Yunfei Guo
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Yunfei Guo has a total of 113 co-authors. Here are the top co-authors by number of shared publications:
- Yun Chen
Hangzhou Dianzi University
9 shared publications - Anke Xue
Hangzhou Dianzi University, Hangzhou, China
5 shared publications Yun Chen
5 shared publicationsAnke Xue
School of Automation, Hangzhou Dianzi University, Hangzhou, China
4 shared publications- Dongsheng Yang
School of Automation, Hangzhou Dianzi University, Hangzhou, China
4 shared publications Han Shen-Tu
4 shared publications
Latest Publications
- 4 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. PP (01628828) , 1-16
ThermalGaussian++: Improving Alignment and Resolution for ThermalGaussian
Thermography is especially valuable for the military and other users of surveillance cameras. Some recent methods based on Neural Radiance Fields (NeRF) have been proposed to reconstruct thermal scenes in 3D from a set of thermal and RGB images. However, unlike NeRF, 3D Gaussian splatting (3DGS) prevails due to its rapid training and real-time rendering. In this work, we propose ThermalGaussian, the first thermal 3DGS approach capable of rendering high-quality images in RGB and thermal modalities. We first calibrate the RGB camera and the thermal camera to ensure that both modalities are accurately aligned. Subsequently, we use the registered images to learn the multimodal 3D Gaussians. To prevent the overfitting of any single modality, we introduce several multimodal regularization constraints. We also develop smoothing constraints tailored to the physical characteristics of the thermal modality. Besides, we contribute a real-world dataset named RGBT-Scenes, captured by a handheld thermal-infrared camera, facilitating future research on thermal scene reconstruction. Based on ThermalGaussian, we further introduce ThermalGaussian++ to improve the alignment and resolution of ThermalGaussian. To improve multimodal alignment, we design a multimodal pose optimization module. This module enables direct processing of non-aligned multimodal image pairs, reducing the need for professional calibration before each use. To improve thermal resolution, we also propose a multimodal joint super-resolution reconstruction module, which enhances the quality of low-resolution thermal fields. Additionally, we contribute a new dataset: RGBT-Scenes++, which offers higher-resolution thermal images. We conduct comprehensive experiments demonstrating that ThermalGaussian++ achieves photorealistic thermal rendering and improves RGB rendering quality. It significantly enhances both alignment and resolution, enabling better practical deployment. In addition, our multimodal regularization constraints reduce the model's storage requirements. The code and datasets will be released.
- 4 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. PP (01628828) , 1-16
ThermalGaussian++: Improving Alignment and Resolution for ThermalGaussian
Thermography is especially valuable for the military and other users of surveillance cameras. Some recent methods based on Neural Radiance Fields (NeRF) have been proposed to reconstruct thermal scenes in 3D from a set of thermal and RGB images. However, unlike NeRF, 3D Gaussian splatting (3DGS) prevails due to its rapid training and real-time rendering. In this work, we propose ThermalGaussian, the first thermal 3DGS approach capable of rendering high-quality images in RGB and thermal modalities. We first calibrate the RGB camera and the thermal camera to ensure that both modalities are accurately aligned. Subsequently, we use the registered images to learn the multimodal 3D Gaussians. To prevent the overfitting of any single modality, we introduce several multimodal regularization constraints. We also develop smoothing constraints tailored to the physical characteristics of the thermal modality. Besides, we contribute a real-world dataset named RGBT-Scenes, captured by a handheld thermal-infrared camera, facilitating future research on thermal scene reconstruction. Based on ThermalGaussian, we further introduce ThermalGaussian++ to improve the alignment and resolution of ThermalGaussian. To improve multimodal alignment, we design a multimodal pose optimization module. This module enables direct processing of non-aligned multimodal image pairs, reducing the need for professional calibration before each use. To improve thermal resolution, we also propose a multimodal joint super-resolution reconstruction module, which enhances the quality of low-resolution thermal fields. Additionally, we contribute a new dataset: RGBT-Scenes++, which offers higher-resolution thermal images. We conduct comprehensive experiments demonstrating that ThermalGaussian++ achieves photorealistic thermal rendering and improves RGB rendering quality. It significantly enhances both alignment and resolution, enabling better practical deployment. In addition, our multimodal regularization constraints reduce the model's storage requirements. The code and datasets will be released.
