Junping Yao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Junping Yao has a total of 50 co-authors. Here are the top co-authors by number of shared publications:
- Xiaojun Li
Rocket Force University of Engineering, Xi’an, China
10 shared publications - Yi Su
National University of Defense Technology
4 shared publications Haifeng Sun
Rocket Force University of Engineering, Xi’an, China
3 shared publications- Chengrong Dong
Xi’an Research Institute of High Technology, Xi’an, China
2 shared publications - Chengrong Dong2 shared publications
- Haifeng Sun
Rocket Force University of Engineering, Xi'an, Shaanxi, China
2 shared publications
Latest Publications
- 26 September 2026
- Published by Elsevier in Future Generation Computer Systems
- Vol. 188
FedHGA: Federated segmentation framework for heterogeneous granularity annotations
- 26 September 2026
- Published by Elsevier in Future Generation Computer Systems
- Vol. 188
FedHGA: Federated segmentation framework for heterogeneous granularity annotations
- Open Access
- 31 August 2026
- Published by Springer Nature in Scientific Reports
SGSR: structure-guided semantic recovery for event-based spiking semantic segmentation
- 31 August 2026
- Published by Springer Nature in Scientific Reports
Open AccessSGSR: structure-guided semantic recovery for event-based spiking semantic segmentation
- Open Access
- 24 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 584-590
SGHD: A Spatially Guided Hypergraph Diffusion Network for Next POI Recommendation
- 24 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 584-590
Open AccessSGHD: A Spatially Guided Hypergraph Diffusion Network for Next POI Recommendation
- 7 February 2026
- Published by Elsevier in Neurocomputing
- Vol. 675
CGEM: A cognitive-guided network for human-aligned entity matching
- 7 February 2026
- Published by Elsevier in Neurocomputing
- Vol. 675
CGEM: A cognitive-guided network for human-aligned entity matching
- Open Access
- 25 July 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 27093
Contextual semantics graph attention network model for entity resolution
Entity resolution technology is the process of distinguishing whether data from different knowledge bases refer to the same entity in the real world. Existing research takes entity pairs as input and makes judgments based on the characteristics of entity pairs. However, there is insufficient utilization of contextual semantics, as existing methods fail to effectively model the token-attribute associations within data sources and cross-attribute semantic hierarchical relationships, which weakens the discriminative power of key attributes. What' more, they exhibit failure in handling polysemous ambiguities, as conventional graph neural network adopts rigid node representations that cannot dynamically adjust word meanings according to attribute-specific contexts. To address this issue, this paper proposes the Contextual Semantics Graph Attention Network (CSGAT), which extracts contextual information at token and attribute levels to generate semantically fused embeddings. The advantages of CSGAT are: 1) Leveraging the Transformer self-attention mechanism to extract feature vectors of words, model sequence relationships, and calculate the degree of relevance with other words; 2) Employing the attention mechanism on contextual information at the attribute level to extract semantic embeddings to enrich attribute embeddings, forming more discriminative attribute embeddings; 3) Utilizing the graph attention network to generate residual vectors for final entity resolution decisions. Experimental on Amazon-Google and BeerAdvo-RateBeer datasets show that, as compared with the competing methods, CSGAT can achieve significant improved performance on F1-score with fine Precision and Recall. Code is available at https://github.com/xhtech2024/csgat .
- 25 July 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 27093
Open AccessContextual semantics graph attention network model for entity resolution
Entity resolution technology is the process of distinguishing whether data from different knowledge bases refer to the same entity in the real world. Existing research takes entity pairs as input and makes judgments based on the characteristics of entity pairs. However, there is insufficient utilization of contextual semantics, as existing methods fail to effectively model the token-attribute associations within data sources and cross-attribute semantic hierarchical relationships, which weakens the discriminative power of key attributes. What' more, they exhibit failure in handling polysemous ambiguities, as conventional graph neural network adopts rigid node representations that cannot dynamically adjust word meanings according to attribute-specific contexts. To address this issue, this paper proposes the Contextual Semantics Graph Attention Network (CSGAT), which extracts contextual information at token and attribute levels to generate semantically fused embeddings. The advantages of CSGAT are: 1) Leveraging the Transformer self-attention mechanism to extract feature vectors of words, model sequence relationships, and calculate the degree of relevance with other words; 2) Employing the attention mechanism on contextual information at the attribute level to extract semantic embeddings to enrich attribute embeddings, forming more discriminative attribute embeddings; 3) Utilizing the graph attention network to generate residual vectors for final entity resolution decisions. Experimental on Amazon-Google and BeerAdvo-RateBeer datasets show that, as compared with the competing methods, CSGAT can achieve significant improved performance on F1-score with fine Precision and Recall. Code is available at https://github.com/xhtech2024/csgat .
