Zehua Gao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Zehua Gao has a total of 44 co-authors. Here are the top co-authors by number of shared publications:
Chuwen Lan
School of Information and Communication Engineering, Beijing Laboratory of Advanced Information Networks, Beijing Key Laboratory of Network System Architecture and Convergence, Beijing University of Posts and Telecommunications, Beijing, 100876, China
6 shared publications- Chuwen Lan
Beijing University of Posts and Telecommunications
5 shared publications - Bo Li
Advanced Materials Institute, Shenzhen Graduate School, Tsinghua University, Shenzhen, China
2 shared publications Di Zhu
Beijing Laboratory of Advanced Information Networks & Beijing Key Laboratory of Network System Architecture and Convergence Acknowledgment, School of Information and Communication Engineering, Beijing University of Posts and Telecommunications, Beijing 100876, China
2 shared publications- Ke Bi
State Key Laboratory of Information Photonics and Optical Communications, School of Physical Science and Technology, Beijing University of Posts and Telecommunications, Beijing, China
2 shared publications Caiyong Wang
School of Electrical and Information Engineering Beijing University of Civil Engineering and Architecture Beijing China
1 shared publication
Latest Publications
- Open Access
- 5 August 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (15) , 7811
OTG: A Physics-Informed Hybrid Interpolation Framework for High-Precision 3D S-Wave Velocity Modeling
Three-dimensional (3D) S-wave velocity field modeling is a critical task in seismic exploration, but balancing modeling accuracy and geological rationality remains challenging due to sparse observation data and inherent limitations of existing methods. To address this issue, we propose an Ordinary Kriging-Thin Plate Spline-Graph Convolutional Network (OTG) fusion model. It dynamically integrates the global trend capture capability of Ordinary Kriging, the local smooth processing ability of Thin Plate Spline, and the nonlinear feature fitting performance of Graph Convolutional Network (GCN) via an adaptive gating fusion mechanism. A multi-dimensional physical constraint loss function is further designed to ensure the geophysical plausibility of interpolation results. Validated on a dataset from 105 seismic stations in Southwest China using spatial cross-validation and random repeated experiments, the full OTG model achieves a root mean square error (RMSE) of 0.1148 km/s, a mean absolute percentage error (MAPE) of 1.9614%, and a Pearson correlation coefficient (PCC) of 0.9801. Compared with the optimal traditional method (OK) and the state-of-the-art hybrid method (DeepKriging), the proposed model reduces the root mean square error (RMSE) by 42.8% and 8.2%, respectively. This study demonstrates that the OTG model realizes the complementary advantages of traditional geoscientific methods and deep learning, providing a reliable engineering solution for high-precision 3D S-wave velocity structure interpolation in seismic exploration.
- 5 August 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (15) , 7811
Open AccessOTG: A Physics-Informed Hybrid Interpolation Framework for High-Precision 3D S-Wave Velocity Modeling
Three-dimensional (3D) S-wave velocity field modeling is a critical task in seismic exploration, but balancing modeling accuracy and geological rationality remains challenging due to sparse observation data and inherent limitations of existing methods. To address this issue, we propose an Ordinary Kriging-Thin Plate Spline-Graph Convolutional Network (OTG) fusion model. It dynamically integrates the global trend capture capability of Ordinary Kriging, the local smooth processing ability of Thin Plate Spline, and the nonlinear feature fitting performance of Graph Convolutional Network (GCN) via an adaptive gating fusion mechanism. A multi-dimensional physical constraint loss function is further designed to ensure the geophysical plausibility of interpolation results. Validated on a dataset from 105 seismic stations in Southwest China using spatial cross-validation and random repeated experiments, the full OTG model achieves a root mean square error (RMSE) of 0.1148 km/s, a mean absolute percentage error (MAPE) of 1.9614%, and a Pearson correlation coefficient (PCC) of 0.9801. Compared with the optimal traditional method (OK) and the state-of-the-art hybrid method (DeepKriging), the proposed model reduces the root mean square error (RMSE) by 42.8% and 8.2%, respectively. This study demonstrates that the OTG model realizes the complementary advantages of traditional geoscientific methods and deep learning, providing a reliable engineering solution for high-precision 3D S-wave velocity structure interpolation in seismic exploration.