- 29 March 2026
- Published by Elsevier in Digital Signal Processing
- Vol. 177
Gaussian belief propagation with information-domain measurement refinement for multi-sensor multi-target tracking
- 29 March 2026
- Published by Elsevier in Digital Signal Processing
- Vol. 177
Gaussian belief propagation with information-domain measurement refinement for multi-sensor multi-target tracking
- Open Access
- 7 January 2026
- Published by MDPI AG in Algorithms
- Vol. 19 (1) , 56
Unified Complementary Learning with Feature Perturbation for Semi-Supervised Learning
Semi-supervised learning has attracted widespread attention due to its ability to utilize both labeled and unlabeled data, leading to significant progress in recent years. Conventional semi-supervised learning approaches often rely on a strategy that combines weak and strong image-level augmentations and employs pseudo-labeling techniques, where high-confidence predictions are selected as pseudo-labels while low-confidence ones are discarded. However, such methods tend to overlook the useful information contained in low-confidence samples. Moreover, existing augmentation strategies are mostly limited to the image level and lack feature-level perturbations. To address these limitations, this paper proposes a semi-supervised learning method that integrates complementary labels and auxiliary feature perturbations, aiming to extract valuable information from low-confidence samples and expand the scope of feature-space perturbations. Experiments on the standard CIFAR-10 dataset show that under the extreme setting of only 4 labels per class, our method improves accuracy by 2.58% compared to Fixmatch. On the more complex STL-10 dataset, also with only 4 labels per class, the Top-1 accuracy is improved by 2.59%. In addition, systematic ablation studies are conducted to verify the effectiveness of each component.
- 7 January 2026
- Published by MDPI AG in Algorithms
- Vol. 19 (1) , 56
Open AccessUnified Complementary Learning with Feature Perturbation for Semi-Supervised Learning
Semi-supervised learning has attracted widespread attention due to its ability to utilize both labeled and unlabeled data, leading to significant progress in recent years. Conventional semi-supervised learning approaches often rely on a strategy that combines weak and strong image-level augmentations and employs pseudo-labeling techniques, where high-confidence predictions are selected as pseudo-labels while low-confidence ones are discarded. However, such methods tend to overlook the useful information contained in low-confidence samples. Moreover, existing augmentation strategies are mostly limited to the image level and lack feature-level perturbations. To address these limitations, this paper proposes a semi-supervised learning method that integrates complementary labels and auxiliary feature perturbations, aiming to extract valuable information from low-confidence samples and expand the scope of feature-space perturbations. Experiments on the standard CIFAR-10 dataset show that under the extreme setting of only 4 labels per class, our method improves accuracy by 2.58% compared to Fixmatch. On the more complex STL-10 dataset, also with only 4 labels per class, the Top-1 accuracy is improved by 2.59%. In addition, systematic ablation studies are conducted to verify the effectiveness of each component.
