Chao Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Chao Liu has a total of 31 co-authors. Here are the top co-authors by number of shared publications:
- Hui Sun
PLA Academy of Military Science, Beijing, China, China
4 shared publications - Haoxiang Zhang3 shared publications
Hui Sun
3 shared publications- Jiangtao Wei
Fudan University
3 shared publications - Yunpeng Zhao3 shared publications
- Hongtao Song
College of Computer Science and Technology, Harbin Engineering University, Harbin 150001, China
2 shared publications
Latest Publications
- 16 October 2026
- Published by Elsevier in Optics & Laser Technology
- Vol. 204
CCR-STI: A coarse-to-fine framework with semantic guidance for cross-scenario robust registration of ship infrared images
- 16 October 2026
- Published by Elsevier in Optics & Laser Technology
- Vol. 204
CCR-STI: A coarse-to-fine framework with semantic guidance for cross-scenario robust registration of ship infrared images
- Open Access
- 22 September 2026
- Published by MDPI AG in Big Data and Cognitive Computing
- Vol. 10 (10) , 323
BiMLI: Bidirectional Link–Modality Interaction for Link Prediction in Attribute-Based Multimodal Knowledge Graphs
In attribute-based multimodal knowledge graphs, images and text serve as auxiliary entity attributes, providing semantic evidence beyond graph topology for link prediction. However, bidirectional conditional dependencies between an entity’s multimodal representations and its local links—the one-hop factual triples formed by relations and neighboring entities—remain insufficiently modeled: different local links rely on structural, visual, and textual evidence to varying degrees, and conversely, the relevance of each modality differs across local links. Existing methods have progressed from modality alignment and fusion to relation-aware, structure-aware, and adaptive selection; yet they mainly let information flow from links to modalities, while conditioning on each modality to distinguish and weight individual surrounding links remains underexplored, leaving the two selection directions disconnected. We therefore propose BiMLI, a bidirectional link–modality interaction framework. Link→Modality selects relevant structural, visual, and textual evidence conditioned on each local link, whereas Modality→Link dynamically weights an entity’s surrounding links conditioned on each modality representation. Interaction-aware feature fusion and neighborhood aggregation integrate both representations, with cross-modal contrastive regularization as an auxiliary constraint. Under a controlled comparison using ten matched random seeds, BiMLI attains mean MRR scores of 41.39% and 44.28% on DB15K and MKG-W, exceeding ROAD by 1.79 and 6.18 percentage points, respectively. Ablation and fine-grained analyses show that the two directions are functionally asymmetric yet complementary, providing evidence for the value of bidirectional conditional modeling.
- 22 September 2026
- Published by MDPI AG in Big Data and Cognitive Computing
- Vol. 10 (10) , 323
Open AccessBiMLI: Bidirectional Link–Modality Interaction for Link Prediction in Attribute-Based Multimodal Knowledge Graphs
In attribute-based multimodal knowledge graphs, images and text serve as auxiliary entity attributes, providing semantic evidence beyond graph topology for link prediction. However, bidirectional conditional dependencies between an entity’s multimodal representations and its local links—the one-hop factual triples formed by relations and neighboring entities—remain insufficiently modeled: different local links rely on structural, visual, and textual evidence to varying degrees, and conversely, the relevance of each modality differs across local links. Existing methods have progressed from modality alignment and fusion to relation-aware, structure-aware, and adaptive selection; yet they mainly let information flow from links to modalities, while conditioning on each modality to distinguish and weight individual surrounding links remains underexplored, leaving the two selection directions disconnected. We therefore propose BiMLI, a bidirectional link–modality interaction framework. Link→Modality selects relevant structural, visual, and textual evidence conditioned on each local link, whereas Modality→Link dynamically weights an entity’s surrounding links conditioned on each modality representation. Interaction-aware feature fusion and neighborhood aggregation integrate both representations, with cross-modal contrastive regularization as an auxiliary constraint. Under a controlled comparison using ten matched random seeds, BiMLI attains mean MRR scores of 41.39% and 44.28% on DB15K and MKG-W, exceeding ROAD by 1.79 and 6.18 percentage points, respectively. Ablation and fine-grained analyses show that the two directions are functionally asymmetric yet complementary, providing evidence for the value of bidirectional conditional modeling.
