15 September 2026, Volume 46 Issue 5
-
Select all|
-
Journal of Hangzhou Dianzi University. 2026, 46(5): 1-10. https://doi.org/10.13954/j.cnki.hdu.2026.05.001To address the issue of parameter redundancy in rotated object detection of SAR images for ship targets, this paper proposes an efficient lightweight detection network named LRSAR-YOLO. The framework implements three key innovations: a lightweight high-low frequency feature fusion module based on grouped and depthwise separable convolutions, a lightweight progressive fusion pyramid preserving image details through hierarchical feature integration, and a rotation detection head employing shared and reparameterized convolutions to maintain accuracy while reducing parameters.The proposed Efficiency Compression Rate (ECR) metric is designed to quantify the effectiveness of model compression.Experimental results demonstrate 54.2% computation reduction and 71.5% parameter reduction compared with YOLOv8 variants on SSDD+ and RSDD datasets, achieving 53.5% and 55.6% ECR values respectively. Deployment verification on the RK3588 embedded platform demonstrates that, compared to the baseline model, the proposed model improves precision, recall, and mean average precision by 2.1%, 5.0%, and 5.3%, respectively, while increasing detection speed by 51.7%. The findings validate ECR's effectiveness and confirm the algorithm's deployment advantages in resource-constrained environments.
-
Journal of Hangzhou Dianzi University. 2026, 46(5): 11-20. https://doi.org/10.13954/j.cnki.hdu.2026.05.002With the introduction of new energy into the network, such as wind power and solar energy, the intermittency and uncertainty inevitably increase the frequency fluctuation of the system. Compared with traditional thermal power unit FM with large limitations and long response time, electric vehicle (EV) has become a research hotspot of auxiliary power system FM due to its flexible scheduling and large scale. Aiming at the problem of unbalanced state of charge (SOC) when EVs participated in the secondary frequency modulation of power grid, a control strategy including SOC balance was proposed. The EV power was adjusted according to the weight of SOC level to achieve the purpose of SOC balance and improve the frequent switching of EV states of charge and discharge,and a comparative analysis is made with the traditional SOC balancing strategy. The comparison results show that the proposed scheme has a more significant SOC balancing effect on EV.
-
Journal of Hangzhou Dianzi University. 2026, 46(5): 21-29. https://doi.org/10.13954/j.cnki.hdu.2026.05.003Cardiotocography (CTG) is widely used to assess fetal health during pregnancy by recording fetal heart rate (FHR) and uterine contraction (UC) signals, assisting obstetricians in identifying fetal hypoxia or other potential risks. Intelligent CTG based on deep learning can improve the accuracy of FHR analysis and reduce false positives by automating the process. This paper proposes the GraphFHR-Based Abnormal Fetal Heart Rate Detection Network (GFNet), which uses a Transformer to extract time-domain features, and applies a Graph Attention Network (GAT) for graph-level classification.Through extensive experiments on the public dataset CTU-UHB, GFNet achieved an accuracy of 84.38%, precision of 81.40%, recall of 83.33%, F1 score of 82.35%, and area under the curve of 81.32%. These results indicate that the proposed approach delivers strong performance in detecting abnormal FHR patterns.
-
Journal of Hangzhou Dianzi University. 2026, 46(5): 30-38. https://doi.org/10.13954/j.cnki.hdu.2026.05.004To further balance spectral and energy efficiency while reducing the computational overhead, a low-complexity and high-energy-efficiency hybrid precoding design is proposed for millimeter-wave massive MIMO with hybrid connection architecture. First, apply the singular value decomposition (SVD) technique to the optimal fully-digital precoding matrix; then, select vectors from the left singular matrix obtained through decomposition to construct the analog precoding matrix, according to the quantitative relationship between the data streams and the number of radio frequency chains; finally, based on the given analog precoding matrix, transform the matrix decomposition problem into a matrix trace maximization problem, approximate the objective function by using the Hölder inequality and the Schatten norm, and solve for the optimal digital precoding matrix. This algorithm does not require the number of data streams to match that of radio frequency chains, and has no constraint that the digital precoding matrix must satisfy a block-diagonal structure. Meanwhile, the solution of the analog precoding matrix is independent of the digital precoding matrix, which avoids cyclic iterative operations and reduces the computational complexity. Simulation results show that the proposed hybrid connection architecture algorithm achieves higher energy efficiency with lower computational complexity while maintaining high spectral efficiency.
