Meng Xu
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
Meng Xu has a total of 65 co-authors. Here are the top co-authors by number of shared publications:
- Sen Jia
College of Computer Science and Software Engineering, Shenzhen University, Shenzhen, China
31 shared publications - Xiuping Jia
University of New South Wales
15 shared publications - Jun Zhou
Griffith University
8 shared publications - Qingquan Li
Shenzhen Key Laboratory of Spatial Smart Sensing and Services, Shenzhen University, Shenzhen 518060, China
7 shared publications - Shuyu Zhang
College of Computer Science and Software Engineering, Shenzhen University, Shenzhen, China
7 shared publications - Xiyou Fu
College of Computer Science and Software Engineering, Shenzhen University, Shenzhen, China
7 shared publications
Latest Publications
- Open Access
- 21 June 2026
- Published by Taylor & Francis in GIScience & Remote Sensing
- Vol. 63  (1)
PLCSeg: sar-guided progressive learning for semantic segmentation of cloudy remote sensing images
- 21 June 2026
- Published by Taylor & Francis in GIScience & Remote Sensing
- Vol. 63  (1)
Open AccessPLCSeg: sar-guided progressive learning for semantic segmentation of cloudy remote sensing images
- 12 March 2026
- Published by Elsevier in Information Fusion
- Vol. 133
NAM-CR: Noise-aware multisensor cloud removal for optical remote sensing imagery by fusing SAR data
- 12 March 2026
- Published by Elsevier in Information Fusion
- Vol. 133
NAM-CR: Noise-aware multisensor cloud removal for optical remote sensing imagery by fusing SAR data
- 13 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64  (01962892) , 1-17
DAHNet: A Decomposition and Aggregation Hybrid Network for Hyperspectral Object Detection
Hyperspectral object detection is important in hyperspectral image analysis. However, it remains challenging because of spectral redundancy and feature coupling. Current studies are limited and lack systematic and general detection frameworks. To address this issue, we propose a hybrid spectral– spatial network named the decomposition and aggregation hybrid network (DAHNet), designed to improve the accuracy and efficiency of hyperspectral object detection. DAHNet begins with the spectral–spatial compression and enhancement (SSCE) module, which combines depthwise separable and 3D convolutions to preserve spectral information while capturing key spatial structures, ensuring compact yet discriminative representations. This is followed by feature refinement through the 3D spectral bottleneck enhancement (3DSBE) module and the scalable spatial bottleneck refinement (SSBR) module, which progressively extract complementary spectral and spatial cues. To effectively integrate the refined features, the cross-domain synergistic fusion (CDSF) module employs an attention mechanism to align spectral and spatial representations, enhancing feature complementarity and semantic consistency. To validate the proposed approach, we constructed an aerial hyperspectral dataset, SZU_CAR, and conducted extensive experiments on two additional public datasets. Experimental results demonstrate that DAHNet consistently outperforms existing methods, achieving mean average precisions (mAPs) of 61.5%, 60.5%, and 27.2% on the SZU_CAR, HOD3K, and M2SODAI datasets, respectively, with consistent performance gains over representative state-of-the-art approaches. The source code is available at http://szu-hsilab.com/.
- 13 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64  (01962892) , 1-17
DAHNet: A Decomposition and Aggregation Hybrid Network for Hyperspectral Object Detection
Hyperspectral object detection is important in hyperspectral image analysis. However, it remains challenging because of spectral redundancy and feature coupling. Current studies are limited and lack systematic and general detection frameworks. To address this issue, we propose a hybrid spectral– spatial network named the decomposition and aggregation hybrid network (DAHNet), designed to improve the accuracy and efficiency of hyperspectral object detection. DAHNet begins with the spectral–spatial compression and enhancement (SSCE) module, which combines depthwise separable and 3D convolutions to preserve spectral information while capturing key spatial structures, ensuring compact yet discriminative representations. This is followed by feature refinement through the 3D spectral bottleneck enhancement (3DSBE) module and the scalable spatial bottleneck refinement (SSBR) module, which progressively extract complementary spectral and spatial cues. To effectively integrate the refined features, the cross-domain synergistic fusion (CDSF) module employs an attention mechanism to align spectral and spatial representations, enhancing feature complementarity and semantic consistency. To validate the proposed approach, we constructed an aerial hyperspectral dataset, SZU_CAR, and conducted extensive experiments on two additional public datasets. Experimental results demonstrate that DAHNet consistently outperforms existing methods, achieving mean average precisions (mAPs) of 61.5%, 60.5%, and 27.2% on the SZU_CAR, HOD3K, and M2SODAI datasets, respectively, with consistent performance gains over representative state-of-the-art approaches. The source code is available at http://szu-hsilab.com/.