- 11 August 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 333
Thermal target reconstruction and novel view synthesis from sparse UAV observations using 3D Gaussian splatting
- 11 August 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 333
Thermal target reconstruction and novel view synthesis from sparse UAV observations using 3D Gaussian splatting
- Open Access
- 11 August 2026
- Published by MDPI AG in Preprints.org
BiMLI: Bidirectional Link–Modality Interaction for Link Prediction in Attribute-Based Multimodal Knowledge Graphs
In attribute-based multimodal knowledge graphs, images and text serve as auxiliary entity attributes, providing semantic evidence beyond graph topology for link prediction. However, bidirectional conditional dependencies between an entity's multimodal representations and its local links—the one-hop factual triples formed by relations and neighboring entities—remain insufficiently modeled: different local links rely on structural, visual, and textual evidence to varying degrees, and conversely, the relevance of each modality differs across local links. Existing methods have progressed from modality alignment and fusion to relation-aware, structure-aware, and adaptive selection; yet they mainly let information flow from links to modalities, while conditioning on each modality to distinguish and weight individual surrounding links remains underexplored, leaving the two selection directions disconnected. We therefore propose BiMLI, a Bidirectional Link–Modality Interaction framework. Link→Modality selects relevant structural, visual, and textual evidence conditioned on each local link, whereas Modality→Link dynamically weights an entity's surrounding links conditioned on each modality representation. Interaction-aware feature fusion and neighborhood aggregation integrate both representations, with cross-modal contrastive regularization as an auxiliary constraint. On DB15K and MKG-W, BiMLI achieves MRR scores of 42.35% and 43.55%, corresponding to relative improvements of 6.43% and 11.98% over the strongest respective baselines. Ablation and fine-grained analyses show that the two directions are functionally asymmetric yet complementary, validating the necessity of bidirectional conditional modeling.
- 11 August 2026
- Published by MDPI AG in Preprints.org
Open AccessBiMLI: Bidirectional Link–Modality Interaction for Link Prediction in Attribute-Based Multimodal Knowledge Graphs
In attribute-based multimodal knowledge graphs, images and text serve as auxiliary entity attributes, providing semantic evidence beyond graph topology for link prediction. However, bidirectional conditional dependencies between an entity's multimodal representations and its local links—the one-hop factual triples formed by relations and neighboring entities—remain insufficiently modeled: different local links rely on structural, visual, and textual evidence to varying degrees, and conversely, the relevance of each modality differs across local links. Existing methods have progressed from modality alignment and fusion to relation-aware, structure-aware, and adaptive selection; yet they mainly let information flow from links to modalities, while conditioning on each modality to distinguish and weight individual surrounding links remains underexplored, leaving the two selection directions disconnected. We therefore propose BiMLI, a Bidirectional Link–Modality Interaction framework. Link→Modality selects relevant structural, visual, and textual evidence conditioned on each local link, whereas Modality→Link dynamically weights an entity's surrounding links conditioned on each modality representation. Interaction-aware feature fusion and neighborhood aggregation integrate both representations, with cross-modal contrastive regularization as an auxiliary constraint. On DB15K and MKG-W, BiMLI achieves MRR scores of 42.35% and 43.55%, corresponding to relative improvements of 6.43% and 11.98% over the strongest respective baselines. Ablation and fine-grained analyses show that the two directions are functionally asymmetric yet complementary, validating the necessity of bidirectional conditional modeling.
- Open Access
- 7 July 2026
- Published by MDPI AG in Remote Sensing
- Vol. 18 (13) , 2247
Physically Consistent SAR Image Generation for Unseen Aspect Angles via Attributed Scattering Center Evolution
Synthetic aperture radar (SAR) target images are highly sensitive to aspect angle, while practical data acquisition usually provides only sparse observations over limited viewpoints. This leads to severe data scarcity at unseen aspect angles and makes cross-angle generation prone to scattering-structure distortion and background statistical mismatch. Existing SAR image generation methods either focus on distribution matching without sufficiently exploiting scattering-related structural cues, or emphasize angle conditioning while failing to jointly preserve physically plausible dominant scattering-response variations and realistic background speckle statistics at unseen aspect angles. To address this issue, we propose a physically consistent framework for SAR image generation at unseen aspect angles. The proposed method introduces an ASC-inspired sparse scattering-structure prior to approximate the dominant scattering responses in the SAR image plane. Rather than performing full parametric ASC inversion, this prior serves as a differentiable and angle-aware structural proxy that guides the generator toward synthesizing SAR images with structurally plausible scattering layouts. In addition, a dual-consistency scheme is introduced to jointly enforce target-region scattering consistency and background-region statistical consistency, thereby improving the physical realism of the generated results in both the target and background regions. Extensive experiments under strict unseen-angle interpolation and hold-out protocols demonstrate that the proposed method consistently outperforms representative baselines in image fidelity, target-region scattering consistency, background statistical consistency, and angle-conditioned consistency. Further visualization and ablation studies verify the critical role of the ASC-inspired sparse scattering-structure prior in physically consistent SAR view completion.
