Xudong Yao
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
Xudong Yao has a total of 12 co-authors. Here are the top co-authors by number of shared publications:
- Haopeng Zhang
Beihang University
4 shared publications - Zhenwei Shi
Beihang University
4 shared publications - Sizhe Wen
Department of Aerospace Information Engineering (Image Processing Center), School of Astronautics, Beihang University, Beijing, China
3 shared publications - Zhiguo Jiang
Beihang University
3 shared publications - Fengying Xie
Beihang University
2 shared publications - Qing Guo
Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing, China
2 shared publications
Latest Publications
- 17 June 2026
- Published by Elsevier in Aerospace Science and Technology
- Vol. 177
A hybrid cross-modal fusion framework for satellite attitude estimation
- 17 June 2026
- Published by Elsevier in Aerospace Science and Technology
- Vol. 177
A hybrid cross-modal fusion framework for satellite attitude estimation
- Top Cited Papers
- 29 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-16
Uncertainty Estimation Optimization for Reliable Remote Sensing Single-Image Super-Resolution
Remote sensing(RS) single-image super-resolution (SISR) is a classic ill-posed problem in low-level vision, characterized by a one-to-many mapping. This ill-posedness can lead to aleatoric uncertainty, thereby rendering the reconstruction results unreliable. Given that remote sensing images typically carry strict physical significance, quantifying uncertainty to evaluate pixel-wise reliability in RSSISR results is of critical importance. In this paper, we estimate image uncertainty based on differential entropy under a given assumed data distribution, and propose a general framework that integrates uncertainty into the loss function for SISR. Within this framework, we derive various uncertainty-aware loss functions under commonly assumed distributions. However, the learning of uncertainty in these loss functions is performed in an unsupervised manner, which frequently results in instability during training. To address this, we design an uncertainty estimation branch incorporating a sparse prior and introduce a similarity measure based on perceptual space features, thereby enhancing the stability of the uncertainty learning process. To mitigate the issue of loss attenuation caused by assigning lower weights to high-uncertainty pixels during uncertainty learning, we introduce an uncertainty-guided loss function after the uncertainty estimation stage. This loss function assigns greater importance to high-uncertainty pixels, thereby enhancing the reconstruction quality and reliability of the SISR results. Our proposed framework can be seamlessly integrated into popular SISR networks, enabling the simultaneous generation of high-quality SR images and reliable uncertainty maps. Extensive experiments demonstrate that our method outperforms existing state-of-the-art approaches in terms of reconstruction accuracy and perceptual quality, while simultaneously preserving meaningful uncertainty estimates. The corresponding code is available at: https://github.com/yaoxudong241/UEO.
- 29 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-16
Top Cited PapersUncertainty Estimation Optimization for Reliable Remote Sensing Single-Image Super-Resolution
Remote sensing(RS) single-image super-resolution (SISR) is a classic ill-posed problem in low-level vision, characterized by a one-to-many mapping. This ill-posedness can lead to aleatoric uncertainty, thereby rendering the reconstruction results unreliable. Given that remote sensing images typically carry strict physical significance, quantifying uncertainty to evaluate pixel-wise reliability in RSSISR results is of critical importance. In this paper, we estimate image uncertainty based on differential entropy under a given assumed data distribution, and propose a general framework that integrates uncertainty into the loss function for SISR. Within this framework, we derive various uncertainty-aware loss functions under commonly assumed distributions. However, the learning of uncertainty in these loss functions is performed in an unsupervised manner, which frequently results in instability during training. To address this, we design an uncertainty estimation branch incorporating a sparse prior and introduce a similarity measure based on perceptual space features, thereby enhancing the stability of the uncertainty learning process. To mitigate the issue of loss attenuation caused by assigning lower weights to high-uncertainty pixels during uncertainty learning, we introduce an uncertainty-guided loss function after the uncertainty estimation stage. This loss function assigns greater importance to high-uncertainty pixels, thereby enhancing the reconstruction quality and reliability of the SISR results. Our proposed framework can be seamlessly integrated into popular SISR networks, enabling the simultaneous generation of high-quality SR images and reliable uncertainty maps. Extensive experiments demonstrate that our method outperforms existing state-of-the-art approaches in terms of reconstruction accuracy and perceptual quality, while simultaneously preserving meaningful uncertainty estimates. The corresponding code is available at: https://github.com/yaoxudong241/UEO.