- 9 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64  (01962892) , 1-21
Tensor-Based Unified Geometry and Reflectance Optimization for 3-D Reconstruction in Large-Scale Urban Scenes
Urban-scale 3-D reconstruction presents a significant challenge due to its complex geometry and diverse material properties. Existing methods struggle to handle this complexity. Neural radiance fields (NeRFs) often produce blurry results, while hybrid approaches based on scene tiling not only impose high computational costs but also lead to spatial inconsistencies. We introduce UniUrban3D, a scalable framework that unifies geometry and appearance modeling within a holistic hierarchical hash-encoded tensor representation combined with dual-path rendering strategies. Our approach leverages an adaptive spatial decomposition and feature aggregation to handle scene complexity, a multiresolution hash-encoded tensor representation that efficiently captures fine geometric details and appearance attributes, and a dual-path rendering pipeline to fuse implicit and explicit representations for enhanced photorealism. Experiments on large-scale datasets demonstrate that UniUrban3D significantly outperforms state-of-the-art methods in reconstruction accuracy and visual fidelity. Our method produces sharper and artifact-free renderings while requiring substantially less memory. Our work enables realistic and scalable city modeling, opening new avenues for applications in virtual reality, digital twins, and urban simulation.
- 9 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64  (01962892) , 1-21
Tensor-Based Unified Geometry and Reflectance Optimization for 3-D Reconstruction in Large-Scale Urban Scenes
Urban-scale 3-D reconstruction presents a significant challenge due to its complex geometry and diverse material properties. Existing methods struggle to handle this complexity. Neural radiance fields (NeRFs) often produce blurry results, while hybrid approaches based on scene tiling not only impose high computational costs but also lead to spatial inconsistencies. We introduce UniUrban3D, a scalable framework that unifies geometry and appearance modeling within a holistic hierarchical hash-encoded tensor representation combined with dual-path rendering strategies. Our approach leverages an adaptive spatial decomposition and feature aggregation to handle scene complexity, a multiresolution hash-encoded tensor representation that efficiently captures fine geometric details and appearance attributes, and a dual-path rendering pipeline to fuse implicit and explicit representations for enhanced photorealism. Experiments on large-scale datasets demonstrate that UniUrban3D significantly outperforms state-of-the-art methods in reconstruction accuracy and visual fidelity. Our method produces sharper and artifact-free renderings while requiring substantially less memory. Our work enables realistic and scalable city modeling, opening new avenues for applications in virtual reality, digital twins, and urban simulation.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64, 4706213
STAR: Spatio-Temporal Alignment and Refinement for Cross-View Geo-Localization
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64, 4706213
STAR: Spatio-Temporal Alignment and Refinement for Cross-View Geo-Localization
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. PP, 1-15
TMMSAM2: Tracker-Aided Multitemporal Memory SAM2 for Hyperspectral Object Tracking
Segment Anything Model 2 (SAM2) demonstrates outstanding performance in prompt-based visual segmentation. However, directly applying it to hyperspectral object-tracking tasks still faces numerous challenges. This article proposes TMMSAM2, a novel SAM2-based framework for hyperspectral object tracking that requires no additional training, comprising a multitemporal memory bank (MMB) and a three-dimensional constraint error mechanism (TCEM). Specifically, the MMB overcomes limitations of traditional strategies based on fixed historical frames by collaboratively integrating memory information across short-, medium-, and long-term temporal scales. Combined with feature similarity calculations and redundant frame filtering mechanisms, it constructs historical memory with optimal spatiotemporal diversity. The TCEM implements real-time quality assessment of tracking results by modeling physical constraints in three dimensions: velocity, size, and motion direction. When anomalies are detected, the system automatically triggers correction using a ViPT-based hyperspectral tracker, thereby establishing an end-to-end tracking quality assurance system. Extensive experiments on three public datasets demonstrate that our method exhibits superior performance in complex scenarios involving rapid target movement and frequent occlusions.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. PP, 1-15
TMMSAM2: Tracker-Aided Multitemporal Memory SAM2 for Hyperspectral Object Tracking
Segment Anything Model 2 (SAM2) demonstrates outstanding performance in prompt-based visual segmentation. However, directly applying it to hyperspectral object-tracking tasks still faces numerous challenges. This article proposes TMMSAM2, a novel SAM2-based framework for hyperspectral object tracking that requires no additional training, comprising a multitemporal memory bank (MMB) and a three-dimensional constraint error mechanism (TCEM). Specifically, the MMB overcomes limitations of traditional strategies based on fixed historical frames by collaboratively integrating memory information across short-, medium-, and long-term temporal scales. Combined with feature similarity calculations and redundant frame filtering mechanisms, it constructs historical memory with optimal spatiotemporal diversity. The TCEM implements real-time quality assessment of tracking results by modeling physical constraints in three dimensions: velocity, size, and motion direction. When anomalies are detected, the system automatically triggers correction using a ViPT-based hyperspectral tracker, thereby establishing an end-to-end tracking quality assurance system. Extensive experiments on three public datasets demonstrate that our method exhibits superior performance in complex scenarios involving rapid target movement and frequent occlusions.