- 2 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Antennas and Wireless Propagation Letters
- Vol. 25 (3) , 944-948
A Multiband Stealth Metasurface With Smart Temperature Regulation
multifunctional stealth metasurface (MSM) integrating multiband stealth and thermal regulation is proposed. The proposed structure demonstrates multispectral stealth capability by combining over 90% microwave absorption within 6.74-13.06 GHz, low infrared emissivity below 0.26 across 5-18 μm, and maintained optical transparency in the visible range. The combined use of low-emissivity surface and an air-gap structure ensures efficient thermal insulation. Its simple monolayer design considerably simplifies fabrication and installation. This work demonstrates significant potential for multispectral stealth and thermal management applications.
- 2 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Antennas and Wireless Propagation Letters
- Vol. 25 (3) , 944-948
A Multiband Stealth Metasurface With Smart Temperature Regulation
multifunctional stealth metasurface (MSM) integrating multiband stealth and thermal regulation is proposed. The proposed structure demonstrates multispectral stealth capability by combining over 90% microwave absorption within 6.74-13.06 GHz, low infrared emissivity below 0.26 across 5-18 μm, and maintained optical transparency in the visible range. The combined use of low-emissivity surface and an air-gap structure ensures efficient thermal insulation. Its simple monolayer design considerably simplifies fabrication and installation. This work demonstrates significant potential for multispectral stealth and thermal management applications.
- Open Access
- 18 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 1213-1218
A Liquid Refractive Index Sensor Based On Perforated Structure
This article proposes a rectangular cavity resonator that can achieve liquid refractive index sensing. Place a square solid dielectric block inside the cavity, drill uniformly distributed holes in front of and on top of the dielectric block, and observe the small changes in the refractive index of the liquid by measuring the changes in resonant frequency and microwave transmission coefficient S21. After analyzing various structural parameters, the article achieved a sensitivity of 4122MHz/RIU near 26.3GHz, which is 2.37 times higher than the sensitivity of solid structure sensors. This result also provides a new approach for the design and performance improvement of future high-sensitivity refractive index sensors.
- 18 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 1213-1218
Open AccessA Liquid Refractive Index Sensor Based On Perforated Structure
This article proposes a rectangular cavity resonator that can achieve liquid refractive index sensing. Place a square solid dielectric block inside the cavity, drill uniformly distributed holes in front of and on top of the dielectric block, and observe the small changes in the refractive index of the liquid by measuring the changes in resonant frequency and microwave transmission coefficient S21. After analyzing various structural parameters, the article achieved a sensitivity of 4122MHz/RIU near 26.3GHz, which is 2.37 times higher than the sensitivity of solid structure sensors. This result also provides a new approach for the design and performance improvement of future high-sensitivity refractive index sensors.
- Open Access
- 3 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Photonics Journal
- Vol. 16 (6) , 1-6
Large-Area, Broadband, and Flexible Terahertz Antireflection Thin Film Based on an All-Dielectric Metasurface
Terahertz electromagnetic waves are widely used in various fields, but their practical application often suffers from significant interfacial losses. In this paper, we present a terahertz antireflection film based on an all-dielectric metasurface consisting of a zirconia microsphere array assembled using a template-assisted method. Simulations and experiments show that the film has a broadband, outstanding transmission-enhancing effect. Because the antireflection film has the advantages of ultralow cost, high compatibility, flexibility, and easy industrialization, it may have potential applications in various fields.
- 3 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Photonics Journal
- Vol. 16 (6) , 1-6
Open AccessLarge-Area, Broadband, and Flexible Terahertz Antireflection Thin Film Based on an All-Dielectric Metasurface
Terahertz electromagnetic waves are widely used in various fields, but their practical application often suffers from significant interfacial losses. In this paper, we present a terahertz antireflection film based on an all-dielectric metasurface consisting of a zirconia microsphere array assembled using a template-assisted method. Simulations and experiments show that the film has a broadband, outstanding transmission-enhancing effect. Because the antireflection film has the advantages of ultralow cost, high compatibility, flexibility, and easy industrialization, it may have potential applications in various fields.
