Yuan Sun
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yuan Sun has a total of 27 co-authors. Here are the top co-authors by number of shared publications:
- Li Fu
JD AI Research, Beijing University of Posts and Telecommunications, Beijing, China
3 shared publications - Lu Yin
School of Electronic Engineering, Beijing University of Posts and Telecommunications, Beijing, China
3 shared publications - Jianqin Yin
School of Intelligent Engineering and Automation, Beijing University of Posts and Telecommunications, Beijing, China
2 shared publications - Li Fu
JD AI Research, Beijing, China
2 shared publications - Qiang Ni
Lancaster University
2 shared publications - Shang Li
School of Electronic Engineering, Beijing University of Posts and Telecommunications, Beijing, China
2 shared publications
Latest Publications
- 26 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75 (00189456) , 8508910
DUST: Discrete Unit Sequence Transformer for Ultrawideband Non-Line-of-Sight Identification
Ultra-Wideband (UWB) technology has emerged as a leading solution for indoor positioning; however, its accuracy is often compromised by non-line-of-sight (NLOS) propagation effects. To address the challenges of robust feature representation and model overfitting in NLOS identification, we propose DUST: a Discrete Unit Sequence Transformer framework featuring three key innovations. First, we introduce an adaptive quantization strategy that dynamically discretizes continuous Channel Impulse Response (CIR) signals using a kernel-based spatio-temporal analysis, effectively suppressing noise while preserving critical signal features. Second, we develop a dual-domain feature fusion mechanism that synergistically integrates time-domain and frequency-domain representations, thereby enhancing the extraction of discriminative patterns from complementary perspectives. Third, we design a hybrid architecture that combines Transformer-based contextual feature modeling with Support Vector Machine (SVM) classification, leveraging self-attention for richer context awareness and improved generalization. Extensive experiments conducted across diverse indoor scenarios demonstrate that DUST consistently outperforms conventional methods in NLOS identification, achieving up to a 14.4% absolute improvement in accuracy1.
- 26 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75 (00189456) , 8508910
DUST: Discrete Unit Sequence Transformer for Ultrawideband Non-Line-of-Sight Identification
Ultra-Wideband (UWB) technology has emerged as a leading solution for indoor positioning; however, its accuracy is often compromised by non-line-of-sight (NLOS) propagation effects. To address the challenges of robust feature representation and model overfitting in NLOS identification, we propose DUST: a Discrete Unit Sequence Transformer framework featuring three key innovations. First, we introduce an adaptive quantization strategy that dynamically discretizes continuous Channel Impulse Response (CIR) signals using a kernel-based spatio-temporal analysis, effectively suppressing noise while preserving critical signal features. Second, we develop a dual-domain feature fusion mechanism that synergistically integrates time-domain and frequency-domain representations, thereby enhancing the extraction of discriminative patterns from complementary perspectives. Third, we design a hybrid architecture that combines Transformer-based contextual feature modeling with Support Vector Machine (SVM) classification, leveraging self-attention for richer context awareness and improved generalization. Extensive experiments conducted across diverse indoor scenarios demonstrate that DUST consistently outperforms conventional methods in NLOS identification, achieving up to a 14.4% absolute improvement in accuracy1.