-
Journal of Hangzhou Dianzi University. 2026, 46(5): 39-48. https://doi.org/10.13954/j.cnki.hdu.2026.05.005To address the challenges of high hardware complexity and high channel sensitivity of terahertz band communication, this paper proposes a high-speed parallel timing synchronization algorithm suitable for 20 Gbps rate, and completes its hardware implementation and successfully deploys it in 220 GHz terahertz high-speed communication system. The algorithm achieves a precise reduction in hardware resource consumption while maintaining interpolation accuracy by designing parallel Farrow interpolation filter based on address query, parallel Mueller&Muller timing error detection module, a pruning numerically controlled oscillator (NCO), a pruning loop filter and other optimization units. Test results show that the equivalent throughput of the algorithm can reach 16 Gsps, and the resource consumption of the digital signal processor is reduced by 21.8%. The proposed algorithm effectively balances the high-speed processing requirements and hardware resource constraints of terahertz communication, successfully supports the ultra-high-speed real-time signal processing in the 220 GHz system, and provides a high-performance and high-efficiency timing synchronization solution for the engineering implementation of modern ultra-high-speed communication systems.
-
Journal of Hangzhou Dianzi University. 2026, 46(5): 49-57. https://doi.org/10.13954/j.cnki.hdu.2026.05.006The miniaturization of lasers not only promotes the development of various small-scale lasers but also provides a platform for studying the light-matter interactions at the microscopic level. Based on a stochastic simulation method, this paper numerically simulates the nonlinear dynamical behaviors of lasers at different scales, with a particular focus on the influence of the spontaneous emission factor on stimulated emission behavior. The evolution of time-domain dynamical behaviors in different lasers under external modulation is analyzed. The study reveals that under free running conditions, as the spontaneous emission factor increases, the efficiency of converting intracavity spontaneous emission photons into stimulated radiation gradually improves, resulting in a continuous reduction of the laser threshold. Furthermore, by introducing external sinusoidal modulation, the impact of the spontaneous emission factor on the modulation response of lasers is investigated. This research can provide theoretical references for further exploring the nonlinear dynamical characteristics of small-scale lasers.
-
Journal of Hangzhou Dianzi University. 2026, 46(5): 58-67. https://doi.org/10.13954/j.cnki.hdu.2026.05.007Vision-guided flexible manufacturing systems are progressively integrating multiple computation-intensive tasks, including visual perception, intelligent algorithms, and real-time motion control. Therefore, heterogeneous computing units must be incorporated. However, the dependencies among tasks are complex, and the execution efficiency varies considerably across different computing resources. To improve the overall task execution efficiency, it is essential to optimize the task execution order and computing resource allocation via task scheduling algorithms. Existing heterogeneous scheduling methods mainly fall into two categories: rule-based heuristics and meta-heuristics. The former suffer from limited search capability, whereas the latter still has room for improvement in in both search-space exploration and solution-quality enhancement. To address these issues, a hierarchical heuristic algorithm (HHA) is proposed for DAG task scheduling. Through the collaboration of a genetic algorithm and a discrete ant colony optimization algorithm (DACO), the proposed approach achieves achieves stage-wise optimization of task scheduling solutions. While ensuring scheduling feasibility, the proposed method effectively improves the solution quality. Comparative experimental results demonstrate that, compared with typical algorithms such as GA-DE, NGA, HGA, and HEFT, the proposed HHA achieves performance improvements of 0.77%, 0.87%, 4.82%, and 8.59% in terms of makespan, respectively. These results validate the effectiveness of the proposed method in heterogeneous control system scheduling problems.