- 18 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 63  (01962892) , 1-16
SAGT: Structure-Adaptive Graph Transformer for Hyperspectral Image Classification
Hyperspectral images (HSIs) are vital for scene analysis, as they capture detailed spatial and spectral information to characterize surface materials. However, accurate HSI classification is challenged by significant intra-class spectral variability and spatial complexity. To address this, we leverage the fact that pixels of the same class typically form irregular local regions. We propose a structure-adaptive graph transformer (SAGT) that dynamically captures irregular spatial topologies and homogeneous spectral information to achieve adaptive HSI representation and precise classification. Specifically, a structure-aware self-attention (SASA) module is developed to embed graph structures into the self-attention mechanism as a robust positional indicator, which can be extended easily and effectively. SASA comprehensively accounts for the spatial structures and spectral autocorrelation of ground objects, facilitating the aggregation of homogeneous spectral information for noise-robust spectral representations. Additionally, a structure-adaptive pooling (SAP) module is designed to dynamically adjust graph structures by discarding irrelevant edges, thus better indicating spatial relationships. By coupling the SASA and SAP modules, our proposed SAGT model significantly alleviates spectral variability and tolerates prior noise. Furthermore, data augmentation techniques of random discard and random offset are built, which randomly drop and shift graph nodes to generate more diverse samples during preprocessing. In postprocessing, multiview decision-making integrates results from multiple contextual views to provide more robust predictions. Experimental results on three benchmark datasets consistently demonstrate that SAGT is more effective and reliable than other state-of-the-art methods. To facilitate reproduction, we will release the source code for SAGT at https://github.com/ShuGuoJ/SAGT.git.
- 18 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 63  (01962892) , 1-16
SAGT: Structure-Adaptive Graph Transformer for Hyperspectral Image Classification
Hyperspectral images (HSIs) are vital for scene analysis, as they capture detailed spatial and spectral information to characterize surface materials. However, accurate HSI classification is challenged by significant intra-class spectral variability and spatial complexity. To address this, we leverage the fact that pixels of the same class typically form irregular local regions. We propose a structure-adaptive graph transformer (SAGT) that dynamically captures irregular spatial topologies and homogeneous spectral information to achieve adaptive HSI representation and precise classification. Specifically, a structure-aware self-attention (SASA) module is developed to embed graph structures into the self-attention mechanism as a robust positional indicator, which can be extended easily and effectively. SASA comprehensively accounts for the spatial structures and spectral autocorrelation of ground objects, facilitating the aggregation of homogeneous spectral information for noise-robust spectral representations. Additionally, a structure-adaptive pooling (SAP) module is designed to dynamically adjust graph structures by discarding irrelevant edges, thus better indicating spatial relationships. By coupling the SASA and SAP modules, our proposed SAGT model significantly alleviates spectral variability and tolerates prior noise. Furthermore, data augmentation techniques of random discard and random offset are built, which randomly drop and shift graph nodes to generate more diverse samples during preprocessing. In postprocessing, multiview decision-making integrates results from multiple contextual views to provide more robust predictions. Experimental results on three benchmark datasets consistently demonstrate that SAGT is more effective and reliable than other state-of-the-art methods. To facilitate reproduction, we will release the source code for SAGT at https://github.com/ShuGuoJ/SAGT.git.
