Junchen Ye
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Junchen Ye has a total of 61 co-authors. Here are the top co-authors by number of shared publications:
- Bowen Du
Beihang University
20 shared publications - Bohao Li
The State Key Laboratory of Complex and Critical Software Environment, Beihang University, Beijing, China
6 shared publications - Leilei Sun
Beihang University
6 shared publications - Tao Zou
State Key Laboratory of Complex and Critical Software Environment, Beihang University, Beijing, China
6 shared publications - Ke Cheng
CCSE Lab, Beihang University, Beijing, China
5 shared publications Bowen Du
4 shared publications
Latest Publications
- 1 December 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Knowledge and Data Engineering
- Vol. 38 (12) , 8264-8278
Hierarchical Transformer for Long-Term Traffic Flow Forecasting
- 1 December 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Knowledge and Data Engineering
- Vol. 38 (12) , 8264-8278
Hierarchical Transformer for Long-Term Traffic Flow Forecasting
- Open Access
- 3 September 2026
- Published by MDPI AG in Remote Sensing
- Vol. 18 (17) , 2983
Boot-Shaped Terrain Screening and Deep Learning Semantic Segmentation for Landslide-Hazard Candidate Extraction from Airborne LiDAR DEM: A Case Study in Zhenxiong County, China
Automated screening of geomorphologically defined landslide potential-hazard candidates from high-resolution topographic data remains challenging in mountainous regions where comprehensive field inventories are unavailable. This study proposes a two-stage framework for extracting rule-defined boot-shaped terrain candidates from airborne LiDAR digital elevation model (DEM) data. First, an expert-informed screening rule formalizes a steep-upper–gentle-lower terrain morphology using representative longitudinal profiles of slope units, and the screened units are converted into rule-derived reference masks. Second, semantic segmentation models are trained to approximate these reference patterns directly from DEM-derived raster inputs. Four architectures—U-Net, U-Net++, DeepLabV3+, and SegFormer-B0—were evaluated using 406 patches of 256 × 256 pixels at 2 m resolution from four LiDAR-covered subregions in Zhenxiong County, China. Under spatially grouped three-fold cross-validation, DeepLabV3+ with DEM + slope-gradient input and Dice + Focal loss achieved a mean pixel-level F1-score of 0.351, mIoU of 0.557, and Patch-F1 of 0.814. Input-feature experiments showed that slope-gradient information was particularly informative, whereas the incremental contribution of aspect was configuration-dependent; DEM + slope was retained as a parsimonious two-channel input. Sensitivity analysis showed that the rule-derived candidate definition changed materially with the screening parameters. Leave-one-subregion-out evaluation yielded a macro-averaged F1 of 0.321, indicating measurable within-county cross-subregion transfer. However, whole-area evaluation under natural candidate prevalence reduced the macro-average F1 to 0.072 at a fixed threshold and 0.095 using validation-derived operating thresholds. These results indicate that the proposed model is best interpreted as a raster-based surrogate for rule-derived geomorphological screening rather than as an independently validated landslide detector.