- 23 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 23 (1545598X) , 1-5
Activation Map-Based Visual Explanation for Remote Sensing Image Super-Resolution
Deep learning has enhanced image super-resolution (SR) performance, yet SR networks are often treated as black boxes with limited explainability. Notably, remote sensing(RS) images possess explicit physical semantics, imposing stringent demands on the explainability of their SR methods. To address this, we propose AMVE-SR, a framework that systematically transfers activation map-based visual explanation (AMVE) from classification networks to SR networks. Distinct from classification-oriented methods (e.g., Grad-CAM) that struggle with shallow-layer visualization or computationally intensive perturbation-based approaches, AMVE-SR introduces a layer-wise recursive propagation mechanism. By exploiting inter-layer correlations, it traces activation maps backward to the input, ensuring high-fidelity explanations even in shallow layers where conventional methods fail to transfer effectively. Furthermore, we compute gradient responses via textural discrepancy rather than raw pixel intensities, capturing high-frequency details (e.g., physical boundaries) essential for physical semantics. To address the lack of quantitative metrics, we propose Average Texture Variation (ATV) by adapting the Average Drop concept in classification tasks. Validated by visual consistency, ATV serves as a novel fidelity metric for SR explanations. Experiments on the AID dataset using the EDSR model demonstrate that our method produces visually reliable explanation maps while achieving superior performance on the proposed ATV metric with high running efficiency. In general, this study extends AMVE from image classification to SR, thereby establishing a new paradigm to explain SR networks.
- 23 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 23 (1545598X) , 1-5
Activation Map-Based Visual Explanation for Remote Sensing Image Super-Resolution
Deep learning has enhanced image super-resolution (SR) performance, yet SR networks are often treated as black boxes with limited explainability. Notably, remote sensing(RS) images possess explicit physical semantics, imposing stringent demands on the explainability of their SR methods. To address this, we propose AMVE-SR, a framework that systematically transfers activation map-based visual explanation (AMVE) from classification networks to SR networks. Distinct from classification-oriented methods (e.g., Grad-CAM) that struggle with shallow-layer visualization or computationally intensive perturbation-based approaches, AMVE-SR introduces a layer-wise recursive propagation mechanism. By exploiting inter-layer correlations, it traces activation maps backward to the input, ensuring high-fidelity explanations even in shallow layers where conventional methods fail to transfer effectively. Furthermore, we compute gradient responses via textural discrepancy rather than raw pixel intensities, capturing high-frequency details (e.g., physical boundaries) essential for physical semantics. To address the lack of quantitative metrics, we propose Average Texture Variation (ATV) by adapting the Average Drop concept in classification tasks. Validated by visual consistency, ATV serves as a novel fidelity metric for SR explanations. Experiments on the AID dataset using the EDSR model demonstrate that our method produces visually reliable explanation maps while achieving superior performance on the proposed ATV metric with high running efficiency. In general, this study extends AMVE from image classification to SR, thereby establishing a new paradigm to explain SR networks.
- Open Access
- 22 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 18 (19391404) , 1302-1316
Single-Image Superresolution for RGB Remote Sensing Imagery via Multiscale CNN-Transformer Feature Fusion
Single-image superresolution (SISR) of remote sensing images aims to improve image resolution through algorithmic means while restoring rich high-frequency detailed information. Previously, convolutional neural network (CNN) achieves impressive progress in SISR due to its strong local feature extraction capability. However, the CNN is difficult to model long-range dependencies, which limits the performance of SISR. Transformer has a greater advantage over the CNN in long-range dependence modeling. Moreover, remote sensing elements typically exhibit multiscale characteristics and maintain strong coupling relationships with the surrounding environment. Paying attention to multiscale local and global information can help further improve the performance of SISR. Existing methods encounter challenges in leveraging diverse feature sets within single architecture networks. In this article, we propose a multiscale CNN-Transformer feature fusion (MSCT) network for remote sensing SISR. MSCT is a hybrid structural network, which consists of the adaptive residual dense block (ARDB), the multiscale Transformer block (MSTB), and the local–global information enhancement block (LGEB). Among them, ARDB can extract abundant multiscale local features via dense connected convolutional layers and adaptive residual learning. MSTB generates multiscale tokens with multiple convolutional layers to obtain multiscale global information. LGEB further enhance the extraction of local–global information in spatial and frequency domain. Extensive experiments on multiple datasets show that MSCT achieves competitive results on RGB remote sensing imagery compared with other representative SISR methods. In addition, SR experiments on Sentinel-2 images show that MSCT has a certain generalization ability on multispectral images. Extended experiments also demonstrate that MSCT can effectively improve the performance of object detection.
