Licheng Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Licheng Liu has a total of 82 co-authors. Here are the top co-authors by number of shared publications:
- Chun Lung Philip Chen
South China University of Technology
16 shared publications - Yaonan Wang
Hunan University
13 shared publications - Bin Yang
School of Artificial Intelligence and Robotics, Hunan University, Changsha, China
10 shared publications - Xinxin Liu
School of Artificial Intelligence and Robotics, Hunan University, Changsha, China
7 shared publications - Tingyun Liu
College of Electrical and Information Engineering, Hunan University, Changsha, China
6 shared publications - Shutao Li
School of Artificial Intelligence and Robotics, Hunan University, Changsha, China
5 shared publications
Latest Publications
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. 12 (5) , 1938-1949
Context-Aware Dual-Graph Neural Network for Cellular State Inference Under Genetic Perturbations
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. 12 (5) , 1938-1949
Context-Aware Dual-Graph Neural Network for Cellular State Inference Under Genetic Perturbations
- 22 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Systems, Man, and Cybernetics: Systems
- Vol. PP (21682216) , 1-15
Dual-Branch Transformation Broad Learning for Multiclass Recognition
Least-squares-regression (LSRs)-based classification methods typically fail to achieve satisfactory performance due to their limited ability to learn discriminative features, while broad-learning-system (BLS)-based models are prone to overfitting as they are required to map high-dimensional features directly to rigid binary labels. To address these limitations, this article proposes a novel dual-branch transformation broad learning (DTBL) model. Concretely, DTBL incorporates two complementary transformation branches: the main branch ensures supervision fidelity by aligning feature representations with ground-truth binary labels, while the auxiliary branch adaptively constructs a relaxed label space that accommodates intraclass variations and enhances interclass separability. A learnable structural similarity constraint is employed to semantically couple the two branches, enabling the sharing of discriminative patterns without undermining their individual functional roles. To further improve model interpretability and generalization ability, we regularize the relaxed label matrix with sparsity and non-negativity constraints. Comprehensive experiments on multiple image classification benchmarks demonstrate that DTBL consistently outperforms existing LSR-based and BLS-based methods, particularly in small-sample scenarios.
- 22 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Systems, Man, and Cybernetics: Systems
- Vol. PP (21682216) , 1-15
Dual-Branch Transformation Broad Learning for Multiclass Recognition
Least-squares-regression (LSRs)-based classification methods typically fail to achieve satisfactory performance due to their limited ability to learn discriminative features, while broad-learning-system (BLS)-based models are prone to overfitting as they are required to map high-dimensional features directly to rigid binary labels. To address these limitations, this article proposes a novel dual-branch transformation broad learning (DTBL) model. Concretely, DTBL incorporates two complementary transformation branches: the main branch ensures supervision fidelity by aligning feature representations with ground-truth binary labels, while the auxiliary branch adaptively constructs a relaxed label space that accommodates intraclass variations and enhances interclass separability. A learnable structural similarity constraint is employed to semantically couple the two branches, enabling the sharing of discriminative patterns without undermining their individual functional roles. To further improve model interpretability and generalization ability, we regularize the relaxed label matrix with sparsity and non-negativity constraints. Comprehensive experiments on multiple image classification benchmarks demonstrate that DTBL consistently outperforms existing LSR-based and BLS-based methods, particularly in small-sample scenarios.
- 5 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. PP (10518215) , 1
WaveletMixer: When Wavelet Transform Meets Self-Attention in Single Image Super-Resolution
Transformer-based methods have shown strong potential for single image super-resolution (SISR) due to their ability to model long-range dependencies. However, most existing approaches primarily focus on global representation learning while underexploring high-frequency priors that are crucial for recovering fine textures and sharp edges. In this work, we present a wavelet-transform-based analysis showing that transformers are effective at modeling low-frequency structures but comparatively less effective than convolutional neural networks (CNNs) in recovering high-frequency details. Motivated by this observation, we propose WaveletMixer, a novel SISR framework that integrates wavelet decomposition into feature learning to exploit the complementary strengths of transformers and CNNs in the frequency domain. WaveletMixer adopts a dual-branch architecture for low- and high-frequency wavelet feature extraction, introduces Wavelet Channel Attention (WCA) to adaptively emphasize informative wavelet subbands, and employs Deformable Cross Attention (DCA) for cross-wavelet subbands interaction and fusion. In addition, an Omni Channel Attention (Omni-CA) module is designed to enhance spatial-channel feature aggregation through multi-axis interaction. Extensive experiments on benchmark datasets demonstrate that WaveletMixer consistently outperforms several state-of-the-art methods and achieves a favorable trade-off between reconstruction performance and model complexity.