- 4 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75 (00189456) , 1-9
ATM: Adaptive Temporal Kalman Filter-Based Solution Separation Method for Positioning and Integrity Monitoring
The Kalman filter is widely used in integrated navigation frameworks combining the global navigation satellite system (GNSS) and inertial navigation system (INS). However, its assumption of Gaussian white noise is often violated in practice, as error sources such as satellite clock offsets and multipath effects show strong temporal correlations. This can affect both positioning accuracy and system integrity. To address this, we propose ATM: an adaptive temporal Kalman filter-based solution separation method for positioning and integrity monitoring. An adaptive temporal Kalman filter mainly consists of measurement noise modeling (MNM) and the combination of the state vector and the measurement (CSM) vector. ATM models colored noise as a first-order adaptive Markov process, capturing temporal dependencies. The Kalman filter update is reformulated as a weighted least-squares problem, enabling propagation to a new measurement vector and a refined measurement equation. This framework supports state solution separation in the positioning domain, improving fault detection and exclusion (FDE). Experiments on the public UrbanNav dataset show that the ATM greatly improves positioning accuracy and reduces protection levels compared to baseline methods in medium and deep urban canyons. Specifically, horizontal positioning accuracy improves by 4.80% and 4.57%, while the horizontal protection level (HPL) decreases by 13.32% and 18.32%, and the vertical protection level (VPL) decreases by 18.51% and 26.74%.
- 4 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75 (00189456) , 1-9
ATM: Adaptive Temporal Kalman Filter-Based Solution Separation Method for Positioning and Integrity Monitoring
The Kalman filter is widely used in integrated navigation frameworks combining the global navigation satellite system (GNSS) and inertial navigation system (INS). However, its assumption of Gaussian white noise is often violated in practice, as error sources such as satellite clock offsets and multipath effects show strong temporal correlations. This can affect both positioning accuracy and system integrity. To address this, we propose ATM: an adaptive temporal Kalman filter-based solution separation method for positioning and integrity monitoring. An adaptive temporal Kalman filter mainly consists of measurement noise modeling (MNM) and the combination of the state vector and the measurement (CSM) vector. ATM models colored noise as a first-order adaptive Markov process, capturing temporal dependencies. The Kalman filter update is reformulated as a weighted least-squares problem, enabling propagation to a new measurement vector and a refined measurement equation. This framework supports state solution separation in the positioning domain, improving fault detection and exclusion (FDE). Experiments on the public UrbanNav dataset show that the ATM greatly improves positioning accuracy and reduces protection levels compared to baseline methods in medium and deep urban canyons. Specifically, horizontal positioning accuracy improves by 4.80% and 4.57%, while the horizontal protection level (HPL) decreases by 13.32% and 18.32%, and the vertical protection level (VPL) decreases by 18.51% and 26.74%.
- Open Access
Boosting the Class-Incremental Learning in 3D Point Clouds via Zero-Collection-Cost Basic Shape Pre-Training
Existing class-incremental learning methods in 3D point clouds rely on exemplars (samples of former classes) to resist the catastrophic forgetting of models, and exemplar-free settings will greatly degrade the performance. For exemplar-free incremental learning, the pre-trained model methods have achieved state-of-the-art results in 2D domains. However, these methods cannot be migrated to the 3D domains due to the limited pre-training datasets and insufficient focus on fine-grained geometric details. This paper breaks through these limitations, proposing a basic shape dataset with zero collection cost for model pre-training. It helps a model obtain extensive knowledge of 3D geometries. Based on this, we propose a framework embedded with 3D geometry knowledge for incremental learning in point clouds, compatible with exemplar-free (-based) settings. In the incremental stage, the geometry knowledge is extended to represent objects in point clouds. The class prototype is calculated by regularizing the data representation with the same category and is kept adjusting in the learning process. It helps the model remember the shape features of different categories. Experiments show that our method outperforms other baseline methods by a large margin on various benchmark datasets, considering both exemplar-free (-based) settings.
Open AccessBoosting the Class-Incremental Learning in 3D Point Clouds via Zero-Collection-Cost Basic Shape Pre-Training
Existing class-incremental learning methods in 3D point clouds rely on exemplars (samples of former classes) to resist the catastrophic forgetting of models, and exemplar-free settings will greatly degrade the performance. For exemplar-free incremental learning, the pre-trained model methods have achieved state-of-the-art results in 2D domains. However, these methods cannot be migrated to the 3D domains due to the limited pre-training datasets and insufficient focus on fine-grained geometric details. This paper breaks through these limitations, proposing a basic shape dataset with zero collection cost for model pre-training. It helps a model obtain extensive knowledge of 3D geometries. Based on this, we propose a framework embedded with 3D geometry knowledge for incremental learning in point clouds, compatible with exemplar-free (-based) settings. In the incremental stage, the geometry knowledge is extended to represent objects in point clouds. The class prototype is calculated by regularizing the data representation with the same category and is kept adjusting in the learning process. It helps the model remember the shape features of different categories. Experiments show that our method outperforms other baseline methods by a large margin on various benchmark datasets, considering both exemplar-free (-based) settings.
