Longfei Ren
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Longfei Ren has a total of 71 co-authors. Here are the top co-authors by number of shared publications:
- Lianru Gao
Chinese Academy of Sciences
10 shared publications - Jocelyn Chanussot
Univ. Grenoble Alpes, Inria, CNRS, Grenoble INP, LJK
6 shared publications - Minghua Wang
Institute of Robotics and Automatic Information System (IRAIS), College of Artificial Intelligence, and Tianjin Key Laboratory of Intelligent Robotics (tjKLIR), Nankai University, Tianjin, China
6 shared publications - Xu Sun
Key Laboratory of Computational Optical Imaging Technology, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China
6 shared publications - Qian Du
Mississippi State University
5 shared publications - Yang-Jun Deng
College of Information and Intelligence, Hunan Agricultural University, Changsha, China
5 shared publications
Latest Publications
- Open Access
- 21 October 2026
- Published by Elsevier in International Journal of Applied Earth Observation and Geoinformation
- Vol. 154
Large-scale building instance segmentation in diverse urban scenes based on LiDAR data
- 21 October 2026
- Published by Elsevier in International Journal of Applied Earth Observation and Geoinformation
- Vol. 154
Open AccessLarge-scale building instance segmentation in diverse urban scenes based on LiDAR data
- Open Access
- 25 July 2026
- Published by Elsevier in SSRN Electronic Journal
NSAMamba: Neighborhood Self-Attention Mamba Network for Hyperspectral Image Classification
The extraction and fusion of spatial–spectral features are critical for effective hyperspectral image (HSI) classification. However, most existing deep learning methods, including recent Mamba-based architectures, still struggle to adequately integrate localized spatial–spectral information and capture fine-grained dependencies, which limits classification accuracy and robustness. To this end, we propose NSAMamba, a novel encoder-decoder architecture tailored for HSI classification. Specifically, the encoder-decoder structure is designed to achieve fine-grained fusion of global context and local detail through three key innovations. First, a dual-branch attention mechanism is proposed in the encoder to capture localized dependencies efficiently. The spatial branch adaptively selects optimal neighborhoods by balancing structural continuity and contextual relevance, enhancing spatial feature representation. In parallel, the spectral branch employs neighborhood attention to dynamically emphasize informative spectral bands while suppressing redundancy, improving spectral discrimination. Second, the decoder incorporates a feature fusion module based on element-wise gated attention, which aggregates hierarchical features from multiple encoder stages. This module refines cross-scale feature weights, learns complex inter-feature relationships, and strengthens the fusion of global and local representations. Finally, the synergy between the encoder’s localized feature extraction and the decoder’s multi-scale fusion enables comprehensive modeling of spatial–spectral features and optimization of global contextual representation. Extensive experiments on three public HSI datasets demonstrate that NSAMamba consistently outperforms state-of-the-art methods in terms of classification accuracy.
- 25 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessNSAMamba: Neighborhood Self-Attention Mamba Network for Hyperspectral Image Classification
The extraction and fusion of spatial–spectral features are critical for effective hyperspectral image (HSI) classification. However, most existing deep learning methods, including recent Mamba-based architectures, still struggle to adequately integrate localized spatial–spectral information and capture fine-grained dependencies, which limits classification accuracy and robustness. To this end, we propose NSAMamba, a novel encoder-decoder architecture tailored for HSI classification. Specifically, the encoder-decoder structure is designed to achieve fine-grained fusion of global context and local detail through three key innovations. First, a dual-branch attention mechanism is proposed in the encoder to capture localized dependencies efficiently. The spatial branch adaptively selects optimal neighborhoods by balancing structural continuity and contextual relevance, enhancing spatial feature representation. In parallel, the spectral branch employs neighborhood attention to dynamically emphasize informative spectral bands while suppressing redundancy, improving spectral discrimination. Second, the decoder incorporates a feature fusion module based on element-wise gated attention, which aggregates hierarchical features from multiple encoder stages. This module refines cross-scale feature weights, learns complex inter-feature relationships, and strengthens the fusion of global and local representations. Finally, the synergy between the encoder’s localized feature extraction and the decoder’s multi-scale fusion enables comprehensive modeling of spatial–spectral features and optimization of global contextual representation. Extensive experiments on three public HSI datasets demonstrate that NSAMamba consistently outperforms state-of-the-art methods in terms of classification accuracy.