- 10 June 2025
- Published by Springer Nature in Arabian Journal for Science and Engineering
- Vol. 50 (23) , 19879-19894
FedMTL: Federated Semantic Segmentation in Multi-type Label Scenarios
- 10 June 2025
- Published by Springer Nature in Arabian Journal for Science and Engineering
- Vol. 50 (23) , 19879-19894
FedMTL: Federated Semantic Segmentation in Multi-type Label Scenarios
- Open Access
- 14 March 2025
- Published by Association for Computing Machinery (ACM)
- p. 545-552
Label-noise-robust Two-Stage Federated Learning in Sensor-Based Human Activity Recognition
- 14 March 2025
- Published by Association for Computing Machinery (ACM)
- p. 545-552
Open AccessLabel-noise-robust Two-Stage Federated Learning in Sensor-Based Human Activity Recognition
- Open Access
- 15 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 612-616
A Joint Generative Discriminant Network for Person Re-identification with Limited Computing Resources
- 15 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 612-616
Open AccessA Joint Generative Discriminant Network for Person Re-identification with Limited Computing Resources
- 23 July 2024
- Published by Elsevier in Knowledge-Based Systems
- Vol. 301
LCFE-X: Pluggable Local Channel Feature Extraction model enhancement method for hyperspectral image classification
- 23 July 2024
- Published by Elsevier in Knowledge-Based Systems
- Vol. 301
LCFE-X: Pluggable Local Channel Feature Extraction model enhancement method for hyperspectral image classification
- Open Access
- 1 May 2024
- Published by Institution of Engineering and Technology (IET) in IET Image Processing
- Vol. 18 (10) , 2553-2567
Factor annealing decoupling compositional training method for imbalanced hyperspectral image classification
Due to differences in the quantity and size of observed targets, hyperspectral images are characterized by class imbalance. The standard deep learning classification model training scheme optimizes the overall classification error, which may lead to performance imbalance between classes in hyperspectral image classification frameworks. Therefore, a novel factor annealing decoupling compositional training method is proposed in this paper. Without requiring resampling or reweighting, it implicitly modulates the training process, so standard models can sufficiently learn the representation of the minority classes and further be trained as robust classifiers. Specifically, the label‐distribution‐aware margin loss is combined with the error‐rate‐based cross‐entropy loss via combination factor, which considers both imbalanced data representation learning and classifier overall performance. Then, a factor annealing optimization training scheme is designed to adjust the combination factor, which solves the stage division problem of two‐stage decoupling learning. Experimental results on two hyperspectral image datasets demonstrate that, as compared with other competing approaches, the proposed method can continuously and stably optimize the model parameters, achieving improvements in class average metrics and difficult classes without affecting overall classification performance.
- 1 May 2024
- Published by Institution of Engineering and Technology (IET) in IET Image Processing
- Vol. 18 (10) , 2553-2567
Open AccessFactor annealing decoupling compositional training method for imbalanced hyperspectral image classification
Due to differences in the quantity and size of observed targets, hyperspectral images are characterized by class imbalance. The standard deep learning classification model training scheme optimizes the overall classification error, which may lead to performance imbalance between classes in hyperspectral image classification frameworks. Therefore, a novel factor annealing decoupling compositional training method is proposed in this paper. Without requiring resampling or reweighting, it implicitly modulates the training process, so standard models can sufficiently learn the representation of the minority classes and further be trained as robust classifiers. Specifically, the label‐distribution‐aware margin loss is combined with the error‐rate‐based cross‐entropy loss via combination factor, which considers both imbalanced data representation learning and classifier overall performance. Then, a factor annealing optimization training scheme is designed to adjust the combination factor, which solves the stage division problem of two‐stage decoupling learning. Experimental results on two hyperspectral image datasets demonstrate that, as compared with other competing approaches, the proposed method can continuously and stably optimize the model parameters, achieving improvements in class average metrics and difficult classes without affecting overall classification performance.