- 7 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Wireless Communications Letters
- Vol. 14 (5) , 1561-1565
Novel Frequency-Efficient Communication-Positioning Integrated Signal and Ranging Methodology for Future Aviation
The future of aviation is characterized by the fusion of digital technologies, data-driven decision-making, and seamless connectivity. Accurate and reliable positioning plays a pivotal role in the future aviation technologies such as autonomous flight, advanced air traffic management (ATM) systems, and augmented reality (AR) for pilot assistance. In case of global navigation satellite system (GNSS) failure, the alternative positioning, navigation, and timing (APNT) system is needed. Current methods for aviation positioning either lack the accuracy, reliability or the capacity for future air traffic development because their signal designs are outdated or not optimal for positioning. To tackle this problem, a novel communication-positioning integrated signal is designed. A continuous positioning signal is superposed onto the L-band digital aeronautical communication system (L-DACS) signal using non-orthogonal multiple access (NOMA) principle to fully utilize the frequency resources and minimize its interference to communication. A novel cross channel measuring (CCM) algorithm is proposed to measure pseudorange when the positioning signal is spread across multiple L-DACS1 channels for improved ranging accuracy. Simulation results show the novel communication-positioning integrated signal and its measuring method outperforms the existing L-DACS1 method, almost achieving meter-level ranging accuracy. And the positioning signal causes minimal interference to communication.
- 7 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Wireless Communications Letters
- Vol. 14 (5) , 1561-1565
Novel Frequency-Efficient Communication-Positioning Integrated Signal and Ranging Methodology for Future Aviation
The future of aviation is characterized by the fusion of digital technologies, data-driven decision-making, and seamless connectivity. Accurate and reliable positioning plays a pivotal role in the future aviation technologies such as autonomous flight, advanced air traffic management (ATM) systems, and augmented reality (AR) for pilot assistance. In case of global navigation satellite system (GNSS) failure, the alternative positioning, navigation, and timing (APNT) system is needed. Current methods for aviation positioning either lack the accuracy, reliability or the capacity for future air traffic development because their signal designs are outdated or not optimal for positioning. To tackle this problem, a novel communication-positioning integrated signal is designed. A continuous positioning signal is superposed onto the L-band digital aeronautical communication system (L-DACS) signal using non-orthogonal multiple access (NOMA) principle to fully utilize the frequency resources and minimize its interference to communication. A novel cross channel measuring (CCM) algorithm is proposed to measure pseudorange when the positioning signal is spread across multiple L-DACS1 channels for improved ranging accuracy. Simulation results show the novel communication-positioning integrated signal and its measuring method outperforms the existing L-DACS1 method, almost achieving meter-level ranging accuracy. And the positioning signal causes minimal interference to communication.