- 3 September 2026
- Published by MDPI AG in Remote Sensing
- Vol. 18 (17) , 2983
Open AccessBoot-Shaped Terrain Screening and Deep Learning Semantic Segmentation for Landslide-Hazard Candidate Extraction from Airborne LiDAR DEM: A Case Study in Zhenxiong County, China
Automated screening of geomorphologically defined landslide potential-hazard candidates from high-resolution topographic data remains challenging in mountainous regions where comprehensive field inventories are unavailable. This study proposes a two-stage framework for extracting rule-defined boot-shaped terrain candidates from airborne LiDAR digital elevation model (DEM) data. First, an expert-informed screening rule formalizes a steep-upper–gentle-lower terrain morphology using representative longitudinal profiles of slope units, and the screened units are converted into rule-derived reference masks. Second, semantic segmentation models are trained to approximate these reference patterns directly from DEM-derived raster inputs. Four architectures—U-Net, U-Net++, DeepLabV3+, and SegFormer-B0—were evaluated using 406 patches of 256 × 256 pixels at 2 m resolution from four LiDAR-covered subregions in Zhenxiong County, China. Under spatially grouped three-fold cross-validation, DeepLabV3+ with DEM + slope-gradient input and Dice + Focal loss achieved a mean pixel-level F1-score of 0.351, mIoU of 0.557, and Patch-F1 of 0.814. Input-feature experiments showed that slope-gradient information was particularly informative, whereas the incremental contribution of aspect was configuration-dependent; DEM + slope was retained as a parsimonious two-channel input. Sensitivity analysis showed that the rule-derived candidate definition changed materially with the screening parameters. Leave-one-subregion-out evaluation yielded a macro-averaged F1 of 0.321, indicating measurable within-county cross-subregion transfer. However, whole-area evaluation under natural candidate prevalence reduced the macro-average F1 to 0.072 at a fixed threshold and 0.095 using validation-derived operating thresholds. These results indicate that the proposed model is best interpreted as a raster-based surrogate for rule-derived geomorphological screening rather than as an independently validated landslide detector.
- 21 May 2026
- Published by Springer Nature
- p. 395-409
Dual-Level Contrastive Learning for Patient Condition Representation with Multimodal Electronic Health Records
- 21 May 2026
- Published by Springer Nature
- p. 395-409
Dual-Level Contrastive Learning for Patient Condition Representation with Multimodal Electronic Health Records
- 21 May 2026
- Published by Springer Nature
- p. 410-423
Text-Enhanced Panoptic Symbol Spotting in CAD Drawings
- 21 May 2026
- Published by Springer Nature
- p. 410-423
Text-Enhanced Panoptic Symbol Spotting in CAD Drawings
- Open Access
- 20 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 243-254
Incident-Guided Spatiotemporal Traffic Forecasting
Recent years have witnessed the rapid development of deep-learning-based, graph-neural-network-based forecasting methods for modern intelligent transportation systems. However, most existing work focuses exclusively on capturing spatio-temporal dependencies from historical traffic data, while overlooking the fact that suddenly occurring transportation incidents, such as traffic accidents and adverse weather, serve as external disturbances that can substantially alter temporal patterns. We argue that this issue has become a major obstacle to modeling the dynamics of traffic systems and improving prediction accuracy, but the unpredictability of incidents makes it difficult to observe patterns from historical sequences. To address these challenges, this paper proposes a novel framework named the Incident-Guided Spatiotemporal Graph Neural Network (IGSTGNN). IGSTGNN explicitly models the incident's impact through two core components: an Incident-Context Spatial Fusion (ICSF) module to capture the initial heterogeneous spatial influence, and a Temporal Incident Impact Decay (TIID) module to model the subsequent dynamic dissipation. To facilitate research on the spatio-temporal impact of incidents on traffic flow, a large-scale dataset is constructed and released, featuring incident records that are time-aligned with traffic time series. On this new benchmark, the proposed IGSTGNN framework is demonstrated to achieve state-of-the-art performance. Furthermore, the generalizability of the ICSF and TIID modules is validated by integrating them into various existing models.
