solution. In simulations, the average secrecy rate of the proposed system increases by approximately 12.5% compared with the benchmark scheme. This performance is a significant improvement compared with those of the fixed RIS and no-RIS schemes. Additionally, this system demonstrates unique advantages in terms of security and computational complexity, with a computational complexity of O(MN+KN3.5).
-
Computer Engineering. 2026, 52(8): 302-314. https://doi.org/10.19678/j.issn.1000-3428.0070400
Aiming at the problem of tracking dense target groups, especially multi-targets with micro-motion characteristics, this study proposes a dense target tracking algorithm based on label hierarchical gradual updating with micro-motion characteristics. In narrowband radar detection of dense micro-motion targets, multi-targets enter into the same resolution unit or cross-overlap, making it difficult to distinguish targets. Consequently, misdetection and omission occur in the detection and tracking processes, such as trajectory intersection and trajectory ablation, affecting detection and tracking performance. Because the micro-motion characteristics of a target have high research value in the group target detection and separation process, this study utilizes the Sinusoidal Frequency Modulated Fourier-Bessel Transform (SFMFBT) to separate the micro-motion target; simultaneously extracts the micro-motion parameters of the target, identifies the number, and extracts the target state parameters in the distance-Doppler-angle domain through signal processing; and builds the filtering updating parameter set jointly. To improve the performance of the tracking algorithm, based on the multi-target tracking filter updating mechanism of the DeepSORT framework, a labeled hierarchical gradual updating algorithm is designed for the dense target tracking trajectory intersection and other problems. A nonlinear estimation of the target state is carried out by the Unscented Kalman Filtering (UKF) to construct a multi-UKF tracker for fast and stable tracking of isotropic dense multi-targets, and the tracking parameters are extracted and numbered by signal processing. Simulation results show that the proposed algorithm can improve tracking accuracy and stability in dense target tracking tasks and has potential engineering applications.
-
Computer Engineering. 2026, 52(8): 315-326. https://doi.org/10.19678/j.issn.1000-3428.0070572
To address the challenges of high noise levels, strong volatility, and difficulties in extracting periodic information from the loads of Integrated Energy Systems (IES), a multivariate load forecasting method based on Pattern Cross-Correlation and Temporal Patch Association (PCC-TPA) mechanisms is proposed. This method analyzes the cross-lag relationships between external influencing factors and multivariate loads using a cross-correlation function to determine the most relevant time lags for data reconstruction and embedding. Building on this, a PCC mechanism is introduced to abstract data into patterns based on their variation trends, thus mitigating the effects of fluctuations and noise. This is followed by the identification and extraction of key moments and periodic information based on the cross-correlation theory. A TPA mechanism is designed to divide a sequence into multiple subsequences using mutual information methods to analyze and filter the subsequences, thereby enhancing the ability of the model to capture local continuity information in sequences. Multiple ablation and comparison experiments are conducted using the comprehensive energy system dataset from the Arizona state university, Tempe campus. The ablation experimental results show that the data reconstructed through cross-lag analysis effectively improved the prediction accuracy of the model. The PCC-TPA mechanisms enhance the ability of the model to identify key moments and capture local information, respectively. The comparison of the experimental results indicates that the proposed method outperforms five mainstream prediction models in terms of multiple evaluation metrics, demonstrating a higher prediction accuracy.
-
Computer Engineering. 2026, 52(8): 327-338. https://doi.org/10.19678/j.issn.1000-3428.0070372
To address the limitation of previous methods in recognizing various types of Chinese character errors and measuring structural similarity, this paper proposes a Chinese character Label Chinese Complete Graph (LCCG) matching method and innovatively introduces an encoding method based on the relative positions and topological structures of strokes. First, the construction process of the Chinese character LCCG is elaborated, and the similarity calculation methods for nodes and edges are defined. The applicable scenarios and specific cost settings for each editing operation are then clearly defined for various error types. Finally, the Max-Min Ant System (MMAS) algorithm is used to determine the optimal matching path, to achieve efficient graph matching. Experimental results show that this method can accurately recognize various types of Chinese character errors and outperforms existing methods in terms of accuracy, matching cost, and running time. Compared to traditional methods, this method not only improves the recognition efficiency significantly, but also enhances the accuracy of stroke matching and calculation of structural similarity through precise editing cost definitions and innovative stroke and structure similarity calculation, demonstrating high practical value.