- 4 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 63 (01962892) , 1-15
MCKTNet: Multiscale Cross-Modal Knowledge Transfer Network for Semantic Segmentation of Remote Sensing Images
Multimodal data fusion can provide valuable and diverse information for remote sensing image segmentation. However, different modal data have different feature distributions, which causes some conflicts and redundancies in cross-modal feature fusion. In addition, existing multimodal fusion networks usually adopt a two-branch structure with a large number of parameters and high computational cost. To address these problems, we propose a multi-scale cross-modal knowledge transfer network for remote sensing image segmentation. First, we use a cross-modal migration learning method that combines channel discrete loss and spatial discrete loss to facilitate cross-modal migration of geometric and semantic features and reduces redundancy by minimizing the differences in feature distribution. Then, we use the polarized cross-self-attention mechanism to establish long-range correlations between different modal features across spatial and channel dimensions, and only a small number of parameters need to be added to achieve complementary cross-modal feature fusion. Finally, to accurately capture object edges, we propose a multi-scale edge perception module to optimize the edge details in the prediction results at the pixel level. Extensive experiments demonstrate that the proposed method is effective, robust, and generalizable, and achieves state-of-the-art performance in multiple remote sensing image semantic segmentation tasks with only 13.51 million parameters. The code will be available at https://github.com/NUAALISILab/MCKTNet.
- 4 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 63 (01962892) , 1-15
MCKTNet: Multiscale Cross-Modal Knowledge Transfer Network for Semantic Segmentation of Remote Sensing Images
Multimodal data fusion can provide valuable and diverse information for remote sensing image segmentation. However, different modal data have different feature distributions, which causes some conflicts and redundancies in cross-modal feature fusion. In addition, existing multimodal fusion networks usually adopt a two-branch structure with a large number of parameters and high computational cost. To address these problems, we propose a multi-scale cross-modal knowledge transfer network for remote sensing image segmentation. First, we use a cross-modal migration learning method that combines channel discrete loss and spatial discrete loss to facilitate cross-modal migration of geometric and semantic features and reduces redundancy by minimizing the differences in feature distribution. Then, we use the polarized cross-self-attention mechanism to establish long-range correlations between different modal features across spatial and channel dimensions, and only a small number of parameters need to be added to achieve complementary cross-modal feature fusion. Finally, to accurately capture object edges, we propose a multi-scale edge perception module to optimize the edge details in the prediction results at the pixel level. Extensive experiments demonstrate that the proposed method is effective, robust, and generalizable, and achieves state-of-the-art performance in multiple remote sensing image semantic segmentation tasks with only 13.51 million parameters. The code will be available at https://github.com/NUAALISILab/MCKTNet.
- 17 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 74 (2) , 2480-2490
NICL: Non-Line-of-Sight Identification in Global Navigation Satellite Systems With Continual Learning
The widespread adoption of Global Navigation Satellite Systems (GNSS) has revolutionized high-precision positioning services. However, the accuracy of these systems is significantly compromised by Non-Line-of-Sight (NLOS) signals, especially in urban canyons. Specifically, Line-of-Sight (LOS) signals are obstructed, and the reflections lead to substantial errors. Machine learning has emerged as a promising tool for identifying NLOS signals. However, traditional methods often train models on extensive datasets from various environments and fine-tune them with new data to adjust to new scenarios. This fine-tuning can result in catastrophic forgetting, which impairs the models' generalization performance across different environments. To address the limitations of model fine-tuning, we introduce NICL, an approach for N LOS I dentification with C ontinual L earning. NICL preserves performance on previously encountered scenarios while adapting to new environments by employing simple regularization techniques that do not require access to the data from the old scenarios. Specifically, our method incorporates Kullback-Leibler divergence regularization and explainability-based regularization to maintain the model's prior outputs and the logic behind those outputs, thus mitigating catastrophic forgetting. Experimental results on real-world GNSS datasets have shown that our NICL method outperforms existing baseline methods, achieving an absolute improvement of 5%-12% in NLOS identification accuracy.