-
Journal of Hangzhou Dianzi University. 2026, 46(5): 68-76. https://doi.org/10.13954/j.cnki.hdu.2026.05.008To recover high-temperature exhaust heat from gas turbines, this study designs and tests a novel micro-tube air-to-air heat exchanger featuring a micro-tube array and segmented baffle plates Through comparative tests of six heat exchanger configurations, an optimization study of the micro-tube heat exchanger structure was conducted. Results indicate that Structure 5 (tube diameter 1.4 mm, 8 baffles) exhibits the best overall performance. Compared with the baseline Structure 1, Structure 5 achieved a 39.5% increase in overall heat transfer coefficient and an 80.4% improvement in Q·/ΔP. Based on Structure 5, variable operating condition experiments were conducted to analyze the influence of flow rate on the heat exchanger's thermal-hydraulic performance. Results show that the overall heat transfer coefficient increases with rising flow rate before gradually leveling off, whereas pressure drop had a linear increase. Furthermore, the heat exchanger's overall performance was evaluated through shell-to-tube flow ratio analysis, indicating that the system exhibited optimal heat-hydraulic comprehensive performance ranges from 1.0 to 1.2. Finally, a performance comparison with traditional micro-tube heat exchangers was conducted, which verified the performance advantages and engineering practicability of the newly designed heat exchanger.
-
Journal of Hangzhou Dianzi University. 2026, 46(5): 77-83. https://doi.org/10.13954/j.cnki.hdu.2026.05.009The input-to-state stability problem of interval type-2 fuzzy systems with parameter uncertainties under denial-of-service attacks based on event-triggering mechanism was investigated. First, a new event-triggering mechanism was designed for networked control systems without considering denial-of-service attacks, and the input-to-state stability of the closed-loop system was guaranteed. Then, to address the issue of signal transmission interruption caused by denial-of-service attacks, a predictor was used to compensate for the lost signals in the networked control system. By combining event-triggering and the predictor, a Lyapunov function was appropriately chosen to derive the sufficient conditions for ensuring input-to-state stability of the closed-loop system. Finally, simulation results verified the effectiveness of the proposed control strategy.
-
Journal of Hangzhou Dianzi University. 2026, 46(5): 84-91. https://doi.org/10.13954/j.cnki.hdu.2026.05.010The support tensor machine model is a high-order extension of support vector machine. It has attracted considerable attention due to its ability to directly handle tensor data classification and is widely used in fields such as image processing, video analysis, and machine learning. To address the distribution uncertainty problem existing in tensor data classification, this paper proposes a distributionally robust chance-constrained support tensor machine model based on rank-one tensor. Considering the distribution sets characterized by the second-order moment information of random variables, the original problem is transformed into a second-order cone programming problem using Chebyshev's inequality, and solved using the alternating optimization method, which effectively solved the classification problem in fuzzy data scenarios. Finally, numerical experiments are conducted to verify the robustness of the model.
-
Journal of Hangzhou Dianzi University. 2026, 46(5): 92-102. https://doi.org/10.13954/j.cnki.hdu.2026.05.011Vessel magnetic signal data plays an important role in research fields such as vessel demagnetization and aeromagnetic exploration. However, due to the influence of site, equipment and geomagnetic environment, it is not easy to obtain vessel magnetic signal data in a wide range. Aiming at the limitation of direct measurement of magnetic field around vessel and the problem that the measured data are easily affected by environmental noise, a method of modeling magnetic field around vessel based on the magnetic gradient tensor data of vessel near field plane is studied. Firstly, a cross flux gate array is used to convert the magnetic induction intensity data to the magnetic gradient tensor data. Secondly, the magnetic field equivalent model of multi-magnetic dipole vessel based on magnetic gradient tensor is proposed. And then, the parameters of the magnetic dipole in the model are solved by using stepwise regression method and multi-objective adaptive elite mutation differential evolution algorithm. Finally, the magnetic field around the vessel is predicted based on the multi-magnetic dipole model. Experimental results show that in the scenario of simulating measuring vessel magnetic fields, the maximum prediction error of this method for the magnetic field of hollow steel pipes is 331.758 nT, with a prediction accuracy of 90%, which largely meets practical application requirements.
