Likun Lu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Likun Lu has a total of 35 co-authors. Here are the top co-authors by number of shared publications:
- Qingtao Zeng
School of Information Engineering, Beijing Institute of Graphic Communication, Beijing, China
7 shared publications - Yeli Li
School of Information Engineering, Beijing Institute of Graphic Communication, Beijing 102600, China
3 shared publications - JunFei Li
School of Information Engineering, Beijing Institute of Graphic Communication, Beijing, China
2 shared publications - Liqin Yu
School of Information Engineering, Beijing Institute of Graphic Communication, Beijing, China
2 shared publications Qingtao Zeng
Beijing key Laboratory of Signal and Information Processing for High-end Printing Equipment, Beijing Institute of Graphic Communication, Beijing, 102600, China
2 shared publications- Wenjing Li
State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications, Beijing, China
2 shared publications
Latest Publications
- Open Access
- 7 September 2026
- Published by MDPI AG in Mathematics
- Vol. 14 (17) , 3234
Frequency-Modulated Spiral Manifold: A Model for Temporal Knowledge Graph Completion
Temporal knowledge graph completion (TKGC) infers missing facts by modeling the temporal evolution of relations. Existing methods typically encode time through low-dimensional geometric transformations or frequency-domain decomposition. However, periodically recurring relations can still be mapped to overlapping trajectories in the same two-dimensional working space, which may reduce temporal separability and produce conflicting optimization signals. To address this limitation, we propose the Frequency-Modulated Spiral Manifold (FMSM) model, which includes the following: (1) a relation-adaptive spectral partition and relation-dimension gate for fusing long- and short-term relation channels; (2) an independent global phase embedding and nonlinear spiral push that lift entangled planar trajectories onto separated three-dimensional manifold layers; and (3) a spiral norm regularizer that stabilizes temporal evolution while preserving valid burst signals. The artificial-intelligence contribution of FMSM is a phase-conditioned spectral-geometric representation that separates recurring temporal facts while retaining multi-scale relation dynamics. Its engineering application is the completion of time-stamped event records for dynamic knowledge-based decision-support systems. Experiments on ICEWS14, ICEWS05-15, and GDELT show that, compared with the strongest reported baseline TeRDy, FMSM yields relative MRR improvements of 0.62%, 0.86%, and 13.28%.
- 7 September 2026
- Published by MDPI AG in Mathematics
- Vol. 14 (17) , 3234
Open AccessFrequency-Modulated Spiral Manifold: A Model for Temporal Knowledge Graph Completion
Temporal knowledge graph completion (TKGC) infers missing facts by modeling the temporal evolution of relations. Existing methods typically encode time through low-dimensional geometric transformations or frequency-domain decomposition. However, periodically recurring relations can still be mapped to overlapping trajectories in the same two-dimensional working space, which may reduce temporal separability and produce conflicting optimization signals. To address this limitation, we propose the Frequency-Modulated Spiral Manifold (FMSM) model, which includes the following: (1) a relation-adaptive spectral partition and relation-dimension gate for fusing long- and short-term relation channels; (2) an independent global phase embedding and nonlinear spiral push that lift entangled planar trajectories onto separated three-dimensional manifold layers; and (3) a spiral norm regularizer that stabilizes temporal evolution while preserving valid burst signals. The artificial-intelligence contribution of FMSM is a phase-conditioned spectral-geometric representation that separates recurring temporal facts while retaining multi-scale relation dynamics. Its engineering application is the completion of time-stamped event records for dynamic knowledge-based decision-support systems. Experiments on ICEWS14, ICEWS05-15, and GDELT show that, compared with the strongest reported baseline TeRDy, FMSM yields relative MRR improvements of 0.62%, 0.86%, and 13.28%.
- 28 June 2026
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 82 (9) , 539
Innovative weighted clustering for categorical matrix-object data: new distance and cluster center considering data distribution
- 28 June 2026
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 82 (9) , 539
Innovative weighted clustering for categorical matrix-object data: new distance and cluster center considering data distribution
- 9 March 2026
- Published by Springer Nature in Multimedia Systems
- Vol. 32 (3) , 164
Multi-domain feature enhanced adaptive fusion network for multi-modal fake news detection
- 9 March 2026
- Published by Springer Nature in Multimedia Systems
- Vol. 32 (3) , 164
Multi-domain feature enhanced adaptive fusion network for multi-modal fake news detection
- Open Access
- 17 July 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 25966
Machine learning based multi-parameter droplet optimisation model study
Continuous inkjet technology, as a key technology in the field of industrial printing, is favoured for its excellent printing speed, precision and versatility. In order to achieve the accurate generation of ideal droplets in continuous inkjet devices, this paper proposes a new parameter optimisation method, BO-GP, which combines the Bayesian optimisation algorithm with computer vision, and after 50 rounds of iterations, it can converge to the optimal values of the control parameters, and successfully constructs the Pareto frontier of the control parameters. In this paper, experiments were conducted on two different device droplet image datasets, a millimetre-scale inkjet device and a microfluidic device, respectively. Compared with the original BO in Loop method, the optimised minimum objective function value is reduced from 0.378 to 0.331 in the millimetre-scale device, and from 0.073 to 0.046 in the microfluidic device. Moreover, the Pareto solution of the 10 sets of predicted parameters output using the BO-GP method tends to be stable with fluctuations around 0.1, and it takes only 1 h to derive the control conditions for achieving high roundness, high yield and uniform size droplets.