- 20 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 243-254
Open AccessIncident-Guided Spatiotemporal Traffic Forecasting
Recent years have witnessed the rapid development of deep-learning-based, graph-neural-network-based forecasting methods for modern intelligent transportation systems. However, most existing work focuses exclusively on capturing spatio-temporal dependencies from historical traffic data, while overlooking the fact that suddenly occurring transportation incidents, such as traffic accidents and adverse weather, serve as external disturbances that can substantially alter temporal patterns. We argue that this issue has become a major obstacle to modeling the dynamics of traffic systems and improving prediction accuracy, but the unpredictability of incidents makes it difficult to observe patterns from historical sequences. To address these challenges, this paper proposes a novel framework named the Incident-Guided Spatiotemporal Graph Neural Network (IGSTGNN). IGSTGNN explicitly models the incident's impact through two core components: an Incident-Context Spatial Fusion (ICSF) module to capture the initial heterogeneous spatial influence, and a Temporal Incident Impact Decay (TIID) module to model the subsequent dynamic dissipation. To facilitate research on the spatio-temporal impact of incidents on traffic flow, a large-scale dataset is constructed and released, featuring incident records that are time-aligned with traffic time series. On this new benchmark, the proposed IGSTGNN framework is demonstrated to achieve state-of-the-art performance. Furthermore, the generalizability of the ICSF and TIID modules is validated by integrating them into various existing models.
- 31 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. PP, 1-13
Trans2Mamba: A Hybrid Attention-Mamba Architecture for Time Series Forecasting
Time series forecasting, particularly for data collected via Internet of Things (IoT) sensing innovations in transportation and energy sectors, plays a crucial role in understanding complex social behaviors and enhancing social intelligence. Accurate prediction of these sensor-driven signals is of great significance for optimizing societal resource allocation, proactively managing social dynamics, and solving complex urban challenges. However, because of the high nonlinearity, dynamism, and long-term dependence of time series, accurate prediction has become a challenging task. The existing models based on graph neural networks are challenging to adapt to dynamic dependencies that change over time, particularly when using static graph structures to represent dependency relationships. In addition, traditional structures have limitations in capturing long-term dependencies. Although attention-based methods have alleviated the above problems to some extent, their high computational complexity and memory overhead limit their practical application in large-scale scenarios. To this end, we propose a hybrid Attention-Mamba architecture named Trans2Mamba that combines attention and Mamba modules for efficient time series forecasting. This model can effectively capture the dynamic and remote dependency characteristics in time series. The model we propose consists of two main modules: an input embedding layer that adaptively characterizes the features of different time steps and variables through a learnable spatiotemporal embedding layer; a stacked ST Block consisting of an attention-based temporal block and a Mamba-based spatial block, aimed at accurately modeling spatiotemporal dependencies through division of labor and collaboration. We validated the effectiveness of the model on seven datasets covering electricity, meteorology, and transportation. In addition, the ablation experiment further verified the positive contribution and rational design of each key module in the model to the overall performance.
- 31 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. PP, 1-13
Trans2Mamba: A Hybrid Attention-Mamba Architecture for Time Series Forecasting
Time series forecasting, particularly for data collected via Internet of Things (IoT) sensing innovations in transportation and energy sectors, plays a crucial role in understanding complex social behaviors and enhancing social intelligence. Accurate prediction of these sensor-driven signals is of great significance for optimizing societal resource allocation, proactively managing social dynamics, and solving complex urban challenges. However, because of the high nonlinearity, dynamism, and long-term dependence of time series, accurate prediction has become a challenging task. The existing models based on graph neural networks are challenging to adapt to dynamic dependencies that change over time, particularly when using static graph structures to represent dependency relationships. In addition, traditional structures have limitations in capturing long-term dependencies. Although attention-based methods have alleviated the above problems to some extent, their high computational complexity and memory overhead limit their practical application in large-scale scenarios. To this end, we propose a hybrid Attention-Mamba architecture named Trans2Mamba that combines attention and Mamba modules for efficient time series forecasting. This model can effectively capture the dynamic and remote dependency characteristics in time series. The model we propose consists of two main modules: an input embedding layer that adaptively characterizes the features of different time steps and variables through a learnable spatiotemporal embedding layer; a stacked ST Block consisting of an attention-based temporal block and a Mamba-based spatial block, aimed at accurately modeling spatiotemporal dependencies through division of labor and collaboration. We validated the effectiveness of the model on seven datasets covering electricity, meteorology, and transportation. In addition, the ablation experiment further verified the positive contribution and rational design of each key module in the model to the overall performance.