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 164112-164120
Robust Range-Only Tracking via Angle-Parameterized Maximum Correntropy Square-Root Cubature Kalman Filtering
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 164112-164120
Open AccessRobust Range-Only Tracking via Angle-Parameterized Maximum Correntropy Square-Root Cubature Kalman Filtering
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Oceanic Engineering
- Vol. 51 (1) , 723-733
Target Motion Analysis Using Labeled Multi-Bernoulli Filter With Twofold Gaussian Mixture
To enhance real-time passive multitarget tracking performance in clutter with a single sensor, a twofold Gaussian mixture model (GMM) labeled multi-Bernoulli (TGMM-LMB) filter is proposed in this article. In the proposed algorithm, a TGMM is designed for track initialization. One GMM is employed to initialize target position with bearing measurements, and the other is used for velocity initialization with Doppler measurements. Using TGMM, the estimated covariance is reduced, and hence the tracking error and the track confirmation time are decreased. Furthermore, a one-step ranked assignment strategy is proposed to obtain the first-order approximation of the updated LMB density, and the linear Gibbs sampling algorithm is used to implement the ranked assignment with linear complexity in the number of labels. This strategy eliminates the density transformation between the LMB density and the $\delta$-generalized LMB density, thereby improving real-time performance. The effectiveness of the proposed algorithm is verified in simulation.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Oceanic Engineering
- Vol. 51 (1) , 723-733
Target Motion Analysis Using Labeled Multi-Bernoulli Filter With Twofold Gaussian Mixture
To enhance real-time passive multitarget tracking performance in clutter with a single sensor, a twofold Gaussian mixture model (GMM) labeled multi-Bernoulli (TGMM-LMB) filter is proposed in this article. In the proposed algorithm, a TGMM is designed for track initialization. One GMM is employed to initialize target position with bearing measurements, and the other is used for velocity initialization with Doppler measurements. Using TGMM, the estimated covariance is reduced, and hence the tracking error and the track confirmation time are decreased. Furthermore, a one-step ranked assignment strategy is proposed to obtain the first-order approximation of the updated LMB density, and the linear Gibbs sampling algorithm is used to implement the ranked assignment with linear complexity in the number of labels. This strategy eliminates the density transformation between the LMB density and the $\delta$-generalized LMB density, thereby improving real-time performance. The effectiveness of the proposed algorithm is verified in simulation.
- 30 October 2025
- Published by Zhejiang University Press in Frontiers of Information Technology & Electronic Engineering
- Vol. 26 (10) , 2016-2029
Distributed kernel mean embedding Gaussian belief propagation for underwater multi-sensor multi-target passive tracking
To address the problem of underwater multi-sensor multi-target passive tracking in clutter, a distributed kernel mean embedding-based Gaussian belief propa
- 30 October 2025
- Published by Zhejiang University Press in Frontiers of Information Technology & Electronic Engineering
- Vol. 26 (10) , 2016-2029
Distributed kernel mean embedding Gaussian belief propagation for underwater multi-sensor multi-target passive tracking
To address the problem of underwater multi-sensor multi-target passive tracking in clutter, a distributed kernel mean embedding-based Gaussian belief propa
- 23 September 2025
- Published by Elsevier in Digital Signal Processing
- Vol. 168
An adaptive virtual measurement STGP algorithm for self-occluded 3D extended object tracking with time-varying shapes
- 23 September 2025
- Published by Elsevier in Digital Signal Processing
- Vol. 168
An adaptive virtual measurement STGP algorithm for self-occluded 3D extended object tracking with time-varying shapes
- 22 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Signal and Information Processing over Networks
- Vol. 11, 1100-1111
Variance-Constrained Distributed Filtering Under Limited Bit Rates for Time-Varying Systems
This article concentrates on the varianceconstrained distributed filtering problem with the constraint of limited bit rates and imperfect measurements for nonlinear time-varying systems. The measurement outputs undergo the phenomena of sensor saturations and nonlinearities occurring in a random way. An encoding-decoding mechanism (EDM) is implemented to regulate the transmission procedures over shared communication network. The main purpose of this article is to formulate a suitable distributed filtering algorithm to enable the fulfillment of both stochastic H∞ performance and variance constraint for the resultant filtering error system over a finite horizon. The sufficient conditions are initially established to satisfy the prescribed performance constraints, following which the proper filter parameters are derived by means of the solutions to a sequence of iterative matrix inequalities. Furthermore, based on the variance constraint analysis for filtering error, the genetic algorithm (GA) is utilized to optimize the bit rate allocation among every node by minimizing the value of triggered decoding error. Finally, the validity of the proposed distributed filtering scheme is testified by a numerical example.