- 20 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Network
- Vol. 39 (1) , 220-226
Network Planning for Time-Sensitive Communications
Network planning is a critical process in the construction of communication networks. Time-sensitive communications require networks with time-aware scheduling and transmission mechanisms to guarantee the deterministic requirements of services, which is difficult to be satisfied by traditional network planning. This article proposes an SDN-enabled network planning solution for time-sensitive communications. Specifically, the solution introduces a layered network architecture by decoupling network planes with different functions, designs a planning strategy for deterministic services, and sorts out a time-sensitive network planning procedure that integrates topology construction, routing, and traffic scheduling. This article uses the scalefree topology construction method to construct delay-bounded network topology. Experimental results show that the solution can guarantee the bounded delay, improve the schedulability, and reduce the algorithm runtime. In addition, the solution is able to utilize network resources according to the capacity of each node.
- 20 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Network
- Vol. 39 (1) , 220-226
Network Planning for Time-Sensitive Communications
Network planning is a critical process in the construction of communication networks. Time-sensitive communications require networks with time-aware scheduling and transmission mechanisms to guarantee the deterministic requirements of services, which is difficult to be satisfied by traditional network planning. This article proposes an SDN-enabled network planning solution for time-sensitive communications. Specifically, the solution introduces a layered network architecture by decoupling network planes with different functions, designs a planning strategy for deterministic services, and sorts out a time-sensitive network planning procedure that integrates topology construction, routing, and traffic scheduling. This article uses the scalefree topology construction method to construct delay-bounded network topology. Experimental results show that the solution can guarantee the bounded delay, improve the schedulability, and reduce the algorithm runtime. In addition, the solution is able to utilize network resources according to the capacity of each node.
- 12 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Informatics
- Vol. 20 (9) , 11485-11495
User-Centric HetNet Handover in Industrial Context Based on Pareto-Efficient Multiagent Transformer
Expanding industrial components and network density raise challenges in the domain of mobility management of multiagent systems (MASs), such as multirobot cooperative transportation. This article investigates the heterogeneous network (HetNet) handover problem in the industrial context involves jointly optimizing data rate, block rate, and handover frequency among large-scale mobile user Terminals. Specifically, by introducing user-centric conditional handover features, we leverage Pareto-efficient solutions to address the multiobjective optimization problem of balancing data rate and block probability. The optimization problem is reformulated into a multiagent learning-based Markov cooperative game to cope with dynamic context conditions, introducing a handover penalty factor to enhance service continuity. Furthermore, we develop a Pareto-efficient multiagent transformer with efficient advantage decomposition, leveraging sequential modeling, and distributed computing power of MAS. Extensive simulations demonstrate the superiority of the proposed algorithm, implementing user-centric optimal handover decisions, while also obtaining an additional fairness gain.
- 12 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Informatics
- Vol. 20 (9) , 11485-11495
User-Centric HetNet Handover in Industrial Context Based on Pareto-Efficient Multiagent Transformer
Expanding industrial components and network density raise challenges in the domain of mobility management of multiagent systems (MASs), such as multirobot cooperative transportation. This article investigates the heterogeneous network (HetNet) handover problem in the industrial context involves jointly optimizing data rate, block rate, and handover frequency among large-scale mobile user Terminals. Specifically, by introducing user-centric conditional handover features, we leverage Pareto-efficient solutions to address the multiobjective optimization problem of balancing data rate and block probability. The optimization problem is reformulated into a multiagent learning-based Markov cooperative game to cope with dynamic context conditions, introducing a handover penalty factor to enhance service continuity. Furthermore, we develop a Pareto-efficient multiagent transformer with efficient advantage decomposition, leveraging sequential modeling, and distributed computing power of MAS. Extensive simulations demonstrate the superiority of the proposed algorithm, implementing user-centric optimal handover decisions, while also obtaining an additional fairness gain.
- Open Access
- 1 September 2021
- Published by IOP Publishing in Journal of Physics: Conference Series
- Vol. 2031 (1) , 012013
Residual Attention Fusion Network for Single Image Super-Resolution
Recently, a very deep convolutional neural network (CNN) demonstrated influential performance in the field of single image super-resolution (SISR). However, most of the CNN-based methods focus on designing deeper and wider network structures alone and do not use the hierarchical and global features in the input image. Therefore, we proposed a residual attention fusion network (RAFN), which is an improved residual fusion (RF) framework, to effectively extract hierarchical features for use in single-image super-resolution. The proposed framework comprises two residual fusion structures composed of several residual and fusion modules, and a continuous memory mechanism is realized by adding a long and short jump connection. The network focuses on learning more effective features. Furthermore, to maximize the power of the RF framework, we introduced global context attention (GCA) module that can model the global context and capture long-distance dependencies. The final RAFN was constructed by applying the proposed RF framework to the GCA blocks. Extensive experiments showed that the proposed network achieved improved performance in the SISR method with fewer parameters, as compared to the methods proposed in previous studies.