- 5 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. PP (10518215) , 1
WaveletMixer: When Wavelet Transform Meets Self-Attention in Single Image Super-Resolution
Transformer-based methods have shown strong potential for single image super-resolution (SISR) due to their ability to model long-range dependencies. However, most existing approaches primarily focus on global representation learning while underexploring high-frequency priors that are crucial for recovering fine textures and sharp edges. In this work, we present a wavelet-transform-based analysis showing that transformers are effective at modeling low-frequency structures but comparatively less effective than convolutional neural networks (CNNs) in recovering high-frequency details. Motivated by this observation, we propose WaveletMixer, a novel SISR framework that integrates wavelet decomposition into feature learning to exploit the complementary strengths of transformers and CNNs in the frequency domain. WaveletMixer adopts a dual-branch architecture for low- and high-frequency wavelet feature extraction, introduces Wavelet Channel Attention (WCA) to adaptively emphasize informative wavelet subbands, and employs Deformable Cross Attention (DCA) for cross-wavelet subbands interaction and fusion. In addition, an Omni Channel Attention (Omni-CA) module is designed to enhance spatial-channel feature aggregation through multi-axis interaction. Extensive experiments on benchmark datasets demonstrate that WaveletMixer consistently outperforms several state-of-the-art methods and achieves a favorable trade-off between reconstruction performance and model complexity.
- 29 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 35 (10577149) , 4961-4974
LSRNet: A Novel Interpretable Low-rank Sparse Representation Guided Fusion Network for Polarization and Intensity Images
Polarization and intensity images fusion (PIF) has extracted extensive attentions as it can generate images with clear scene information and salient texture details of the object surface that are important for downstream applications. However, existing deep learning-based PIF methods usually lack interpretability and ignore the interactions among multi-modal features. To this end, we propose a novel interpretable low-rank sparse representation guided fusion network for polarization and intensity images (termed LSRNet). Specifically, a low-rank sparse representation deep unfolding module is designed to acquire the base and detail features of the source images, with the ability of improving the interpretability of the network. In addition, a cross-modal connection complementary feature extraction module is proposed, which aims to establish dependency among features of multi-modalities to fully extract complementary features of the source images. In order to demonstrate the validity of our LSRNet and take into account shortcomings of existing datasets for PIF, a multi-scene polarization and intensity image dataset, named MSPI dataset, is constructed, which includes 1034 high-resolution aligned image pairs. According to the best of our knowledge, this is the most comprehensive dataset for PIF that with a large number of image pairs, high resolution and multiple scene types. Extensive experiments on our MSPI dataset and two publicly available datasets (i.e., 12CFC and HCP) demonstrate the superior fusion performance, generalization ability, and desirable running efficiency of our LSRNet. Our codes and dataset will be publicly available at https://github.com/thebinyang/LSRNet.
- 29 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 35 (10577149) , 4961-4974
LSRNet: A Novel Interpretable Low-rank Sparse Representation Guided Fusion Network for Polarization and Intensity Images
Polarization and intensity images fusion (PIF) has extracted extensive attentions as it can generate images with clear scene information and salient texture details of the object surface that are important for downstream applications. However, existing deep learning-based PIF methods usually lack interpretability and ignore the interactions among multi-modal features. To this end, we propose a novel interpretable low-rank sparse representation guided fusion network for polarization and intensity images (termed LSRNet). Specifically, a low-rank sparse representation deep unfolding module is designed to acquire the base and detail features of the source images, with the ability of improving the interpretability of the network. In addition, a cross-modal connection complementary feature extraction module is proposed, which aims to establish dependency among features of multi-modalities to fully extract complementary features of the source images. In order to demonstrate the validity of our LSRNet and take into account shortcomings of existing datasets for PIF, a multi-scene polarization and intensity image dataset, named MSPI dataset, is constructed, which includes 1034 high-resolution aligned image pairs. According to the best of our knowledge, this is the most comprehensive dataset for PIF that with a large number of image pairs, high resolution and multiple scene types. Extensive experiments on our MSPI dataset and two publicly available datasets (i.e., 12CFC and HCP) demonstrate the superior fusion performance, generalization ability, and desirable running efficiency of our LSRNet. Our codes and dataset will be publicly available at https://github.com/thebinyang/LSRNet.