-
Computer Engineering. 2026, 52(8): 339-350. https://doi.org/10.19678/j.issn.1000-3428.0070573The resource allocation problem in smart grids has gained attention in practical applications, with the goal of minimizing total system generation costs while meeting various constraints. To address capacity safety constraints and initial input limitations and to ensure supply-demand balance in distributed microgrid resource allocation, this study investigates a resource allocation problem with coupled equality and local inequality constraints. A distributed optimization algorithm independent of the initial system conditions is proposed to achieve optimal resource allocation across multiple microgrid nodes. The algorithm is based on penalty functions and the differential inclusion theory. It constructs suitable penalty parameters to ensure that the state variables of all agents enter and remain within the feasible region defined by the inequality constraints in finite time, thus effectively addressing capacity constraints and initial input limitations. In addition, under the condition of an undirected connected network, the algorithm introduces an innovative consensus mechanism to ensure that all agents satisfy the coupled equality constraints. This enables effective resource coordination among agent nodes, achieving a balance between the power supply from generators and demand from users. An analysis using Lyapunov stability theory shows that the algorithm guarantees the convergence of agents to the optimal solution of the resource allocation problem. Compared with existing distributed algorithms, the proposed algorithm features a simple structure, low computational cost, flexible initial point selection, and improved privacy protection. In addition, it does not require the computation of an exact penalty parameter in advance. Finally, two simulation case studies of power systems are conducted, which validate the effectiveness of the algorithm for solving distributed microgrid resource allocation problems.
-
Computer Engineering. 2026, 52(8): 351-363. https://doi.org/10.19678/j.issn.1000-3428.0070458
The arrival of passengers at airports exhibits certain aggregation and randomness, and an non-stationary arrival process can easily lead to wastage of service resources. To dynamically allocate service resources at airport security checks, accurate prediction of passenger flow at various times based on flight schedules is necessary. By the time the security equipment reads the passenger identification information, the passenger has already completed the queuing process, which implies that the actual arrival time at the security check is earlier than the recorded security check time. The current process of security checks cannot record the exact time at which each passenger arrives at the inspection area. Existing research on airport passenger flow forecasting has replaced the time at which passengers begin to undergo security checks with the time at which they arrive at the security check area, ignoring the queuing time for security checks, which leads to lower prediction accuracy. To address this issue, a method for forecasting passenger flow at airport security checks based on a non-stationary queuing model is proposed. First, a security check queuing model with a general distribution, non-stationary random arrival, and service processes is established, and an approximate algorithm is proposed to solve the system's queuing performance indicators. Second, a heuristic iterative algorithm based on a non-stationary queuing model is designed to iteratively calculate passenger arrival times. Finally, a passenger arrival distribution model is established based on a time-segmented clustering and fitting method. The analysis results show that, compared to existing methods of calculating passenger arrival times, the proposed method reduces the error by approximately 15.07%. The proposed model demonstrates good predictive accuracy and versatility in multiple case studies. It can also make relatively accurate predictions for specific dates such as holidays and has a good over-forecasting effect, facilitating the allocation of service resources. The forecasting method based on the non-stationary queuing model can provide a data-dependent foundation for the dynamic allocation of service resources at airport security checks.
-
Computer Engineering. 2026, 52(8): 364-375. https://doi.org/10.19678/j.issn.1000-3428.0070595
Brain-Computer Interface (BCI) systems based on Steady-State Visual Evoked Potential (SSVEP) show classification performance limitations due to individual differences and interference from non-target stimuli. Moreover, existing methods have not thoroughly explored the quantitative relationship between visual distraction interference and individual differences. To address these issues, we propose an SSVEP multi-domain collaborative decoding algorithm with adaptive compensation for visual distraction. The algorithm includes an adaptive label smoothing technique for visual distraction and a multi-domain joint decoding model. First, based on the theory of the visual crowding effect, an adaptive quantitative model linking signal amplitude and label noise is constructed. By dynamically adjusting the intensity of label smoothing, the model achieves a quantitative representation of individual visual distraction levels, thereby mitigating model overfitting while reducing interference from non-target stimuli and suppressing the adverse effects of individual differences. Subsequently, a multi-domain joint decoding model is proposed. This model first achieves deep collaboration across time-frequency-spatial domains through a feature extraction framework and then introduces a Bidirectional Long Short-Term Memory (BiLSTM) network to model temporal global dependencies, forming a composite feature representation that combines local receptive fields with long-range contextual awareness. Validation on three publicly available SSVEP datasets using time windows of 0.5 s and 1.0 s demonstrates that, under all experimental settings, the proposed algorithm exhibits superior average recognition accuracy and average information transfer rate compared to other state-of-the-art methods. Ablation experiments reveal that the adaptive compensation mechanism for visual distraction is effective across all experimental settings, with accuracy improvements of up to 18 percentage points in short time windows. The findings indicate that the proposed approach provides new insights for optimizing individual adaptability and fusing time-frequency features in neural decoding.