- 22 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 18 (19391404) , 1302-1316
Open AccessSingle-Image Superresolution for RGB Remote Sensing Imagery via Multiscale CNN-Transformer Feature Fusion
Single-image superresolution (SISR) of remote sensing images aims to improve image resolution through algorithmic means while restoring rich high-frequency detailed information. Previously, convolutional neural network (CNN) achieves impressive progress in SISR due to its strong local feature extraction capability. However, the CNN is difficult to model long-range dependencies, which limits the performance of SISR. Transformer has a greater advantage over the CNN in long-range dependence modeling. Moreover, remote sensing elements typically exhibit multiscale characteristics and maintain strong coupling relationships with the surrounding environment. Paying attention to multiscale local and global information can help further improve the performance of SISR. Existing methods encounter challenges in leveraging diverse feature sets within single architecture networks. In this article, we propose a multiscale CNN-Transformer feature fusion (MSCT) network for remote sensing SISR. MSCT is a hybrid structural network, which consists of the adaptive residual dense block (ARDB), the multiscale Transformer block (MSTB), and the local–global information enhancement block (LGEB). Among them, ARDB can extract abundant multiscale local features via dense connected convolutional layers and adaptive residual learning. MSTB generates multiscale tokens with multiple convolutional layers to obtain multiscale global information. LGEB further enhance the extraction of local–global information in spatial and frequency domain. Extensive experiments on multiple datasets show that MSCT achieves competitive results on RGB remote sensing imagery compared with other representative SISR methods. In addition, SR experiments on Sentinel-2 images show that MSCT has a certain generalization ability on multispectral images. Extended experiments also demonstrate that MSCT can effectively improve the performance of object detection.
- Open Access
- 2 August 2022
- Published by MDPI AG in Remote Sensing
- Vol. 14 (15) , 3701
CD_HIEFNet: Cloud Detection Network Using Haze Optimized Transformation Index and Edge Feature for Optical Remote Sensing Imagery
Clouds in optical remote sensing images are an unavoidable existence that greatly affect the utilization of these images. Therefore, accurate and effective cloud detection is an indispensable step in image preprocessing. To date, most researchers have tried to use deep-learning methods for cloud detection. However, these studies generally use computer vision technology to improve the performances of the models, without considering the unique spectral feature information in remote sensing images. Moreover, due to the complex and changeable shapes of clouds, accurate cloud-edge detection is also a difficult problem. In order to solve these problems, we propose a deep-learning cloud detection network that uses the haze-optimized transformation (HOT) index and the edge feature extraction module for optical remote sensing images (CD_HIEFNet). In our model, the HOT index feature image is used to add the unique spectral feature information from clouds into the network for accurate detection, and the edge feature extraction (EFE) module is employed to refine cloud edges. In addition, we use ConvNeXt as the backbone network, and we improved the decoder to enhance the details of the detection results. We validated CD_HIEFNet using the Landsat-8 (L8) Biome dataset and compared it with the Fmask, FCN8s, U-Net, SegNet, DeepLabv3+ and CloudNet methods. The experimental results showed that our model has excellent performance, even in complex cloud scenarios. Moreover, according to the extended experimental results for the other L8 dataset and the Gaofen-1 data, CD_HIEFNet has strong performance in terms of robustness and generalization, thus helping to provide new ideas for cloud detection-related work.