- 16 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 35 (10577149) , 4208-4222
Resolution Preserving and Utilization Network for Tiny Object Detection in Large-Size Remote Sensing Imagery
Efficient tiny object detection (TOD) in large-size remote sensing imagery (LSRSI) is particularly challenging in real-world remote sensing applications. We observe that as the input size of the remote sensing scene increases, TOD faces more severe foreground signal identification issues. To address this, we are the first to design a backbone network from the perspective of low-level spatial feature preservation and utilization, specifically for tiny object feature extraction in large-size remote sensing scene patches. The proposed architecture, referred to as the resolution preserving and utilization network (RPUN), demonstrates excellent foreground tiny object feature response identification ability when increasing the input size of remote sensing scenes, effectively maintaining detection performance comparable to that of smaller input slices. Additionally, we introduce GF2UBSv2, a large-scale panchromatic satellite imagery dataset focused on tiny urban bridge detection. Extensive experiments conducted on GF2UBSv2, DIOR, SODA-A, and DOTAv2.0 demonstrate the superior performance of RPUN compared with state-of-the-art methods. The code and dataset are available at: https: //github.com//NankleZTW.
- 16 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 35 (10577149) , 4208-4222
Resolution Preserving and Utilization Network for Tiny Object Detection in Large-Size Remote Sensing Imagery
Efficient tiny object detection (TOD) in large-size remote sensing imagery (LSRSI) is particularly challenging in real-world remote sensing applications. We observe that as the input size of the remote sensing scene increases, TOD faces more severe foreground signal identification issues. To address this, we are the first to design a backbone network from the perspective of low-level spatial feature preservation and utilization, specifically for tiny object feature extraction in large-size remote sensing scene patches. The proposed architecture, referred to as the resolution preserving and utilization network (RPUN), demonstrates excellent foreground tiny object feature response identification ability when increasing the input size of remote sensing scenes, effectively maintaining detection performance comparable to that of smaller input slices. Additionally, we introduce GF2UBSv2, a large-scale panchromatic satellite imagery dataset focused on tiny urban bridge detection. Extensive experiments conducted on GF2UBSv2, DIOR, SODA-A, and DOTAv2.0 demonstrate the superior performance of RPUN compared with state-of-the-art methods. The code and dataset are available at: https: //github.com//NankleZTW.
- 3 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-16
Fast Anchor Graph Regularized Relaxation Linear Regression for Classification of Hyperspectral Images
Hyperspectral image (HSI) classification remains challenging due to the inherently high dimensionality and complex intrinsic structure of the data. These characteristics pose significant difficulties for traditional linear regression methods, particularly when it comes to capturing nonlinear relationships and preserving local structural information. To address these issues, this paper proposes a novel classification framework called fast anchor graph regularized relaxation linear regression (FAGRLR). The proposed method integrates a fast anchor graph model into linear regression to encode both spatial and structural neighborhood information inherent in HSIs. By utilizing anchor graphs, the method efficiently preserves critical local relationships within the data, enhancing the discriminative ability of the model. To further avoid overfitting, a label relaxation mechanism is introduced to increase the label matrix’s flexibility in discriminative transformation learning. Additionally, the L2,1-norm constraint is incorporated to promote robustness and sparsity in the learned transformation matrix, thereby enhancing the generalization performance of the proposed method. We validate the effectiveness of FAGRLR through extensive experiments conducted on four benchmark hyperspectral datasets. Experimental results demonstrate that the proposed method achieves superior classification performance compared to several state-of-the-art methods.