- 8 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 63  (01962892) , 1-15
DiLAST: Leveraging Differential RGB Features for Hyperspectral Image Super-Resolution
Hyperspectral image super-resolution aims to reconstruct high-quality spatial-spectral cubes from RGB images. However, the limited spectral coverage of RGB inputs hinders the simultaneous modeling of spatial structures and spectral relationships. To address this limitation, we propose a differential low-rank adaptive spatial-spectral transformer (DiLAST). Initially, a differential operator is employed to enhance RGB features in a bottom-up manner, explicitly amplifying subtle inter-channel differences. The enhanced features are then fed into a U-shaped backbone, which integrates three complementary modules for joint spatial-spectral modeling. Specifically, a center spatial-spectral attention (CSSA) module employs cross-attention mechanisms to capture local-to-global dependencies across both spatial and spectral domains; an adaptive cross-scale fusion (ACF) module utilizes learnable gating weights to establish dynamic interaction pathways between shallow high-frequency details and deep semantic representations; and a low-rank spectral calibration (LRSC) module exploits low-rank matrix priors to reveal low-dimensional manifold structures among spectral bands, thereby enhancing spectral consistency. By leveraging the synergistic effects of spatial non-locality, global spectral correlation, and low-rank properties, the proposed DiLAST achieves PSNR improvements of 33.82 dB, 36.03 dB, and 37.01 dB on benchmark datasets. Moreover, the accuracy and practical applicability of the reconstructed spectra have been effectively validated in remote sensing scenarios and object tracking tasks. The code is accessible at https://github.com/renqi1998/DiLAST.
- 8 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 63  (01962892) , 1-15
DiLAST: Leveraging Differential RGB Features for Hyperspectral Image Super-Resolution
Hyperspectral image super-resolution aims to reconstruct high-quality spatial-spectral cubes from RGB images. However, the limited spectral coverage of RGB inputs hinders the simultaneous modeling of spatial structures and spectral relationships. To address this limitation, we propose a differential low-rank adaptive spatial-spectral transformer (DiLAST). Initially, a differential operator is employed to enhance RGB features in a bottom-up manner, explicitly amplifying subtle inter-channel differences. The enhanced features are then fed into a U-shaped backbone, which integrates three complementary modules for joint spatial-spectral modeling. Specifically, a center spatial-spectral attention (CSSA) module employs cross-attention mechanisms to capture local-to-global dependencies across both spatial and spectral domains; an adaptive cross-scale fusion (ACF) module utilizes learnable gating weights to establish dynamic interaction pathways between shallow high-frequency details and deep semantic representations; and a low-rank spectral calibration (LRSC) module exploits low-rank matrix priors to reveal low-dimensional manifold structures among spectral bands, thereby enhancing spectral consistency. By leveraging the synergistic effects of spatial non-locality, global spectral correlation, and low-rank properties, the proposed DiLAST achieves PSNR improvements of 33.82 dB, 36.03 dB, and 37.01 dB on benchmark datasets. Moreover, the accuracy and practical applicability of the reconstructed spectra have been effectively validated in remote sensing scenarios and object tracking tasks. The code is accessible at https://github.com/renqi1998/DiLAST.
- 14 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 63  (01962892) , 1-16
Enhanced Spatial-Frequency Synergistic Network for Multispectral and Hyperspectral Image Fusion
Multispectral and hyperspectral image fusion (MHIF) seeks to combine high-resolution multispectral images (HR-MSIs) with low-resolution hyperspectral images (LR-HSIs) to create high-resolution hyperspectral images (HR-HSIs). Transformer-based architectures have recently become prominent in MHIF tasks due to their effective global self-attention mechanisms. However, the quadratic computational complexity of the global self-attention in Transformers presents significant challenges for practical applications. In this paper, we propose an enhanced spatial-frequency synergistic (ESFS) approach that leverages both spatial and frequency domain features to enhance fusion quality. Our ESFS framework introduces the condensed spatial augmentation module (CSAM), which condenses window features and employs cross-attention to balance extensive contextual understanding and detailed local feature extraction while reducing computational overhead. Additionally, we develop the selective frequency decomposition module (SFDM), which utilizes global filters composed of phase and amplitude information in the frequency domain to retain features, effectively capturing deep frequency domain characteristics and their interdependencies. Comprehensive experiments on three benchmark MHIF datasets demonstrate that our method achieves superior performance, establishing a new state-of-the-art (SOTA) in both quantitative metrics and visual quality assessments. The code is available at http://szu-hsilab.com/.
- 14 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 63  (01962892) , 1-16
Enhanced Spatial-Frequency Synergistic Network for Multispectral and Hyperspectral Image Fusion
Multispectral and hyperspectral image fusion (MHIF) seeks to combine high-resolution multispectral images (HR-MSIs) with low-resolution hyperspectral images (LR-HSIs) to create high-resolution hyperspectral images (HR-HSIs). Transformer-based architectures have recently become prominent in MHIF tasks due to their effective global self-attention mechanisms. However, the quadratic computational complexity of the global self-attention in Transformers presents significant challenges for practical applications. In this paper, we propose an enhanced spatial-frequency synergistic (ESFS) approach that leverages both spatial and frequency domain features to enhance fusion quality. Our ESFS framework introduces the condensed spatial augmentation module (CSAM), which condenses window features and employs cross-attention to balance extensive contextual understanding and detailed local feature extraction while reducing computational overhead. Additionally, we develop the selective frequency decomposition module (SFDM), which utilizes global filters composed of phase and amplitude information in the frequency domain to retain features, effectively capturing deep frequency domain characteristics and their interdependencies. Comprehensive experiments on three benchmark MHIF datasets demonstrate that our method achieves superior performance, establishing a new state-of-the-art (SOTA) in both quantitative metrics and visual quality assessments. The code is available at http://szu-hsilab.com/.
