Yanbin Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yanbin Li has a total of 27 co-authors. Here are the top co-authors by number of shared publications:
- Wei Zhang
Beijing University of Posts and Telecommunications
5 shared publications - Zhiguo Zhang
School of Electronic Engineering, Beijing University of Posts and Telecommunications, Beijing, China
3 shared publications - Hongyang He
University of Warwick
2 shared publications - Kaige Qu
Beihang University
2 shared publications - Mingming Zhang
School of Integrated Circuit Science and Engineering, Beihang University, Beijing, China
2 shared publications - Qi-Jun Zhang
Carleton University
2 shared publications
Latest Publications
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Electronics
- Vol. 73 (10) , 14749-14761
DOA: A Degeneracy Optimization Agent With Adaptive Pose Compensation Capability Based on Deep Reinforcement Learning
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Electronics
- Vol. 73 (10) , 14749-14761
DOA: A Degeneracy Optimization Agent With Adaptive Pose Compensation Capability Based on Deep Reinforcement Learning
- 12 September 2026
- Published by Springer Nature
- p. 575-591
Geometric Regularization for Long-Tailed Semi-Supervised Learning via Gaussian Feature Bridges
- 12 September 2026
- Published by Springer Nature
- p. 575-591
Geometric Regularization for Long-Tailed Semi-Supervised Learning via Gaussian Feature Bridges
- Open Access
- 28 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 19 (19391404) , 6965-6980
SABR-YOLO: A Lightweight Detector for Power Transmission Tower Detection in Remote Sensing Image
The detection of transmission towers in remote sensing images presents significant challenges due to insufficient feature representation and significant object size variations, particularly at low spatial resolutions where robust feature extraction and fusion capabilities are essential. To address these limitations, we propose YOLO with Sobel attention and dynamic branch reweighting (SABR-YOLO), an efficient detector incorporating two innovative lightweight modules: the Sobel-enhanced convo lutional block attention module (SECBAM) for enhanced local perception through edge-aware feature learning, and the branch reweighting feature pyramid network (BRFPN) for improved multi-scale feature fusion with minimal computational overhead. To verify the effectiveness of SABR-YOLO, we construct a new dataset of transmission tower object detection (TTOD), which includes transmission towers of various types, shapes, and spatial resolutions. To evaluate its performance on other multi-scale object detection tasks, we performed experiments on the public datasets of remote sensing object detection (RSOD) and super tiny object detection for remote sensing (RS-STOD). The experi mental results show that SABR-YOLO achieves mAP50 scores of 0.965 on the RSOD dataset, 0.872 on the RS-STOD dataset, and 0.756 on TTOD dataset, outperforming several baseline models and state-of-the-art methods. The related code and TTOD are public available at https://github.com/twiilight/SABR-YOLO
- 28 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 19 (19391404) , 6965-6980
Open AccessSABR-YOLO: A Lightweight Detector for Power Transmission Tower Detection in Remote Sensing Image
The detection of transmission towers in remote sensing images presents significant challenges due to insufficient feature representation and significant object size variations, particularly at low spatial resolutions where robust feature extraction and fusion capabilities are essential. To address these limitations, we propose YOLO with Sobel attention and dynamic branch reweighting (SABR-YOLO), an efficient detector incorporating two innovative lightweight modules: the Sobel-enhanced convo lutional block attention module (SECBAM) for enhanced local perception through edge-aware feature learning, and the branch reweighting feature pyramid network (BRFPN) for improved multi-scale feature fusion with minimal computational overhead. To verify the effectiveness of SABR-YOLO, we construct a new dataset of transmission tower object detection (TTOD), which includes transmission towers of various types, shapes, and spatial resolutions. To evaluate its performance on other multi-scale object detection tasks, we performed experiments on the public datasets of remote sensing object detection (RSOD) and super tiny object detection for remote sensing (RS-STOD). The experi mental results show that SABR-YOLO achieves mAP50 scores of 0.965 on the RSOD dataset, 0.872 on the RS-STOD dataset, and 0.756 on TTOD dataset, outperforming several baseline models and state-of-the-art methods. The related code and TTOD are public available at https://github.com/twiilight/SABR-YOLO
- 19 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Microwave Theory and Techniques
- Vol. 74 (2) , 1423-1439
A Cross-Channel Parametric CNN Framework for Microwave Topology Modeling
Topology design has emerged as an efficient approach in the domain of electromagnetic (EM) design. Despite extensive research, accurately modeling high-performance topology structures remains a formidable challenge. This article proposes an advanced convolutional neural network (CNN) architecture, named cross-channel parametric small convolution network (CCP-SConvNet), tailored to microwave topology modeling. A CCP pooling mechanism combined with multiple small convolutional filters is developed to facilitate fine-grained spatial abstraction and adaptive interchannel information fusion. As a result, CCP-SConvNet improves predictive accuracy and robustness while reducing the overall parameter count compared to traditional CNNs. To ensure fabrication feasibility, a modified breadth-first search (BFS) algorithm is proposed to eliminate nonphysical structures such as suspended metal elements, thereby reducing the volume of training data required for model development. By integrating dynamic data generation with joint optimization strategies, the proposed framework further alleviates dependence on large datasets and accelerates the overall design process. CCP-SConvNet thus offers a rapid and cost-effective alternative to full-wave EM simulations, enabling efficient modeling and optimization of microwave topological components. The efficacy and robustness of the proposed CCP-SConvNet are validated through three representative cases: a single-pole waveguide filter operating at 11–13 GHz with a prediction error of 3.56%, a three-pole waveguide filter at 7.4–8.8GHz with an error of 2.82%, and a microstrip line filter at 1.2–2.8GHz with an error of 3.36%. The model reduces the required training data by up to 90% through the modified BFS strategy and accelerates EM simulation from minutes to milliseconds. The single-pole filter was fabricated and experimentally measured, exhibiting good agreement between the measured and predicted $S$ -parameters. These results demonstrate the superior capability of CCP-SConvNet in accurate, efficient, and physically validated modeling of advanced microwave topological components.