-
Computer Engineering. 2026, 52(8): 376-389. https://doi.org/10.19678/j.issn.1000-3428.0070352
The research on 2-Dimensional (2D) obstacle flow fields primarily employs numerical computations to simulate fluid motion. Factors such as mesh granularity in the model, solver algorithms, and boundary conditions significantly affect the accuracy of the computational results, as well as the computational resources and time required. Deep learning models trained on flow field data can be utilized for rapid prediction and inference tasks in real-time or large-scale flow fields, thereby substantially reducing computational costs. However, when predicting 2D flow fields, existing deep learning models require the resampling of flow field data to satisfy convolutional operation requirements, transforming originally unevenly distributed meshes into uniformly distributed meshes. This process results in feature loss, which significantly affects prediction accuracy. To address this issue, a convolutional neural network model called CAM-UNet with dynamic geometric adaptability is proposed. A Content-Aware Mixer (CAMixer) module is employed to enhance the dynamic adaptability of the U-Net model to geometric shapes and improve the flow field prediction accuracy in mesh point regions affected by resampling. Additionally, a module with gating units capable of effectively integrating the spatial and channel features of the data is designed. Based on this module, the skip connection structure of the U-Net model is reconstructed to further enhance its feature representation capability. Experiments reveal that this model shows improvement in terms of learning the nonlinear representations between the input data of 2D obstacle flow fields and the fluid velocities in the x and y directions, significantly reducing prediction errors in the horizontal velocity for regions with complex flow phenomena.
-
Computer Engineering. 2026, 52(8): 390-399. https://doi.org/10.19678/j.issn.1000-3428.0070709
Sea ice thickness is a key parameter in global climate change research, and it plays a crucial role in regulating the Earth's climate system, ocean circulation, and heat exchange. However, accurate retrieval of sea ice thickness is significantly challenging because of the highly variable physical properties of sea ice. To address this issue, this paper proposes a passive microwave remote sensing method for sea ice thickness retrieval named SIT-TransNET. This method integrates multi-feature fusion and an improved Transformer, and it utilizes brightness temperature data from the Advanced Microwave Scanning Radiometer 2 (AMSR2) satellite combined with auxiliary data (including snow surface temperature, sea surface salinity, and 1.4 GHz brightness temperature) to explore complex relationships between these data and sea ice thickness. Further, this method analyzes the importance of different features and enhances the effective representation of sea ice thickness by establishing various feature fusion schemes. The self-attention and multi-head attention mechanisms of the SIT-TransNET model enable capturing the contributions and combinations of different features and dynamically adjusting the weights of different features for accurate sea ice thickness retrieval. Compared to the other methods, experimental results demonstrate that the SIT-TransNET method significantly improves the accuracy of sea ice thickness retrieval with a coefficient of determination (R2) of 0.96 and Root Mean Square Error (RMSE) of 9 cm. This confirms that this method is suitable for sea ice thickness retrieval and provides an effective technical method for large-scale sea ice thickness monitoring and climate change research.