- Top Cited Papers
- 1 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 45 (4) , 1293-1310
CiSeg: Unsupervised Cross-Modality Adaptation for 3D Medical Image Segmentation via Causal Intervention
Unsupervised domain adaptation (UDA) addresses the domain shift problem by transferring knowledge from labeled source domain data (e.g. CT) to unlabeled target domain data (e.g. MRI). While state-of-the-art methods reduce domain gaps via image- or feature-level alignment, their reliance on spurious correlations in the training data often limits generalization across domains. To overcome this limitation, we propose the Causal Intervention Segmentation Network (CiSeg), a novel framework that first integrates causal inference into UDA. A Structural Causal Model (SCM) is first constructed for the source domain to disentangle causal variables from bias variables, alleviating the impact of spurious correlations. Based on this SCM, we introduce a Counterfactual Disentanglement (CD) module to decompose the source domain's latent features into distinct causal and bias components, effectively eliminating their mutual dependencies. To enhance cross-domain consistency, two auxiliary components are introduced: Prototype-guided Contrastive Learning (PCL) and Causal-bias Residual Alignment (CBRA). PCL aligns pixel-level representations with their corresponding semantic prototypes, promoting stronger intra-class consistency and clearer inter-class separability. CBRA employs adversarial learning to align causal and bias residual features across domains, further enhancing feature-level invariance. Extensive experiments on cardiac, abdominal multi-organ, and BraTS18 segmentation tasks demonstrate that CiSeg outperforms state-of-the-art methods, achieving superior segmentation performance and robust cross-domain generalization. Code and models are available at https://github.com/lvpeiqing/CiSeg.
- 1 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 45 (4) , 1293-1310
Top Cited PapersCiSeg: Unsupervised Cross-Modality Adaptation for 3D Medical Image Segmentation via Causal Intervention
Unsupervised domain adaptation (UDA) addresses the domain shift problem by transferring knowledge from labeled source domain data (e.g. CT) to unlabeled target domain data (e.g. MRI). While state-of-the-art methods reduce domain gaps via image- or feature-level alignment, their reliance on spurious correlations in the training data often limits generalization across domains. To overcome this limitation, we propose the Causal Intervention Segmentation Network (CiSeg), a novel framework that first integrates causal inference into UDA. A Structural Causal Model (SCM) is first constructed for the source domain to disentangle causal variables from bias variables, alleviating the impact of spurious correlations. Based on this SCM, we introduce a Counterfactual Disentanglement (CD) module to decompose the source domain's latent features into distinct causal and bias components, effectively eliminating their mutual dependencies. To enhance cross-domain consistency, two auxiliary components are introduced: Prototype-guided Contrastive Learning (PCL) and Causal-bias Residual Alignment (CBRA). PCL aligns pixel-level representations with their corresponding semantic prototypes, promoting stronger intra-class consistency and clearer inter-class separability. CBRA employs adversarial learning to align causal and bias residual features across domains, further enhancing feature-level invariance. Extensive experiments on cardiac, abdominal multi-organ, and BraTS18 segmentation tasks demonstrate that CiSeg outperforms state-of-the-art methods, achieving superior segmentation performance and robust cross-domain generalization. Code and models are available at https://github.com/lvpeiqing/CiSeg.