- 7 July 2026
- Published by MDPI AG in Remote Sensing
- Vol. 18 (13) , 2247
Open AccessPhysically Consistent SAR Image Generation for Unseen Aspect Angles via Attributed Scattering Center Evolution
Synthetic aperture radar (SAR) target images are highly sensitive to aspect angle, while practical data acquisition usually provides only sparse observations over limited viewpoints. This leads to severe data scarcity at unseen aspect angles and makes cross-angle generation prone to scattering-structure distortion and background statistical mismatch. Existing SAR image generation methods either focus on distribution matching without sufficiently exploiting scattering-related structural cues, or emphasize angle conditioning while failing to jointly preserve physically plausible dominant scattering-response variations and realistic background speckle statistics at unseen aspect angles. To address this issue, we propose a physically consistent framework for SAR image generation at unseen aspect angles. The proposed method introduces an ASC-inspired sparse scattering-structure prior to approximate the dominant scattering responses in the SAR image plane. Rather than performing full parametric ASC inversion, this prior serves as a differentiable and angle-aware structural proxy that guides the generator toward synthesizing SAR images with structurally plausible scattering layouts. In addition, a dual-consistency scheme is introduced to jointly enforce target-region scattering consistency and background-region statistical consistency, thereby improving the physical realism of the generated results in both the target and background regions. Extensive experiments under strict unseen-angle interpolation and hold-out protocols demonstrate that the proposed method consistently outperforms representative baselines in image fidelity, target-region scattering consistency, background statistical consistency, and angle-conditioned consistency. Further visualization and ablation studies verify the critical role of the ASC-inspired sparse scattering-structure prior in physically consistent SAR view completion.
- Open Access
- 24 June 2026
- Published by Elsevier in SSRN Electronic Journal
CoFIR-GAN: Visible-to-Infrared Image Generation with Multi-Scale Dual-Function Discrimination and Posterior Physical Constraints
Visible-to-infrared image generation aims to synthesize infrared-like imagery from easily accessible visible images, thereby alleviating the scarcity of infrared samples and supporting nighttime perception, intelligent driving, and cross-modal recognition. Existing methods mainly focus on cross-domain appearance mapping, while the thermal-radiative formation characteristics of infrared imagery are insufficiently modeled, which may lead to local thermal-texture distortion and physically inconsistent responses. This paper proposes CoFIR-GAN (Coarse-to-Fine Visible-to-Infrared Generative Adversarial Network), a visible-to-infrared generation framework that integrates multi-scale dual-function discrimination with posterior physical constraints. The proposed method adopts a coarse-to-fine adversarial generator as the main synthesis pathway,where global coarse generation, local enhancement, and high-frequency residual compensationjointly reconstruct infrared images. Meanwhile, a multi-scale dual-function discriminator introduces multi-scale structural auxiliary supervision in addition to real/fake discrimination. Afrozen posterior physical decomposition network further constrains the generated results from the perspectives of physical reconstruction consistency and physical-space consistency
- 24 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessCoFIR-GAN: Visible-to-Infrared Image Generation with Multi-Scale Dual-Function Discrimination and Posterior Physical Constraints
Visible-to-infrared image generation aims to synthesize infrared-like imagery from easily accessible visible images, thereby alleviating the scarcity of infrared samples and supporting nighttime perception, intelligent driving, and cross-modal recognition. Existing methods mainly focus on cross-domain appearance mapping, while the thermal-radiative formation characteristics of infrared imagery are insufficiently modeled, which may lead to local thermal-texture distortion and physically inconsistent responses. This paper proposes CoFIR-GAN (Coarse-to-Fine Visible-to-Infrared Generative Adversarial Network), a visible-to-infrared generation framework that integrates multi-scale dual-function discrimination with posterior physical constraints. The proposed method adopts a coarse-to-fine adversarial generator as the main synthesis pathway,where global coarse generation, local enhancement, and high-frequency residual compensationjointly reconstruct infrared images. Meanwhile, a multi-scale dual-function discriminator introduces multi-scale structural auxiliary supervision in addition to real/fake discrimination. Afrozen posterior physical decomposition network further constrains the generated results from the perspectives of physical reconstruction consistency and physical-space consistency
- Open Access
- 21 May 2026
- Published by MDPI AG in Preprints.org
Physically Consistent SAR Image Generation for Unseen Aspect Angles via Attributed Scattering Center Evolution