- Open Access
- 27 February 2026
- Published by MDPI AG in Remote Sensing
- Vol. 18 (5) , 707
A Deformable Dual-Branch Visual State-Space Network for Landslide Identification with Multi-Scale Recognition and Irregular Boundary Enhancement
In recent years, rapid and reliable interpretation for emergency response to landslides and other geological hazards has become increasingly important. This paper presents DFmamba, an improved deformable dual-branch visual state-space network, to address engineering challenges such as missed large landslide bodies, boundary shifts, and loss of small-scale details. DFmamba mitigates the limited effective receptive field and window-partition constraints that often prevent existing methods from balancing large-area semantic consistency, multi-scale detection, precise boundary delineation, and computational efficiency. It employs a parallel encoder with a convolutional branch and a Visual State-Space Model (VSSM) branch to jointly capture local textures and global context. In the decoder, deformable residual blocks (DRB) enhance geometric modeling of irregular boundaries, while multi-scale feature alignment and a shallow high-frequency injection (MFP) mechanism strengthen boundary responses and preserve fine details. Experiments on the public CAS dataset against representative CNN-, Transformer-, and SSM-based baselines show that DFmamba achieves improved Precision, Recall, F1-score, and IoU, with stable performance across multi-scale scenarios, demonstrating strong robustness for landslide segmentation.
- 27 February 2026
- Published by MDPI AG in Remote Sensing
- Vol. 18 (5) , 707
Open AccessA Deformable Dual-Branch Visual State-Space Network for Landslide Identification with Multi-Scale Recognition and Irregular Boundary Enhancement
In recent years, rapid and reliable interpretation for emergency response to landslides and other geological hazards has become increasingly important. This paper presents DFmamba, an improved deformable dual-branch visual state-space network, to address engineering challenges such as missed large landslide bodies, boundary shifts, and loss of small-scale details. DFmamba mitigates the limited effective receptive field and window-partition constraints that often prevent existing methods from balancing large-area semantic consistency, multi-scale detection, precise boundary delineation, and computational efficiency. It employs a parallel encoder with a convolutional branch and a Visual State-Space Model (VSSM) branch to jointly capture local textures and global context. In the decoder, deformable residual blocks (DRB) enhance geometric modeling of irregular boundaries, while multi-scale feature alignment and a shallow high-frequency injection (MFP) mechanism strengthen boundary responses and preserve fine details. Experiments on the public CAS dataset against representative CNN-, Transformer-, and SSM-based baselines show that DFmamba achieves improved Precision, Recall, F1-score, and IoU, with stable performance across multi-scale scenarios, demonstrating strong robustness for landslide segmentation.
- Open Access
- 21 February 2026
- Published by Association for Computing Machinery (ACM)
- p. 89-98
RPE4Rec: Enhancing Dynamic Node Retrieval with Efficient Relative Position Encoding for Recommendation Systems
- 21 February 2026
- Published by Association for Computing Machinery (ACM)
- p. 89-98
Open AccessRPE4Rec: Enhancing Dynamic Node Retrieval with Efficient Relative Position Encoding for Recommendation Systems
- Open Access
Global-Lens Transformers: Adaptive Token Mixing for Dynamic Link Prediction
Dynamic graph learning plays a pivotal role in modeling evolving relationships over time, especially for temporal link prediction tasks in domains such as traffic systems, social networks, and recommendation platforms. While Transformer-based models have demonstrated strong performance by capturing long-range temporal dependencies, their reliance on self-attention results in quadratic complexity with respect to sequence length, limiting scalability on high-frequency or large-scale graphs. In this work, we revisit the necessity of self-attention in dynamic graph modeling. Inspired by recent findings that attribute the success of Transformers more to their architectural design than attention itself, we propose GLFormer, a novel attention-free Transformer-style framework for dynamic graphs. GLFormer introduces an adaptive token mixer that performs context-aware local aggregation based on interaction order and time intervals. To capture long-term dependencies, we further design a hierarchical aggregation module that expands the temporal receptive field by stacking local token mixers across layers. Experiments on six widely-used dynamic graph benchmarks show that GLFormer achieves SOTA performance, which reveals that attention-free architectures can match or surpass Transformer baselines in dynamic graph settings with significantly improved efficiency.