- 9 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Emerging Topics in Computational Intelligence
- Vol. 10 (3) , 2498-2515
Weak-Strong Consistency Learning via Collaborative Space Alignment for Semi-Supervised Medical Image Segmentation
In semi-supervised learning, cross pseudo supervision (CPS) is considered a promising learning approach.However, we observe that CPS easily converges to consensus early in training and degenerates into noisy self-training. To tackle this challenge, we propose a novel semi-supervised segmentation framework that utilizes collaborative spatial complementary alignment to facilitate weak-to-strong consistency learning (CSWS). CSWS comprises two main components: Weak-to-Strong Consistency Learning Strategy (WSCL) and Collaborative Space Complementary Alignment Strategy (CSCA). WSCL is implemented by combining weak augmentation and strong augmentation in a differently initialized network, thus effectively combining the benefits of consistency and self-training. To suppress the impact of low-quality pseudo-labels on consistency learning, CSCA leverages the semantic information in logit space and representation space to offer complementary features for consistency learning in a collaborative manner. In addition, we enrich the model inference view differences by introducing a prototype feature discrepancy loss, which is expected to facilitate consistent learning to obtain more precise predictions. We have conducted extensive experiments on five public datasets, including MRI, CT, and endoscopic imaging modalities. The experimental results show that CSWS has better generalization and high robustness than other state-of-the-art methods.
- 9 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Emerging Topics in Computational Intelligence
- Vol. 10 (3) , 2498-2515
Weak-Strong Consistency Learning via Collaborative Space Alignment for Semi-Supervised Medical Image Segmentation
In semi-supervised learning, cross pseudo supervision (CPS) is considered a promising learning approach.However, we observe that CPS easily converges to consensus early in training and degenerates into noisy self-training. To tackle this challenge, we propose a novel semi-supervised segmentation framework that utilizes collaborative spatial complementary alignment to facilitate weak-to-strong consistency learning (CSWS). CSWS comprises two main components: Weak-to-Strong Consistency Learning Strategy (WSCL) and Collaborative Space Complementary Alignment Strategy (CSCA). WSCL is implemented by combining weak augmentation and strong augmentation in a differently initialized network, thus effectively combining the benefits of consistency and self-training. To suppress the impact of low-quality pseudo-labels on consistency learning, CSCA leverages the semantic information in logit space and representation space to offer complementary features for consistency learning in a collaborative manner. In addition, we enrich the model inference view differences by introducing a prototype feature discrepancy loss, which is expected to facilitate consistent learning to obtain more precise predictions. We have conducted extensive experiments on five public datasets, including MRI, CT, and endoscopic imaging modalities. The experimental results show that CSWS has better generalization and high robustness than other state-of-the-art methods.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 35, 10272-10285
VLMSNet: View Large to Measure Shape
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 35, 10272-10285
VLMSNet: View Large to Measure Shape
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64, 5527116
Interpretable Cross-Sensor Hyperspectral Image Super-Resolution
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64, 5527116
Interpretable Cross-Sensor Hyperspectral Image Super-Resolution
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 35, 1
Learning to Super-Resolve Face Images via Dual-Domain Multi-scale Feature Interaction
Face Super-Resolution (FSR), aiming to improve the quality of Low-Resolution (LR) facial images, has been greatly propelled by the deep learning techniques. However, existing approaches, whether based on Convolutional Neural Networks (CNNs) or Transformers, are either inherently damaging facial structures limited by their architectures or failing to capture essential multi-scale textures due to the rigid receptive fields. To address these concerns, we propose a novel dual-domain feature interaction method called Spatial-frequency Multi-scale feature Learning Network (SMLNet) for FSR by employing a dual-branch architecture. Specifically, the frequency branch captures high-quality global structures and fine high-frequency details, while the spatial branch operates complementarily to preserve fine-grained local texture patterns. Moreover, we further introduce a Multi-scale Spatial-frequency feature Interaction Module (MSIM), which combines a Multi-scale Feature Extraction Block (MFEB) and a Spatial-Frequency feature Interaction Module (SFIM) to interact and aggregate multi-level complementary features from the dual branches. Extensive quantitative experiments and qualitative analyses across multiple datasets, together with evaluations on real-world images, demonstrate that the proposed SMLNet significantly outperforms other state-of-the-art methods.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 35, 1
Learning to Super-Resolve Face Images via Dual-Domain Multi-scale Feature Interaction
Face Super-Resolution (FSR), aiming to improve the quality of Low-Resolution (LR) facial images, has been greatly propelled by the deep learning techniques. However, existing approaches, whether based on Convolutional Neural Networks (CNNs) or Transformers, are either inherently damaging facial structures limited by their architectures or failing to capture essential multi-scale textures due to the rigid receptive fields. To address these concerns, we propose a novel dual-domain feature interaction method called Spatial-frequency Multi-scale feature Learning Network (SMLNet) for FSR by employing a dual-branch architecture. Specifically, the frequency branch captures high-quality global structures and fine high-frequency details, while the spatial branch operates complementarily to preserve fine-grained local texture patterns. Moreover, we further introduce a Multi-scale Spatial-frequency feature Interaction Module (MSIM), which combines a Multi-scale Feature Extraction Block (MFEB) and a Spatial-Frequency feature Interaction Module (SFIM) to interact and aggregate multi-level complementary features from the dual branches. Extensive quantitative experiments and qualitative analyses across multiple datasets, together with evaluations on real-world images, demonstrate that the proposed SMLNet significantly outperforms other state-of-the-art methods.