- 1 September 2021
- Published by IOP Publishing in Journal of Physics: Conference Series
- Vol. 2031 (1) , 012013
Open AccessResidual Attention Fusion Network for Single Image Super-Resolution
Recently, a very deep convolutional neural network (CNN) demonstrated influential performance in the field of single image super-resolution (SISR). However, most of the CNN-based methods focus on designing deeper and wider network structures alone and do not use the hierarchical and global features in the input image. Therefore, we proposed a residual attention fusion network (RAFN), which is an improved residual fusion (RF) framework, to effectively extract hierarchical features for use in single-image super-resolution. The proposed framework comprises two residual fusion structures composed of several residual and fusion modules, and a continuous memory mechanism is realized by adding a long and short jump connection. The network focuses on learning more effective features. Furthermore, to maximize the power of the RF framework, we introduced global context attention (GCA) module that can model the global context and capture long-distance dependencies. The final RAFN was constructed by applying the proposed RF framework to the GCA blocks. Extensive experiments showed that the proposed network achieved improved performance in the SISR method with fewer parameters, as compared to the methods proposed in previous studies.
- 3 May 2021
- Published by Wiley in International Journal of Imaging Systems and Technology
- Vol. 31 (4) , 2158-2169
Deep learningsuper‐resolutionelectron microscopy based on deep residual attention network
Field‐emission scanning electron microscopy has become a fundamental research tool in the fields of medicine and materials science owing to its effectiveness. However, an inherent contradiction exists between the resolution of field‐emission scanning electron microscopy and its field‐of‐view. To solve this problem, we propose a deep learning‐based method for electron microscopy that can simultaneously obtain a large field‐of‐view and ultrahigh resolution. To solve the super‐resolution problem, a deep residual attention network is designed based on residual learning and the attention mechanism, wherein the backbone network employs several residual groups to effectively extract the features; moreover, attention groups append the backbone part to refine and fuse the features. Besides, a high‐frequency information retention module is added to acquire high‐frequency signals, acting as an effective complement to the deep residual attention network. Owing to the lack of super‐resolution datasets for electron microscopy, we created a microscopic butterfly wing dataset. In the experiments, MixUp was also applied to super‐resolution problem as a simple and effective data augmentation method, to provide more data to train the model. To evaluate the proposed method, we used standard and self‐made datasets with peak signal‐to‐noise ratio (PSNR) and structural similarity (SSIM) as criteria. The results show that the proposed model exhibits a state‐of‐the‐art performance with a PSNR of 25.12 dB and an SSIM of 0.64, and makes field‐emission scanning electron microscopy more practical and promising in the field of optical device optimization.
- 3 May 2021
- Published by Wiley in International Journal of Imaging Systems and Technology
- Vol. 31 (4) , 2158-2169
Deep learningsuper‐resolutionelectron microscopy based on deep residual attention network
Field‐emission scanning electron microscopy has become a fundamental research tool in the fields of medicine and materials science owing to its effectiveness. However, an inherent contradiction exists between the resolution of field‐emission scanning electron microscopy and its field‐of‐view. To solve this problem, we propose a deep learning‐based method for electron microscopy that can simultaneously obtain a large field‐of‐view and ultrahigh resolution. To solve the super‐resolution problem, a deep residual attention network is designed based on residual learning and the attention mechanism, wherein the backbone network employs several residual groups to effectively extract the features; moreover, attention groups append the backbone part to refine and fuse the features. Besides, a high‐frequency information retention module is added to acquire high‐frequency signals, acting as an effective complement to the deep residual attention network. Owing to the lack of super‐resolution datasets for electron microscopy, we created a microscopic butterfly wing dataset. In the experiments, MixUp was also applied to super‐resolution problem as a simple and effective data augmentation method, to provide more data to train the model. To evaluate the proposed method, we used standard and self‐made datasets with peak signal‐to‐noise ratio (PSNR) and structural similarity (SSIM) as criteria. The results show that the proposed model exhibits a state‐of‐the‐art performance with a PSNR of 25.12 dB and an SSIM of 0.64, and makes field‐emission scanning electron microscopy more practical and promising in the field of optical device optimization.