- 19 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Microwave Theory and Techniques
- Vol. 74 (2) , 1423-1439
A Cross-Channel Parametric CNN Framework for Microwave Topology Modeling
Topology design has emerged as an efficient approach in the domain of electromagnetic (EM) design. Despite extensive research, accurately modeling high-performance topology structures remains a formidable challenge. This article proposes an advanced convolutional neural network (CNN) architecture, named cross-channel parametric small convolution network (CCP-SConvNet), tailored to microwave topology modeling. A CCP pooling mechanism combined with multiple small convolutional filters is developed to facilitate fine-grained spatial abstraction and adaptive interchannel information fusion. As a result, CCP-SConvNet improves predictive accuracy and robustness while reducing the overall parameter count compared to traditional CNNs. To ensure fabrication feasibility, a modified breadth-first search (BFS) algorithm is proposed to eliminate nonphysical structures such as suspended metal elements, thereby reducing the volume of training data required for model development. By integrating dynamic data generation with joint optimization strategies, the proposed framework further alleviates dependence on large datasets and accelerates the overall design process. CCP-SConvNet thus offers a rapid and cost-effective alternative to full-wave EM simulations, enabling efficient modeling and optimization of microwave topological components. The efficacy and robustness of the proposed CCP-SConvNet are validated through three representative cases: a single-pole waveguide filter operating at 11–13 GHz with a prediction error of 3.56%, a three-pole waveguide filter at 7.4–8.8GHz with an error of 2.82%, and a microstrip line filter at 1.2–2.8GHz with an error of 3.36%. The model reduces the required training data by up to 90% through the modified BFS strategy and accelerates EM simulation from minutes to milliseconds. The single-pole filter was fabricated and experimentally measured, exhibiting good agreement between the measured and predicted $S$ -parameters. These results demonstrate the superior capability of CCP-SConvNet in accurate, efficient, and physically validated modeling of advanced microwave topological components.
- 9 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 10 (7) , 6784-6791
Anti-Degeneracy Scheme for Lidar SLAM Based on Particle Filter in Geometry Feature-Less Environments
Simultaneous localization and mapping (SLAM) based on particle filtering has been extensively employedin indoor scenarios due to its high efficiency. However, in geometry feature-less scenes, the accuracy is severely reduced due to lack of constraints. In this article, we propose an anti-degeneracy system based on deep learning. Firstly, we design a scale-invariant linear mapping to convert coordinates in continuous space into discrete indexes, in which a data augmentation method based on Gaussian model is proposed to ensure the model performance by effectively mitigating the impact of changes in the number of particles on the feature distribution. Secondly, we develop a degeneracy detection model (DD-Model) using residual neural networks (ResNet) and transformer which is able to identify degeneracy by scrutinizing the distribution of the particle population. Thirdly, an adaptive anti-degeneracy strategy is designed, which first performs fusion and perturbation on the resample process to provide rich and accurate initial values for the pose optimization, and use a hierarchical pose optimization combining coarse and fine matching, which is able to adaptively adjust the optimization frequency and the sensor trustworthiness according to the degree of degeneracy, in order to enhance the ability of searching the global optimal pose. Finally, we demonstrate the optimality of model, as well as the improvement of image matrix method and GPU on the inference time through ablation experiments. The effectiveness of our method is verified by comparing with SOTA methods.
- 9 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 10 (7) , 6784-6791
Anti-Degeneracy Scheme for Lidar SLAM Based on Particle Filter in Geometry Feature-Less Environments
Simultaneous localization and mapping (SLAM) based on particle filtering has been extensively employedin indoor scenarios due to its high efficiency. However, in geometry feature-less scenes, the accuracy is severely reduced due to lack of constraints. In this article, we propose an anti-degeneracy system based on deep learning. Firstly, we design a scale-invariant linear mapping to convert coordinates in continuous space into discrete indexes, in which a data augmentation method based on Gaussian model is proposed to ensure the model performance by effectively mitigating the impact of changes in the number of particles on the feature distribution. Secondly, we develop a degeneracy detection model (DD-Model) using residual neural networks (ResNet) and transformer which is able to identify degeneracy by scrutinizing the distribution of the particle population. Thirdly, an adaptive anti-degeneracy strategy is designed, which first performs fusion and perturbation on the resample process to provide rich and accurate initial values for the pose optimization, and use a hierarchical pose optimization combining coarse and fine matching, which is able to adaptively adjust the optimization frequency and the sensor trustworthiness according to the degree of degeneracy, in order to enhance the ability of searching the global optimal pose. Finally, we demonstrate the optimality of model, as well as the improvement of image matrix method and GPU on the inference time through ablation experiments. The effectiveness of our method is verified by comparing with SOTA methods.
- Open Access
- 30 April 2025
- Published by Elsevier in Biomimetic Intelligence and Robotics
- Vol. 5 (4)
An adaptive compensation strategy for sensors based on the degree of degradation
- 30 April 2025
- Published by Elsevier in Biomimetic Intelligence and Robotics
- Vol. 5 (4)
Open AccessAn adaptive compensation strategy for sensors based on the degree of degradation