- 17 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 74 (2) , 2480-2490
NICL: Non-Line-of-Sight Identification in Global Navigation Satellite Systems With Continual Learning
The widespread adoption of Global Navigation Satellite Systems (GNSS) has revolutionized high-precision positioning services. However, the accuracy of these systems is significantly compromised by Non-Line-of-Sight (NLOS) signals, especially in urban canyons. Specifically, Line-of-Sight (LOS) signals are obstructed, and the reflections lead to substantial errors. Machine learning has emerged as a promising tool for identifying NLOS signals. However, traditional methods often train models on extensive datasets from various environments and fine-tune them with new data to adjust to new scenarios. This fine-tuning can result in catastrophic forgetting, which impairs the models' generalization performance across different environments. To address the limitations of model fine-tuning, we introduce NICL, an approach for N LOS I dentification with C ontinual L earning. NICL preserves performance on previously encountered scenarios while adapting to new environments by employing simple regularization techniques that do not require access to the data from the old scenarios. Specifically, our method incorporates Kullback-Leibler divergence regularization and explainability-based regularization to maintain the model's prior outputs and the logic behind those outputs, thus mitigating catastrophic forgetting. Experimental results on real-world GNSS datasets have shown that our NICL method outperforms existing baseline methods, achieving an absolute improvement of 5%-12% in NLOS identification accuracy.
- 9 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Selected Areas in Communications
- Vol. 42 (1) , 223-238
New Signal and Algorithms for 5G/6G High Precision Train Positioning in Tunnel With Leaky Coaxial Cable
High precision train positioning is a crucial component of intelligent transportation systems. Tunnels are commonly encountered in subways and mountainous regions. As part of the communication system infrastructure, Leaky CoaXial (LCX) Cable is widely equipped as antenna in tunnels with many advantages. LCX positioning holds great promise as a technology for rail applications in the upcoming B5G (beyond-5G) and 6G eras. This paper focuses on the LCX positioning methodology and proposes two novel algorithms along with a novel communication-positioning integration signal. Firstly, a novel algorithm called Multiple Slot Distinction (MSD) LCX positioning algorithm is proposed. The algorithm utilizes a generated pseudo spectrum to fully utilize the coupled signals radiated from different slots of LCX. This approach offers higher time resolution compared to traditional methods. To further improve the positioning accuracy to centimeter-level and increase the measuring frequency for fast trains, a novel communication-positioning integration signal is designed. It consists of traditional Positioning Reference Signal (PRS) and a significantly low power Fine Ranging Signal (FRS). FRS is configured to be continuous and superposed onto the cellular signal using Non-Orthogonal Multiple Access (NOMA) principle to minimize its interference to communication. A two-stage LCX positioning method is then executed: At the first stage, the closest slot between the receiver and LCX is estimated by the proposed MSD algorithm using PRS; At the second stage, centimeter-level positioning is achieved by tracking the carrier phase of the continuous FRS. This process is assisted by the closest slot estimation, which helps mitigate interference between neighboring slots and eliminate the integer ambiguities. Simulation results show our proposed LCX position methodology outperforms the existing ones and offer great potentials for future implementations.
- 9 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Selected Areas in Communications
- Vol. 42 (1) , 223-238
New Signal and Algorithms for 5G/6G High Precision Train Positioning in Tunnel With Leaky Coaxial Cable
High precision train positioning is a crucial component of intelligent transportation systems. Tunnels are commonly encountered in subways and mountainous regions. As part of the communication system infrastructure, Leaky CoaXial (LCX) Cable is widely equipped as antenna in tunnels with many advantages. LCX positioning holds great promise as a technology for rail applications in the upcoming B5G (beyond-5G) and 6G eras. This paper focuses on the LCX positioning methodology and proposes two novel algorithms along with a novel communication-positioning integration signal. Firstly, a novel algorithm called Multiple Slot Distinction (MSD) LCX positioning algorithm is proposed. The algorithm utilizes a generated pseudo spectrum to fully utilize the coupled signals radiated from different slots of LCX. This approach offers higher time resolution compared to traditional methods. To further improve the positioning accuracy to centimeter-level and increase the measuring frequency for fast trains, a novel communication-positioning integration signal is designed. It consists of traditional Positioning Reference Signal (PRS) and a significantly low power Fine Ranging Signal (FRS). FRS is configured to be continuous and superposed onto the cellular signal using Non-Orthogonal Multiple Access (NOMA) principle to minimize its interference to communication. A two-stage LCX positioning method is then executed: At the first stage, the closest slot between the receiver and LCX is estimated by the proposed MSD algorithm using PRS; At the second stage, centimeter-level positioning is achieved by tracking the carrier phase of the continuous FRS. This process is assisted by the closest slot estimation, which helps mitigate interference between neighboring slots and eliminate the integer ambiguities. Simulation results show our proposed LCX position methodology outperforms the existing ones and offer great potentials for future implementations.