- 2 August 2022
- Published by MDPI AG in Remote Sensing
- Vol. 14 (15) , 3701
Open AccessCD_HIEFNet: Cloud Detection Network Using Haze Optimized Transformation Index and Edge Feature for Optical Remote Sensing Imagery
Clouds in optical remote sensing images are an unavoidable existence that greatly affect the utilization of these images. Therefore, accurate and effective cloud detection is an indispensable step in image preprocessing. To date, most researchers have tried to use deep-learning methods for cloud detection. However, these studies generally use computer vision technology to improve the performances of the models, without considering the unique spectral feature information in remote sensing images. Moreover, due to the complex and changeable shapes of clouds, accurate cloud-edge detection is also a difficult problem. In order to solve these problems, we propose a deep-learning cloud detection network that uses the haze-optimized transformation (HOT) index and the edge feature extraction module for optical remote sensing images (CD_HIEFNet). In our model, the HOT index feature image is used to add the unique spectral feature information from clouds into the network for accurate detection, and the edge feature extraction (EFE) module is employed to refine cloud edges. In addition, we use ConvNeXt as the backbone network, and we improved the decoder to enhance the details of the detection results. We validated CD_HIEFNet using the Landsat-8 (L8) Biome dataset and compared it with the Fmask, FCN8s, U-Net, SegNet, DeepLabv3+ and CloudNet methods. The experimental results showed that our model has excellent performance, even in complex cloud scenarios. Moreover, according to the extended experimental results for the other L8 dataset and the Gaofen-1 data, CD_HIEFNet has strong performance in terms of robustness and generalization, thus helping to provide new ideas for cloud detection-related work.
- Open Access
- 10 September 2021
- Published by MDPI AG in Remote Sensing
- Vol. 13 (18) , 3617
Light-Weight Cloud Detection Network for Optical Remote Sensing Images with Attention-Based DeeplabV3+ Architecture
Clouds in optical remote sensing images cause spectral information change or loss, that affects image analysis and application. Therefore, cloud detection is of great significance. However, there are some shortcomings in current methods, such as the insufficient extendibility due to using the information of multiple bands, the intense extendibility due to relying on some manually determined thresholds, and the limited accuracy, especially for thin clouds or complex scenes caused by low-level manual features. Combining the above shortcomings and the requirements for efficiency in practical applications, we propose a light-weight deep learning cloud detection network based on DeeplabV3+ architecture and channel attention module (CD-AttDLV3+), only using the most common red–green–blue and near-infrared bands. In the CD-AttDLV3+ architecture, an optimized backbone network-MobileNetV2 is used to reduce the number of parameters and calculations. Atrous spatial pyramid pooling effectively reduces the information loss caused by multiple down-samplings while extracting multi-scale features. CD-AttDLV3+ concatenates more low-level features than DeeplabV3+ to improve the cloud boundary quality. The channel attention module is introduced to strengthen the learning of important channels and improve the training efficiency. Moreover, the loss function is improved to alleviate the imbalance of samples. For the Landsat-8 Biome set, CD-AttDLV3+ achieves the highest accuracy in comparison with other methods, including Fmask, SVM, and SegNet, especially for distinguishing clouds from bright surfaces and detecting light-transmitting thin clouds. It can also perform well on other Landsat-8 and Sentinel-2 images. Experimental results indicate that CD-AttDLV3+ is robust, with a high accuracy and extendibility.
- 10 September 2021
- Published by MDPI AG in Remote Sensing
- Vol. 13 (18) , 3617
Open AccessLight-Weight Cloud Detection Network for Optical Remote Sensing Images with Attention-Based DeeplabV3+ Architecture
Clouds in optical remote sensing images cause spectral information change or loss, that affects image analysis and application. Therefore, cloud detection is of great significance. However, there are some shortcomings in current methods, such as the insufficient extendibility due to using the information of multiple bands, the intense extendibility due to relying on some manually determined thresholds, and the limited accuracy, especially for thin clouds or complex scenes caused by low-level manual features. Combining the above shortcomings and the requirements for efficiency in practical applications, we propose a light-weight deep learning cloud detection network based on DeeplabV3+ architecture and channel attention module (CD-AttDLV3+), only using the most common red–green–blue and near-infrared bands. In the CD-AttDLV3+ architecture, an optimized backbone network-MobileNetV2 is used to reduce the number of parameters and calculations. Atrous spatial pyramid pooling effectively reduces the information loss caused by multiple down-samplings while extracting multi-scale features. CD-AttDLV3+ concatenates more low-level features than DeeplabV3+ to improve the cloud boundary quality. The channel attention module is introduced to strengthen the learning of important channels and improve the training efficiency. Moreover, the loss function is improved to alleviate the imbalance of samples. For the Landsat-8 Biome set, CD-AttDLV3+ achieves the highest accuracy in comparison with other methods, including Fmask, SVM, and SegNet, especially for distinguishing clouds from bright surfaces and detecting light-transmitting thin clouds. It can also perform well on other Landsat-8 and Sentinel-2 images. Experimental results indicate that CD-AttDLV3+ is robust, with a high accuracy and extendibility.