- 17 July 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 25966
Open AccessMachine learning based multi-parameter droplet optimisation model study
Continuous inkjet technology, as a key technology in the field of industrial printing, is favoured for its excellent printing speed, precision and versatility. In order to achieve the accurate generation of ideal droplets in continuous inkjet devices, this paper proposes a new parameter optimisation method, BO-GP, which combines the Bayesian optimisation algorithm with computer vision, and after 50 rounds of iterations, it can converge to the optimal values of the control parameters, and successfully constructs the Pareto frontier of the control parameters. In this paper, experiments were conducted on two different device droplet image datasets, a millimetre-scale inkjet device and a microfluidic device, respectively. Compared with the original BO in Loop method, the optimised minimum objective function value is reduced from 0.378 to 0.331 in the millimetre-scale device, and from 0.073 to 0.046 in the microfluidic device. Moreover, the Pareto solution of the 10 sets of predicted parameters output using the BO-GP method tends to be stable with fluctuations around 0.1, and it takes only 1 h to derive the control conditions for achieving high roundness, high yield and uniform size droplets.
- 25 May 2025
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 81 (8) , 1-36
Task offloading and computing resource allocation of the joint UAV with computing power network
- 25 May 2025
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 81 (8) , 1-36
Task offloading and computing resource allocation of the joint UAV with computing power network
- Open Access
- 8 April 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 1-23
Attentive ink MLP droplet detection algorithm based on the satellite droplets threshold domain
Satellite droplet are trailing droplets caused by improper voltage waveform control in the nozzles, which directly affect the quality of inkjet printing. Traditional empirical tuning methods struggle to effectively control and predict satellite droplet. In response, this paper establishes a simulation model of a ring-shaped piezoelectric ceramic device. establishes a simulation model of a ring-shaped piezoelectric ceramic nozzle through numerical analysis, collecting droplet behavior parameters for 13,650 different waveforms. It was found that the relative differential of the negative pressure peak at the nozzle is the primary cause of satellite droplet formation. Furthermore, the critical values of various behavior parameters leading to satellite droplet formation were investigated, resulting in the construction of a “Satellite droplet formation”. resulting in the construction of a “Satellite droplet threshold domain.” Based on this domain, an attention-based MLP algorithm, the Attentive Ink MLP Based on this domain, an attention-based MLP algorithm, the Attentive Ink MLP, was proposed to automatically predict whether a waveform will generate satellite droplet. algorithm reduces prediction time from 40 min to 5 min, with a validation accuracy of 0.988. The numerical analysis and algorithm were further validated through experiments with a droplet. The numerical analysis and algorithm were further validated through experiments with a droplet observation system.
- 8 April 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 1-23
Open AccessAttentive ink MLP droplet detection algorithm based on the satellite droplets threshold domain
Satellite droplet are trailing droplets caused by improper voltage waveform control in the nozzles, which directly affect the quality of inkjet printing. Traditional empirical tuning methods struggle to effectively control and predict satellite droplet. In response, this paper establishes a simulation model of a ring-shaped piezoelectric ceramic device. establishes a simulation model of a ring-shaped piezoelectric ceramic nozzle through numerical analysis, collecting droplet behavior parameters for 13,650 different waveforms. It was found that the relative differential of the negative pressure peak at the nozzle is the primary cause of satellite droplet formation. Furthermore, the critical values of various behavior parameters leading to satellite droplet formation were investigated, resulting in the construction of a “Satellite droplet formation”. resulting in the construction of a “Satellite droplet threshold domain.” Based on this domain, an attention-based MLP algorithm, the Attentive Ink MLP Based on this domain, an attention-based MLP algorithm, the Attentive Ink MLP, was proposed to automatically predict whether a waveform will generate satellite droplet. algorithm reduces prediction time from 40 min to 5 min, with a validation accuracy of 0.988. The numerical analysis and algorithm were further validated through experiments with a droplet. The numerical analysis and algorithm were further validated through experiments with a droplet observation system.