Open AccessGlobal-Lens Transformers: Adaptive Token Mixing for Dynamic Link Prediction
Dynamic graph learning plays a pivotal role in modeling evolving relationships over time, especially for temporal link prediction tasks in domains such as traffic systems, social networks, and recommendation platforms. While Transformer-based models have demonstrated strong performance by capturing long-range temporal dependencies, their reliance on self-attention results in quadratic complexity with respect to sequence length, limiting scalability on high-frequency or large-scale graphs. In this work, we revisit the necessity of self-attention in dynamic graph modeling. Inspired by recent findings that attribute the success of Transformers more to their architectural design than attention itself, we propose GLFormer, a novel attention-free Transformer-style framework for dynamic graphs. GLFormer introduces an adaptive token mixer that performs context-aware local aggregation based on interaction order and time intervals. To capture long-term dependencies, we further design a hierarchical aggregation module that expands the temporal receptive field by stacking local token mixers across layers. Experiments on six widely-used dynamic graph benchmarks show that GLFormer achieves SOTA performance, which reveals that attention-free architectures can match or surpass Transformer baselines in dynamic graph settings with significantly improved efficiency.
- Open Access
Text-Enhanced Panoptic Symbol Spotting in CAD Drawings
With the widespread adoption of Computer-Aided Design(CAD) drawings in engineering, architecture, and industrial design, the ability to accurately interpret and analyze these drawings has become increasingly critical. Among various subtasks, panoptic symbol spotting plays a vital role in enabling downstream applications such as CAD automation and design retrieval. Existing methods primarily focus on geometric primitives within the CAD drawings to address this task, but they face following major problems: they usually overlook the rich textual annotations present in CAD drawings and they lack explicit modeling of relationships among primitives, resulting in incomprehensive understanding of the holistic drawings. To fill this gap, we propose a panoptic symbol spotting framework that incorporates textual annotations. The framework constructs unified representations by jointly modeling geometric and textual primitives. Then, using visual features extract by pretrained CNN as the initial representations, a Transformer-based backbone is employed, enhanced with a type-aware attention mechanism to explicitly model the different types of spatial dependencies between various primitives. Extensive experiments on the real-world dataset demonstrate that the proposed method outperforms existing approaches on symbol spotting tasks involving textual annotations, and exhibits superior robustness when applied to complex CAD drawings.
Open AccessText-Enhanced Panoptic Symbol Spotting in CAD Drawings
With the widespread adoption of Computer-Aided Design(CAD) drawings in engineering, architecture, and industrial design, the ability to accurately interpret and analyze these drawings has become increasingly critical. Among various subtasks, panoptic symbol spotting plays a vital role in enabling downstream applications such as CAD automation and design retrieval. Existing methods primarily focus on geometric primitives within the CAD drawings to address this task, but they face following major problems: they usually overlook the rich textual annotations present in CAD drawings and they lack explicit modeling of relationships among primitives, resulting in incomprehensive understanding of the holistic drawings. To fill this gap, we propose a panoptic symbol spotting framework that incorporates textual annotations. The framework constructs unified representations by jointly modeling geometric and textual primitives. Then, using visual features extract by pretrained CNN as the initial representations, a Transformer-based backbone is employed, enhanced with a type-aware attention mechanism to explicitly model the different types of spatial dependencies between various primitives. Extensive experiments on the real-world dataset demonstrate that the proposed method outperforms existing approaches on symbol spotting tasks involving textual annotations, and exhibits superior robustness when applied to complex CAD drawings.