- Open Access
- 1 January 2026
- Published by MDPI AG in Remote Sensing
- Vol. 18 (1) , 143
When Deep Learning Meets Broad Learning: A Unified Framework for Change Detection with Synthetic Aperture Radar Images
Change detection (CD) with synthetic aperture radar (SAR) images remains pivotal for environmental monitoring and disaster management. Deep learning has powerful feature extraction capabilities for CD, but suffers from complex architectures and limited interpretability. While BLSs demonstrate advantages in structural simplicity and interpretability, their feature representation capacity remains constrained. In high-precision CD with SAR images, strong feature representation capability is required, along with an uncomplicated framework and high interpretability. Therefore, a novel paradigm named PC-BiBL is proposed which achieves seamless integration of deep learning and broad learning. On the one hand, it employs a hierarchical cross-convolutional encoding (HCCE) module that uses pseudo-random cross-convolution (PCConv) for hierarchical cross-feature representation, aggregating contextual information. PCConv is an untrained convolution layer, which can utilize specialized pseudo-random kernels to extract features from bitemporal SAR images. On the other hand, since back-propagation algorithms are not required, the features can be directly fed into the bifurcated broad learning (BiBL) module for node expansion and direct parameter computation. BiBL constructs dual-branch nodes and computes their difference nodes, explicitly fusing bitemporal features while highlighting change information—an advancement over traditional BLS. Experiments on five SAR datasets demonstrate the state-of-the-art performance of PC-BiBL, surpassing existing methods in accuracy and robustness. Quantitative metrics and visual analyses confirm its superiority in handling speckle noise and preserving boundary information.
- 1 January 2026
- Published by MDPI AG in Remote Sensing
- Vol. 18 (1) , 143
Open AccessWhen Deep Learning Meets Broad Learning: A Unified Framework for Change Detection with Synthetic Aperture Radar Images
Change detection (CD) with synthetic aperture radar (SAR) images remains pivotal for environmental monitoring and disaster management. Deep learning has powerful feature extraction capabilities for CD, but suffers from complex architectures and limited interpretability. While BLSs demonstrate advantages in structural simplicity and interpretability, their feature representation capacity remains constrained. In high-precision CD with SAR images, strong feature representation capability is required, along with an uncomplicated framework and high interpretability. Therefore, a novel paradigm named PC-BiBL is proposed which achieves seamless integration of deep learning and broad learning. On the one hand, it employs a hierarchical cross-convolutional encoding (HCCE) module that uses pseudo-random cross-convolution (PCConv) for hierarchical cross-feature representation, aggregating contextual information. PCConv is an untrained convolution layer, which can utilize specialized pseudo-random kernels to extract features from bitemporal SAR images. On the other hand, since back-propagation algorithms are not required, the features can be directly fed into the bifurcated broad learning (BiBL) module for node expansion and direct parameter computation. BiBL constructs dual-branch nodes and computes their difference nodes, explicitly fusing bitemporal features while highlighting change information—an advancement over traditional BLS. Experiments on five SAR datasets demonstrate the state-of-the-art performance of PC-BiBL, surpassing existing methods in accuracy and robustness. Quantitative metrics and visual analyses confirm its superiority in handling speckle noise and preserving boundary information.