- 24 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 72 (00189456) , 1-13
G5GIM: Integrity Monitoring for GNSS/5G Integrated Navigation of Urban Vehicles
Integrity Monitoring (IM) for navigation systems is critical to ensure the safety of auto-driving vehicles. It refers to accurately raising alarms to a user when navigation positioning errors exceed the performance limitation. Although Receiver Autonomous Integrity Monitoring (RAIM) methods have been widely applied for Global Navigation Satellite System (GNSS) observations via a consistency checking algorithm, the integrity performance for auto-driving would suffer from a large degradation due to the low satellite visibility with signal occlusions and Non-Line-of-Sight (NLOS) receptions in urban environments. To alleviate this problem, the Fifth-generation (5G) communication signals are used as an auxiliary urban navigation source to increase the number of observations, and then an IM method for the integration of GNSS/5G (G5GIM) is proposed to improve the integrity performance. Specifically, the positioning reference signals of the 5G system in urban canyons are modeled to enrich the observations of the vehicle navigation, followed by the Horizontal Protection Level (HPL) estimation method and the fault detection algorithm for the integration of GNSS and 5G observations. Experimental results on the simulation and UrbanNav datasets show that the proposed G5GIM method significantly outperforms the existing GNSS RAIM method in terms of HPL reduction, fault detection rate and positioning accuracy improvement.
- 24 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 72 (00189456) , 1-13
G5GIM: Integrity Monitoring for GNSS/5G Integrated Navigation of Urban Vehicles
Integrity Monitoring (IM) for navigation systems is critical to ensure the safety of auto-driving vehicles. It refers to accurately raising alarms to a user when navigation positioning errors exceed the performance limitation. Although Receiver Autonomous Integrity Monitoring (RAIM) methods have been widely applied for Global Navigation Satellite System (GNSS) observations via a consistency checking algorithm, the integrity performance for auto-driving would suffer from a large degradation due to the low satellite visibility with signal occlusions and Non-Line-of-Sight (NLOS) receptions in urban environments. To alleviate this problem, the Fifth-generation (5G) communication signals are used as an auxiliary urban navigation source to increase the number of observations, and then an IM method for the integration of GNSS/5G (G5GIM) is proposed to improve the integrity performance. Specifically, the positioning reference signals of the 5G system in urban canyons are modeled to enrich the observations of the vehicle navigation, followed by the Horizontal Protection Level (HPL) estimation method and the fault detection algorithm for the integration of GNSS and 5G observations. Experimental results on the simulation and UrbanNav datasets show that the proposed G5GIM method significantly outperforms the existing GNSS RAIM method in terms of HPL reduction, fault detection rate and positioning accuracy improvement.
- Open Access
An end-to-end multi-scale network for action prediction in videos
In this paper, we develop an efficient multi-scale network to predict action classes in partial videos in an end-to-end manner. Unlike most existing methods with offline feature generation, our method directly takes frames as input and further models motion evolution on two different temporal scales.Therefore, we solve the complexity problems of the two stages of modeling and the problem of insufficient temporal and spatial information of a single scale. Our proposed End-to-End MultiScale Network (E2EMSNet) is composed of two scales which are named segment scale and observed global scale. The segment scale leverages temporal difference over consecutive frames for finer motion patterns by supplying 2D convolutions. For observed global scale, a Long Short-Term Memory (LSTM) is incorporated to capture motion features of observed frames. Our model provides a simple and efficient modeling framework with a small computational cost. Our E2EMSNet is evaluated on three challenging datasets: BIT, HMDB51, and UCF101. The extensive experiments demonstrate the effectiveness of our method for action prediction in videos.