Synthetic aperture radar (SAR) target images are highly sensitive to aspect angle, while practical data acquisition usually provides only sparse observations over limited viewpoints. This leads to severe data scarcity at unseen aspect angles and makes cross-angle generation prone to scattering-structure distortion and background-statistics mismatch. Existing SAR image generation methods either focus on distribution matching without explicitly modeling target scattering mechanisms, or emphasize angle conditioning while failing to jointly preserve physically plausible scattering-structure evolution and realistic background speckle statistics at unseen views. To address this issue, we propose a physically consistent framework for SAR image generation at unseen aspect angles. The proposed method employs attributed scattering centers as an intermediate physical representation to explicitly model scattering-structure evolution across aspect angles, and uses this evolving representation to guide the generator toward synthesizing SAR images with structurally plausible scattering layouts. In addition, a dual-consistency scheme is introduced to jointly enforce target-region scattering consistency and background-region statistical consistency, thereby improving the physical realism of the generated results from both target and background perspectives. Extensive experiments under strict unseen-angle interpolation and hold-out protocols demonstrate that the proposed method consistently outperforms representative baselines in image fidelity, target-region scattering consistency, background statistical consistency, and angle-condition consistency. Further visualization and ablation studies verify the critical role of attributed scattering-center evolution modeling in physically consistent SAR view completion.
- 21 May 2026
- Published by MDPI AG in Preprints.org
Open AccessPhysically Consistent SAR Image Generation for Unseen Aspect Angles via Attributed Scattering Center Evolution
Synthetic aperture radar (SAR) target images are highly sensitive to aspect angle, while practical data acquisition usually provides only sparse observations over limited viewpoints. This leads to severe data scarcity at unseen aspect angles and makes cross-angle generation prone to scattering-structure distortion and background-statistics mismatch. Existing SAR image generation methods either focus on distribution matching without explicitly modeling target scattering mechanisms, or emphasize angle conditioning while failing to jointly preserve physically plausible scattering-structure evolution and realistic background speckle statistics at unseen views. To address this issue, we propose a physically consistent framework for SAR image generation at unseen aspect angles. The proposed method employs attributed scattering centers as an intermediate physical representation to explicitly model scattering-structure evolution across aspect angles, and uses this evolving representation to guide the generator toward synthesizing SAR images with structurally plausible scattering layouts. In addition, a dual-consistency scheme is introduced to jointly enforce target-region scattering consistency and background-region statistical consistency, thereby improving the physical realism of the generated results from both target and background perspectives. Extensive experiments under strict unseen-angle interpolation and hold-out protocols demonstrate that the proposed method consistently outperforms representative baselines in image fidelity, target-region scattering consistency, background statistical consistency, and angle-condition consistency. Further visualization and ablation studies verify the critical role of attributed scattering-center evolution modeling in physically consistent SAR view completion.
- 1 May 2026
- Published by Springer Nature
- p. 259-268
Preliminary Study of an Ontology Design Methodology for Target Characteristics Domain
- 1 May 2026
- Published by Springer Nature
- p. 259-268
Preliminary Study of an Ontology Design Methodology for Target Characteristics Domain
- Open Access
- 2 April 2026
- Published by Elsevier in SSRN Electronic Journal
Thermal Target Reconstruction and Novel View Synthesis from Sparse UAV Observations using 3D Gaussian Splatting
In real world unmanned aerial vehicle (UAV) maritime monitoring scenarios, infrared data acquisition is often constrained by limited flight time, sensor capability, and environmental conditions, resulting in sparse observations that pose a fundamental challenge to data-driven reconstruction and rendering methods. To address this issue, we propose SparseIR-GS, a 3D Gaussian Splatting-based framework for thermal scene reconstruction and novel view synthesis from sparse infrared images. Specifically, we introduce an uncertainty aware optimization strategy to constrain dual radiance fields, a thermal target guided densification mechanism to focus Gaussian distribution on high temperature regions, and a view depth guided pruning scheme to suppress geometrically inconsistent floating artifacts. To support realistic evaluation, we construct the Sanya IR dataset, a UAV-based multi-view thermal benchmark of maritime targets, comprising 7,414 infrared images of five ship categories captured under diverse viewpoints and temporal conditions. Extensive experiments on both the Sanya IR dataset and the public TL-NSD dataset demonstrate that SparseIR-GS consistently outperforms state-of-the-art methods under sparse-view settings. In particular, our method achieves the best PSNR and SSIM under 6-view and 12-view configurations, while enabling reliable geometry reconstruction from 6 infrared images. SparseIR-GS provides a practical and efficient solution for data-efficient infrared scene understanding, offering strong potential for real-world UAV applications such as maritime surveillance, thermal target perception, and airborne infrared monitoring. Sanya IR dataset will be released at https://github.com/xxxxxxxx/.