-
Computer Engineering. 2026, 52(8): 400-411. https://doi.org/10.19678/j.issn.1000-3428.0070383
In intelligent coal mine construction, the real-time monitoring of underground personnel in coal mines is important to ensure mine safety. However, several parameters are used for real-time monitoring, and commonly used detection models often struggle to meet these requirements when deployed underground. Moreover, given the complex environment of underground coal mines, issues such as missed or false detections frequently occur during personnel detection. To address these issues, this paper proposes a coal mine underground personnel detection algorithm based on YOLOv8 that integrates attention mechanisms and lightweight networks. First, to address deployment difficulties caused by excessive model parameters, the C2f module in the backbone network of the original model is replaced with a reconstructed lightweight module called C2f_RepGhost. Second, to enhance the detection accuracy of the model, an EMA attention mechanism is incorporated into the backbone network. Third, to strengthen the capability of the model to detect small-target personnel, the DyHead dynamic detection head is introduced. Finally, the original loss function is replaced with Inner-CIoU to optimize the target localization accuracy. Comparative experiments on the PASCAL VOC 2012 dataset and a self-built underground coal mine dataset are conducted. The results demonstrate that, compared to the original model, the improved model achieves a 1.3 percentage point increase in precision and a 1.2 percentage point increase in recall on the PASCAL VOC 2012 dataset. On the self-built dataset, the improved model reduces the parameters by 29.6% compared to the original model, and the precision and recall reach 95.3% and 90%, respectively, which are 2.4 and 3.5 percentage points, higher than those of the original model, respectively. The improved model not only reduces the number of parameters but also mitigates the issues of missed and false detections in the original model, meeting the practical requirements for personnel detection in underground coal mines.
-
Computer Engineering. 2026, 52(8): 412-421. https://doi.org/10.19678/j.issn.1000-3428.0070667
In industrial production, defect classification plays a crucial role in ensuring product quality and safety. However, industrial defect datasets are characterized by significant intra-class variations and minimal inter-class differences. Coupled with a limited number of defect samples, this results in subpar performance of existing defect classification models in real-world industrial settings. To address this issue, an industrial defect classification algorithm called VFD-SA is proposed, which is based on Variational Feature Disentanglement (VFD) and Sharpness-Aware (SA) training. First, a Variational Auto-Encoder (VAE) is introduced to disentangle the defect features into category-discriminative and intra-class variance features. Subsequently, a resampling strategy is employed to enhance the intra-class variance features, which are then combined with the original features to improve the discriminative power of the feature representations. The VFD enables the model to focus more on category-discriminative features of defects while exhibiting tolerance to details and backgrounds unrelated to defect categories, thereby enhancing the model's defect classification performance. Finally, by incorporating the SA training strategy, the geometric shape of the loss function is optimized to further improve the generalization capability of the model. Experimental results on the NEU-CLS dataset for steel rolling defects, GC10-DET dataset for metal defects, and a self-made dataset for fastener defects demonstrate that the VFD-SA achieves accuracy rates of 100%, 93.52%, and 99.48%, respectively, significantly outperforming existing defect classification algorithms and meeting defect classification requirements across various industrial scenarios.
-
Computer Engineering. 2026, 52(8): 422-430. https://doi.org/10.19678/j.issn.1000-3428.0070483
Multisensor data fusion methods can enhance the accuracy of bearing fault diagnoses. However, in the context of motor-bearing fault diagnosis, most existing multisensor data fusion methods suffer from issues such as a single type of sensor data and difficulty in fully exploiting the redundancy and complementarity among different modal data. To address this issue, a bearing fault diagnosis method based on a Multiple-Constraint Modal-Invariant Graph Convolutional Fusion Network (MCMI-GCFN) is proposed. First, features are extracted from the original current and vibration signals using a Convolutional Auto-Encoder (CAE) and Squeeze-and-Excitation block (SE block). Second, a source domain classifier and domain discriminator are introduced to capture the modal invariance among different modal data based on domain adversarial training, thereby fully exploiting the redundancy and complementarity among multimodal data. Finally, the spatial aggregation characteristics of the Graph Convolutional Neural Network (GCN) are utilized to capture the dependency relationships between features at similar time steps in the current and vibration modalities, enabling precise fusion of their contextual semantic information. This method is validated using a publicly available bearing damage current and vibration dataset obtained from the University of Paderborn, Germany. The experimental results demonstrate that the MCMI-GCFN method achieves a bearing fault diagnosis accuracy of 99.6%, which is 9 to 11.4 percentage points higher than that achieved by nonfusion methods, thereby verifying the effectiveness of the proposed model.
Volume 52 No 8
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：Wu Jiangxing
Executive Chief Editor:
Hu Bing
Tel：021-67092514
E-mail: ecice06@ecict.com.cn
CN: 31-1289/TP
ISSN: 1000-3428