Open AccessAn end-to-end multi-scale network for action prediction in videos
In this paper, we develop an efficient multi-scale network to predict action classes in partial videos in an end-to-end manner. Unlike most existing methods with offline feature generation, our method directly takes frames as input and further models motion evolution on two different temporal scales.Therefore, we solve the complexity problems of the two stages of modeling and the problem of insufficient temporal and spatial information of a single scale. Our proposed End-to-End MultiScale Network (E2EMSNet) is composed of two scales which are named segment scale and observed global scale. The segment scale leverages temporal difference over consecutive frames for finer motion patterns by supplying 2D convolutions. For observed global scale, a Long Short-Term Memory (LSTM) is incorporated to capture motion features of observed frames. Our model provides a simple and efficient modeling framework with a small computational cost. Our E2EMSNet is evaluated on three challenging datasets: BIT, HMDB51, and UCF101. The extensive experiments demonstrate the effectiveness of our method for action prediction in videos.
- Open Access
- 28 April 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 71 (00189456) , 1-10
Stacking Ensemble Learning for Non-Line-of-Sight Detection of Global Navigation Satellite System
While the global navigation satellite system (GNSS) has been widely used to provide high-precision location services in many applications, it usually suffers from performance degradation due to non-line-of-sight (NLOS) reception. As the received NLOS signals might have great measurement errors especially in urban canyons, they should be detected to mitigate the errors contaminating the positioning systems. However, the NLOS detection is quite challenging as the accuracy rate is usually highly related to the surrounding environment the receiver is located in. To address this problem, we propose a stacking ensemble learning (SEL) method for the NLOS detection of GNSS. First, satellite measurement features are extracted from the GNSS raw measurements via a designed data processing module. Then, they are input to the SEL module consisting of two levels of machine learning models. In the first level, a support vector machine (SVM) and an extreme gradient boosting (XGBoost) are adopted in parallel, and the outputs of the fist-level models are input to the second-level logistic regression (LR) to obtain NLOS predictions. The proposed SEL module combines the views of different models to the measurement features to address the shortcomings of each single model and improve the model’s generalization. Experimental results on real GNSS observations in urban canyons show that the proposed method outperforms the baseline machine learning methods with obvious detection accuracy improvements.
- 28 April 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 71 (00189456) , 1-10
Open AccessStacking Ensemble Learning for Non-Line-of-Sight Detection of Global Navigation Satellite System
While the global navigation satellite system (GNSS) has been widely used to provide high-precision location services in many applications, it usually suffers from performance degradation due to non-line-of-sight (NLOS) reception. As the received NLOS signals might have great measurement errors especially in urban canyons, they should be detected to mitigate the errors contaminating the positioning systems. However, the NLOS detection is quite challenging as the accuracy rate is usually highly related to the surrounding environment the receiver is located in. To address this problem, we propose a stacking ensemble learning (SEL) method for the NLOS detection of GNSS. First, satellite measurement features are extracted from the GNSS raw measurements via a designed data processing module. Then, they are input to the SEL module consisting of two levels of machine learning models. In the first level, a support vector machine (SVM) and an extreme gradient boosting (XGBoost) are adopted in parallel, and the outputs of the fist-level models are input to the second-level logistic regression (LR) to obtain NLOS predictions. The proposed SEL module combines the views of different models to the measurement features to address the shortcomings of each single model and improve the model’s generalization. Experimental results on real GNSS observations in urban canyons show that the proposed method outperforms the baseline machine learning methods with obvious detection accuracy improvements.