- 2 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessThermal Target Reconstruction and Novel View Synthesis from Sparse UAV Observations using 3D Gaussian Splatting
In real world unmanned aerial vehicle (UAV) maritime monitoring scenarios, infrared data acquisition is often constrained by limited flight time, sensor capability, and environmental conditions, resulting in sparse observations that pose a fundamental challenge to data-driven reconstruction and rendering methods. To address this issue, we propose SparseIR-GS, a 3D Gaussian Splatting-based framework for thermal scene reconstruction and novel view synthesis from sparse infrared images. Specifically, we introduce an uncertainty aware optimization strategy to constrain dual radiance fields, a thermal target guided densification mechanism to focus Gaussian distribution on high temperature regions, and a view depth guided pruning scheme to suppress geometrically inconsistent floating artifacts. To support realistic evaluation, we construct the Sanya IR dataset, a UAV-based multi-view thermal benchmark of maritime targets, comprising 7,414 infrared images of five ship categories captured under diverse viewpoints and temporal conditions. Extensive experiments on both the Sanya IR dataset and the public TL-NSD dataset demonstrate that SparseIR-GS consistently outperforms state-of-the-art methods under sparse-view settings. In particular, our method achieves the best PSNR and SSIM under 6-view and 12-view configurations, while enabling reliable geometry reconstruction from 6 infrared images. SparseIR-GS provides a practical and efficient solution for data-efficient infrared scene understanding, offering strong potential for real-world UAV applications such as maritime surveillance, thermal target perception, and airborne infrared monitoring. Sanya IR dataset will be released at https://github.com/xxxxxxxx/.
- Open Access
Humanoid-inspired Causal Representation Learning for Domain Generalization
This paper proposes the Humanoid-inspired Structural Causal Model (HSCM), a novel causal framework inspired by human intelligence, designed to overcome the limitations of conventional domain generalization models. Unlike approaches that rely on statistics to capture data-label dependencies and learn distortion-invariant representations, HSCM replicates the hierarchical processing and multi-level learning of human vision systems, focusing on modeling fine-grained causal mechanisms. By disentangling and reweighting key image attributes such as color, texture, and shape, HSCM enhances generalization across diverse domains, ensuring robust performance and interpretability. Leveraging the flexibility and adaptability of human intelligence, our approach enables more effective transfer and learning in dynamic, complex environments. Through both theoretical and empirical evaluations, we demonstrate that HSCM outperforms existing domain generalization models, providing a more principled method for capturing causal relationships and improving model robustness. The code is available at https://github.com/lambett/HSCM.
Open AccessHumanoid-inspired Causal Representation Learning for Domain Generalization
This paper proposes the Humanoid-inspired Structural Causal Model (HSCM), a novel causal framework inspired by human intelligence, designed to overcome the limitations of conventional domain generalization models. Unlike approaches that rely on statistics to capture data-label dependencies and learn distortion-invariant representations, HSCM replicates the hierarchical processing and multi-level learning of human vision systems, focusing on modeling fine-grained causal mechanisms. By disentangling and reweighting key image attributes such as color, texture, and shape, HSCM enhances generalization across diverse domains, ensuring robust performance and interpretability. Leveraging the flexibility and adaptability of human intelligence, our approach enables more effective transfer and learning in dynamic, complex environments. Through both theoretical and empirical evaluations, we demonstrate that HSCM outperforms existing domain generalization models, providing a more principled method for capturing causal relationships and improving model robustness. The code is available at https://github.com/lambett/HSCM.