- Open Access
- 1 December 2020
- Published by IOP Publishing in Journal of Physics: Conference Series
- Vol. 1693 (1) , 012164
Deep Residual Fusion Network for Single Image Super-Resolution
Convolutional neural networks have been applied in the field of single-image super-resolution (SISR) and have achieved a series of outstanding results. However, most of the SISR research still attempt to pursue wider and deeper network structure, without paying enough attention to the correlations between different features. In order to solve these problems, deep residual fusion network (DRFN) is proposed for more powerful feature expression and feature learning. Specifically, we propose a feature fusion group (FFG) structure, which can effectively use the relevant features extracted from the residual attention group (RAG) and fuse them to be more discriminative. Residual attention group (RAG) includes channel attention module (CAM) and spatial attention module (SAM), which uses attention mechanism to refine features. DRFN also makes full use of nested residual connections, skipping redundant low-frequency information to enhance circulation, thereby focusing the calculation on more important high-frequency components. Extensive experimental results have proved the effectiveness of our model. And our model finally achieves excellent performance in terms of both quantitative metrics and visual quality.
- 1 December 2020
- Published by IOP Publishing in Journal of Physics: Conference Series
- Vol. 1693 (1) , 012164
Open AccessDeep Residual Fusion Network for Single Image Super-Resolution
Convolutional neural networks have been applied in the field of single-image super-resolution (SISR) and have achieved a series of outstanding results. However, most of the SISR research still attempt to pursue wider and deeper network structure, without paying enough attention to the correlations between different features. In order to solve these problems, deep residual fusion network (DRFN) is proposed for more powerful feature expression and feature learning. Specifically, we propose a feature fusion group (FFG) structure, which can effectively use the relevant features extracted from the residual attention group (RAG) and fuse them to be more discriminative. Residual attention group (RAG) includes channel attention module (CAM) and spatial attention module (SAM), which uses attention mechanism to refine features. DRFN also makes full use of nested residual connections, skipping redundant low-frequency information to enhance circulation, thereby focusing the calculation on more important high-frequency components. Extensive experimental results have proved the effectiveness of our model. And our model finally achieves excellent performance in terms of both quantitative metrics and visual quality.
- Open Access
Deep Learning for Super-Resolution in a Field Emission Scanning Electron Microscope
A field emission scanning electron microscope (FESEM) is a complex scanning electron microscope with ultra-high-resolution image scanning, instant printing, and output storage capabilities. FESEMs have been widely used in fields such as materials science, biology, and medical science. However, owing to the balance between resolution and field of view (FOV), when locating a target using an FESEM, it is difficult to view specific details in an image with a large FOV and high resolution simultaneously. This paper presents a deep neural network to realize super-resolution of an FESEM image. This technology can effectively improve the resolution of the acquired image without changing the physical structure of the FESEM, thus resolving the constraint problem between the resolution and FOV. Experimental results show that the apply of a deep neural network only requires a single image acquired by an FESEM to be the input. A higher resolution image with a large FOV and excellent noise reduction is obtained within a short period of time. To verify the effect of the model numerically, we evaluated the image quality by using the peak signal-to-noise ratio value and structural similarity index value, which can reach 26.88 dB and 0.7740, respectively. We believe that this technology will improve the quality of FESEM imaging and be of significance in various application fields.
Open AccessDeep Learning for Super-Resolution in a Field Emission Scanning Electron Microscope
A field emission scanning electron microscope (FESEM) is a complex scanning electron microscope with ultra-high-resolution image scanning, instant printing, and output storage capabilities. FESEMs have been widely used in fields such as materials science, biology, and medical science. However, owing to the balance between resolution and field of view (FOV), when locating a target using an FESEM, it is difficult to view specific details in an image with a large FOV and high resolution simultaneously. This paper presents a deep neural network to realize super-resolution of an FESEM image. This technology can effectively improve the resolution of the acquired image without changing the physical structure of the FESEM, thus resolving the constraint problem between the resolution and FOV. Experimental results show that the apply of a deep neural network only requires a single image acquired by an FESEM to be the input. A higher resolution image with a large FOV and excellent noise reduction is obtained within a short period of time. To verify the effect of the model numerically, we evaluated the image quality by using the peak signal-to-noise ratio value and structural similarity index value, which can reach 26.88 dB and 0.7740, respectively. We believe that this technology will improve the quality of FESEM imaging and be of significance in various application fields.