- Open Access
- 10 February 2025
- Published by Springer Nature in Research Square
Research on the Barcode Deblurring Algorithm Based on a GAN
With the development of Internet of Things (IoT) technology, barcode automatic recognition systems play a crucial role. Traditional methods often perform poorly when processing blurred barcodes, which affects recognition and application performance. This paper proposes a barcode deblurring algorithm based on generative adversarial networks (GANs), aimed at overcoming the problem of insufficient barcode clarity in traditional image processing. First, the SE attention mechanism is combined with the aggregation residual block ResNeXt to form SE-ResNeXt, replacing the residual block ResNet, which accelerates the model’s convergence speed and enhances the stability of the training process. Second, the channel prior convolutional attention (CPCA) mechanism is introduced to improve the network's feature extraction ability and detection performance. The experimental results show that the proposed model achieves a peak signal-to-noise ratio (PSNR) of 30.48 dB, an improvement of 4.87 dB over the baseline network, and a structural similarity index (SSIM) of 0.9383, an improvement of 7.72%. The subjective visual deblurring effect is also promising, with restored barcode images showing clear edge contours and noticeable detail recovery.
- 10 February 2025
- Published by Springer Nature in Research Square
Open AccessResearch on the Barcode Deblurring Algorithm Based on a GAN
With the development of Internet of Things (IoT) technology, barcode automatic recognition systems play a crucial role. Traditional methods often perform poorly when processing blurred barcodes, which affects recognition and application performance. This paper proposes a barcode deblurring algorithm based on generative adversarial networks (GANs), aimed at overcoming the problem of insufficient barcode clarity in traditional image processing. First, the SE attention mechanism is combined with the aggregation residual block ResNeXt to form SE-ResNeXt, replacing the residual block ResNet, which accelerates the model’s convergence speed and enhances the stability of the training process. Second, the channel prior convolutional attention (CPCA) mechanism is introduced to improve the network's feature extraction ability and detection performance. The experimental results show that the proposed model achieves a peak signal-to-noise ratio (PSNR) of 30.48 dB, an improvement of 4.87 dB over the baseline network, and a structural similarity index (SSIM) of 0.9383, an improvement of 7.72%. The subjective visual deblurring effect is also promising, with restored barcode images showing clear edge contours and noticeable detail recovery.
- 19 November 2024
- Published by Springer Nature in Multimedia Systems
- Vol. 30 (6) , 1-20
Exploiting heterogeneous information isolation and multi-view aggregation for multimodal recommendation
- 19 November 2024
- Published by Springer Nature in Multimedia Systems
- Vol. 30 (6) , 1-20
Exploiting heterogeneous information isolation and multi-view aggregation for multimodal recommendation
- Open Access
- 17 October 2024
- Published by MDPI AG in Electronics
- Vol. 13 (20) , 4076
Intelligent Semantic Communication System Based on Kolmogorov–Arnold Networks Driven by Dynamic Terminal-Side Computing Power Network
With the advent of the 6G era, the number of IoT devices has experienced explosive growth, leading to the generation of massive amounts of data at the network edge. Semantic communication, as an innovative solution to handling this data deluge, can significantly enhance communication efficiency. However, the limited storage and computational resources of terminal devices constrain the widespread application of semantic communication in 6G networks. To address this issue, we propose a terminal-side-computing-driven intelligent semantic communication solution. Specifically, we introduce a semantic communication model based on Kolmogorov–Arnold Networks (KANs), named K-DeepSC. Using image-reconstruction tasks as an example, the proposed K-DeepSC reduces the number of model parameters by 44% compared to semantic communication models based on Multi-Layer Perceptrons (MLPs), while maintaining similar performance. Furthermore, to fully leverage idle terminal computing power for semantic tasks, we explore computation offloading in dynamic Terminal-Side Computing Power Networks. By optimizing task delay minimization, a deep reinforcement learning algorithm is employed to determine the optimal offloading strategy. Simulation results demonstrate that our proposed solution effectively reduces semantic task processing delay.
- 17 October 2024
- Published by MDPI AG in Electronics
- Vol. 13 (20) , 4076
Open AccessIntelligent Semantic Communication System Based on Kolmogorov–Arnold Networks Driven by Dynamic Terminal-Side Computing Power Network
With the advent of the 6G era, the number of IoT devices has experienced explosive growth, leading to the generation of massive amounts of data at the network edge. Semantic communication, as an innovative solution to handling this data deluge, can significantly enhance communication efficiency. However, the limited storage and computational resources of terminal devices constrain the widespread application of semantic communication in 6G networks. To address this issue, we propose a terminal-side-computing-driven intelligent semantic communication solution. Specifically, we introduce a semantic communication model based on Kolmogorov–Arnold Networks (KANs), named K-DeepSC. Using image-reconstruction tasks as an example, the proposed K-DeepSC reduces the number of model parameters by 44% compared to semantic communication models based on Multi-Layer Perceptrons (MLPs), while maintaining similar performance. Furthermore, to fully leverage idle terminal computing power for semantic tasks, we explore computation offloading in dynamic Terminal-Side Computing Power Networks. By optimizing task delay minimization, a deep reinforcement learning algorithm is employed to determine the optimal offloading strategy. Simulation results demonstrate that our proposed solution effectively reduces semantic task processing delay.