- 22 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Signal and Information Processing over Networks
- Vol. 11, 1100-1111
Variance-Constrained Distributed Filtering Under Limited Bit Rates for Time-Varying Systems
This article concentrates on the varianceconstrained distributed filtering problem with the constraint of limited bit rates and imperfect measurements for nonlinear time-varying systems. The measurement outputs undergo the phenomena of sensor saturations and nonlinearities occurring in a random way. An encoding-decoding mechanism (EDM) is implemented to regulate the transmission procedures over shared communication network. The main purpose of this article is to formulate a suitable distributed filtering algorithm to enable the fulfillment of both stochastic H∞ performance and variance constraint for the resultant filtering error system over a finite horizon. The sufficient conditions are initially established to satisfy the prescribed performance constraints, following which the proper filter parameters are derived by means of the solutions to a sequence of iterative matrix inequalities. Furthermore, based on the variance constraint analysis for filtering error, the genetic algorithm (GA) is utilized to optimize the bit rate allocation among every node by minimizing the value of triggered decoding error. Finally, the validity of the proposed distributed filtering scheme is testified by a numerical example.
- 15 July 2025
- Published by Elsevier in Digital Signal Processing
- Vol. 168
Kernel Gaussian processes based extended target tracking in polar coordinate
- 15 July 2025
- Published by Elsevier in Digital Signal Processing
- Vol. 168
Kernel Gaussian processes based extended target tracking in polar coordinate
- 14 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (16) , 31208-31218
Multisource Decision Fusion for Multiview Extended Target Tracking With Self-Occlusion
To address the problem of multi-view extended target tracking with self-occlusion, a Dempster–Shafer evidence theory with generalized covariance intersection (DS-GCI) multi-source decision fusion is proposed. The proposed DS-GCI algorithm develops a multi-view fusion framework for extended target tracking under conditions where sensor observations are asynchronous and local estimates are inconsistent due to target self-occlusion. First, a confidence evaluation strategy based on the observation angle interval and Gaussian kernel similarity is employed to quantify the reliability and consistency of local estimates. The resulting quantified measures are then fused to construct the initial basic probability assignment (BPA) within the Dempster–Shafer evidence theory framework. Second, the belief Jensen–Shannon divergence and Deng entropy are jointly used to characterize conflicts and information among local estimates, and are further employed to adjust the initial BPA. If the difference between the adjusted and initial BPAs exceeds a certain threshold, the adjusted BPA is used as the dynamic fusion weights. Otherwise, the weights are derived using Dempster’s combination rule. Finally, the global estimate is computed using the generalized covariance intersection with these dynamic weights. Simulation and experiment demonstrate that the proposed algorithm effectively improves the accuracy of multi-view extended target tracking, when local estimates are inconsistent due to sensor viewpoint differences and target self-occlusion.
- 14 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (16) , 31208-31218
Multisource Decision Fusion for Multiview Extended Target Tracking With Self-Occlusion
To address the problem of multi-view extended target tracking with self-occlusion, a Dempster–Shafer evidence theory with generalized covariance intersection (DS-GCI) multi-source decision fusion is proposed. The proposed DS-GCI algorithm develops a multi-view fusion framework for extended target tracking under conditions where sensor observations are asynchronous and local estimates are inconsistent due to target self-occlusion. First, a confidence evaluation strategy based on the observation angle interval and Gaussian kernel similarity is employed to quantify the reliability and consistency of local estimates. The resulting quantified measures are then fused to construct the initial basic probability assignment (BPA) within the Dempster–Shafer evidence theory framework. Second, the belief Jensen–Shannon divergence and Deng entropy are jointly used to characterize conflicts and information among local estimates, and are further employed to adjust the initial BPA. If the difference between the adjusted and initial BPAs exceeds a certain threshold, the adjusted BPA is used as the dynamic fusion weights. Otherwise, the weights are derived using Dempster’s combination rule. Finally, the global estimate is computed using the generalized covariance intersection with these dynamic weights. Simulation and experiment demonstrate that the proposed algorithm effectively improves the accuracy of multi-view extended target tracking, when local estimates are inconsistent due to sensor viewpoint differences and target self-occlusion.