- 19 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 62  (01962892) , 1-15
SQformer: Spectral-Query Transformer for Hyperspectral Image Arbitrary-Scale Super-Resolution
Super-resolution is vital for the quality improvement of hyperspectral images (HSIs) under the spatial and spectral resolution trade-off. However, deep learning HSI super-resolution approaches typically adopt the "one model, one scale" scheme that is inefficient in training and storing. This is difficult in maximizing orbit equipment performance and aligning multiple spatial resolution data in remote sensing. So, this paper intends to address HSI arbitrary-scale super-resolution, enabling the scaling of HSIs to arbitrary sizes using a single model. To do this end, we treat HSI arbitrary-scale super-resolution as a retrieval problem. It conceptualizes the HSI as a dictionary of pixel-wise tokens with spatial-spectral features, position information, and scale information. Its objective is to employ a set of initialized tokens related to the high-resolution (HR) HSI as queries to retrieve matched spectral features from low-resolution (LR) one, which is so-called token-based query-to-spectrum. Since these query tokens can be constructed flexibly (e.g., through random initialization), we can generate a desired number of them to reconstruct our HR HSI, thus achieving arbitrary-scale super-resolution. This process considers not only position information but also spectral features so that it can decrease spectral distortion. With the above idea, we developed a HSI arbitrary-scale super-resolution method, dubbed as Spectral-Query transformer (SQformer). Specifically, it begins by converting the LR HSI into a dictionary of LR tokens and then constructs a desired number of HR tokens. To enable flexible token construction, we design an implicit spectral token (particularly a learnable vector) and replicate it α H × α W times to form the HR tokens. Next, the HR and LR tokens are passed into a transformer decoder to find the most matched spectral response for the former by soft-weighting the LR tokens. Finally, the HR tokens are spatially rearranged in order, forming a HR HSI. Extensive experiments have demonstrated its effectiveness on remote sensing data. The code will be released at: https://github.com/ShuGuoJ/SQformer.git .
- 19 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 62  (01962892) , 1-15
SQformer: Spectral-Query Transformer for Hyperspectral Image Arbitrary-Scale Super-Resolution
Super-resolution is vital for the quality improvement of hyperspectral images (HSIs) under the spatial and spectral resolution trade-off. However, deep learning HSI super-resolution approaches typically adopt the "one model, one scale" scheme that is inefficient in training and storing. This is difficult in maximizing orbit equipment performance and aligning multiple spatial resolution data in remote sensing. So, this paper intends to address HSI arbitrary-scale super-resolution, enabling the scaling of HSIs to arbitrary sizes using a single model. To do this end, we treat HSI arbitrary-scale super-resolution as a retrieval problem. It conceptualizes the HSI as a dictionary of pixel-wise tokens with spatial-spectral features, position information, and scale information. Its objective is to employ a set of initialized tokens related to the high-resolution (HR) HSI as queries to retrieve matched spectral features from low-resolution (LR) one, which is so-called token-based query-to-spectrum. Since these query tokens can be constructed flexibly (e.g., through random initialization), we can generate a desired number of them to reconstruct our HR HSI, thus achieving arbitrary-scale super-resolution. This process considers not only position information but also spectral features so that it can decrease spectral distortion. With the above idea, we developed a HSI arbitrary-scale super-resolution method, dubbed as Spectral-Query transformer (SQformer). Specifically, it begins by converting the LR HSI into a dictionary of LR tokens and then constructs a desired number of HR tokens. To enable flexible token construction, we design an implicit spectral token (particularly a learnable vector) and replicate it α H × α W times to form the HR tokens. Next, the HR and LR tokens are passed into a transformer decoder to find the most matched spectral response for the former by soft-weighting the LR tokens. Finally, the HR tokens are spatially rearranged in order, forming a HR HSI. Extensive experiments have demonstrated its effectiveness on remote sensing data. The code will be released at: https://github.com/ShuGuoJ/SQformer.git .