- 3 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-16
Fast Anchor Graph Regularized Relaxation Linear Regression for Classification of Hyperspectral Images
Hyperspectral image (HSI) classification remains challenging due to the inherently high dimensionality and complex intrinsic structure of the data. These characteristics pose significant difficulties for traditional linear regression methods, particularly when it comes to capturing nonlinear relationships and preserving local structural information. To address these issues, this paper proposes a novel classification framework called fast anchor graph regularized relaxation linear regression (FAGRLR). The proposed method integrates a fast anchor graph model into linear regression to encode both spatial and structural neighborhood information inherent in HSIs. By utilizing anchor graphs, the method efficiently preserves critical local relationships within the data, enhancing the discriminative ability of the model. To further avoid overfitting, a label relaxation mechanism is introduced to increase the label matrix’s flexibility in discriminative transformation learning. Additionally, the L2,1-norm constraint is incorporated to promote robustness and sparsity in the learned transformation matrix, thereby enhancing the generalization performance of the proposed method. We validate the effectiveness of FAGRLR through extensive experiments conducted on four benchmark hyperspectral datasets. Experimental results demonstrate that the proposed method achieves superior classification performance compared to several state-of-the-art methods.
- 16 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-22
Endmember Selection With Adaptive Double Prior Model
Selecting a correct endmember subset from a large number of endmembers plays an important role in improving the performance of spectral unmixing. Sparse unmixing (SU) pro-vides a useful tool to solve this endmember selection problem. However, due to the complexity of endmembers in the spectral library and noise in hyperspectral data, this problem is ill-posed and remains a significant challenge. To address this, prior infor-mation can be employed to mitigate the problem. Existing algo-rithms, however, often overlook the possibility that the prior information may be incomplete or inaccurate, and consider only with a single prior model. To this end, this paper proposes an adaptive method designed to handle incorrect prior information and balance the tradeoff between the contributions of two sets of priors. In contrast to the conventional models, our method dy-namically adjusts optimal decision boundaries by introducing decision-making parameters (DMPs), thereby obtaining a better solution. Optimal weight conditions and robustness analysis are also presented to prove the performance of our method. Fur-thermore, experiments conducted on both simulated and real datasets demonstrate that the proposed method delivers robust and reliable performance, highlighting its effectiveness in ad-dressing the unmixing challenges. The code of our method is available at https://github.com/spdelphi/Double-Prior.
- 16 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-22
Endmember Selection With Adaptive Double Prior Model
Selecting a correct endmember subset from a large number of endmembers plays an important role in improving the performance of spectral unmixing. Sparse unmixing (SU) pro-vides a useful tool to solve this endmember selection problem. However, due to the complexity of endmembers in the spectral library and noise in hyperspectral data, this problem is ill-posed and remains a significant challenge. To address this, prior infor-mation can be employed to mitigate the problem. Existing algo-rithms, however, often overlook the possibility that the prior information may be incomplete or inaccurate, and consider only with a single prior model. To this end, this paper proposes an adaptive method designed to handle incorrect prior information and balance the tradeoff between the contributions of two sets of priors. In contrast to the conventional models, our method dy-namically adjusts optimal decision boundaries by introducing decision-making parameters (DMPs), thereby obtaining a better solution. Optimal weight conditions and robustness analysis are also presented to prove the performance of our method. Fur-thermore, experiments conducted on both simulated and real datasets demonstrate that the proposed method delivers robust and reliable performance, highlighting its effectiveness in ad-dressing the unmixing challenges. The code of our method is available at https://github.com/spdelphi/Double-Prior.
- 4 December 2025
- Published by Elsevier in ISPRS Journal of Photogrammetry and Remote Sensing
- Vol. 232, 94-108
Cross-satellite hierarchical multimodal denoising for hyperspectral imagery
- 4 December 2025
- Published by Elsevier in ISPRS Journal of Photogrammetry and Remote Sensing
- Vol. 232, 94-108
Cross-satellite hierarchical multimodal denoising for hyperspectral imagery
- Open Access
- 8 November 2025
- Published by Elsevier in Information Geography
- Vol. 2 (1)
Advances in hyperspectral image unmixing: From algorithmic frameworks to practical applications
- 8 November 2025
- Published by Elsevier in Information Geography
- Vol. 2 (1)
Open AccessAdvances in hyperspectral image unmixing: From algorithmic frameworks to practical applications
- 5 October 2025
- Published by Elsevier in Pattern Recognition
- Vol. 172
HADDNLP: Hyperspectral anomaly detection via double nonlocal priors
- 5 October 2025
- Published by Elsevier in Pattern Recognition
- Vol. 172
HADDNLP: Hyperspectral anomaly detection via double nonlocal priors