ISPRS Journal of Photogrammetry and Remote Sensing最新文献_第7页

MARSNet: A Mamba-driven adaptive framework for robust multisource remote sensing image matching in noisy environments MARSNet：一个mamba驱动的自适应框架，用于噪声环境下的鲁棒多源遥感图像匹配

IF 12.2 1区地球科学 Q1 GEOGRAPHY, PHYSICAL

ISPRS Journal of Photogrammetry and Remote Sensing

Pub Date : 2026-01-05 DOI: 10.1016/j.isprsjprs.2025.12.021

Weipeng Jing , Peilun Kang , Donglin Di , Jian Wang , Yang Song , Chao Li , Lei Fan

Semi-dense matching of multi-source remote sensing images under noise interference remains a challenging task. Existing detector-free methods often exhibit low efficiency and reduced performance when faced with large viewpoint variations and significant noise disturbances. Due to the inherent noise and modality differences in multi-source remote sensing images, the accuracy and robustness of feature matching are substantially compromised. To address this issue, we propose a hybrid network for multi-source remote sensing image matching based on an efficient and robust Mamba framework, named MARSNet. The network achieves efficient and robust matching through the following innovative designs: First, it leverages the efficient Mamba network to capture long-range dependencies within image sequences, enhancing the modeling capability for complex scenes. Second, a frozen pre-trained DINOv2 foundation model is introduced as a robust feature extractor, effectively improving the model’s noise resistance. Finally, an adaptive fusion strategy is employed to integrate features, and the Mamba-like linear attention mechanism is adopted to refine the Transformer-based linear attention, further enhancing the efficiency and expressive power for long-sequence processing. To validate the effectiveness of the proposed method, extensive experiments were conducted on multi-source remote sensing image datasets, covering various scenarios such as noise-free, additive random noise, and periodic stripe noise. The experimental results demonstrate that the proposed method achieves significant improvements in matching accuracy and robustness compared to state-of-the-art methods. Additionally, by performing pose error evaluation on a large-scale general dataset, the superior performance of the proposed method in 3D reconstruction is validated, complementing the test results from the multi-source remote sensing dataset, thereby providing a more comprehensive assessment of the method’s generalization ability and robustness.

噪声干扰下多源遥感图像的半密集匹配一直是一项具有挑战性的任务。现有的无探测器方法在视点变化大、噪声干扰大的情况下，效率低、性能下降。由于多源遥感图像中固有的噪声和模态差异，大大降低了特征匹配的准确性和鲁棒性。为了解决这一问题，我们提出了一种基于高效鲁棒曼巴框架的多源遥感图像匹配混合网络，称为MARSNet。该网络通过以下创新设计实现了高效鲁棒的匹配：首先，利用高效的曼巴网络捕获图像序列内的远程依赖关系，增强了对复杂场景的建模能力。其次，引入冻结预训练的DINOv2基础模型作为鲁棒特征提取器，有效提高了模型的抗噪性；最后，采用自适应融合策略对特征进行融合，并采用类似mamba的线性注意机制对基于transformer的线性注意进行细化，进一步提高了长序列处理的效率和表达能力。为了验证该方法的有效性，在多源遥感图像数据集上进行了大量实验，包括无噪声、加性随机噪声和周期性条纹噪声等多种场景。实验结果表明，与现有方法相比，该方法在匹配精度和鲁棒性方面都有显著提高。此外，通过在大型通用数据集上进行位姿误差评估，验证了该方法在三维重建中的优越性能，与多源遥感数据集的测试结果相补充，从而更全面地评估了该方法的泛化能力和鲁棒性。

{"title":"MARSNet: A Mamba-driven adaptive framework for robust multisource remote sensing image matching in noisy environments","authors":"Weipeng Jing , Peilun Kang , Donglin Di , Jian Wang , Yang Song , Chao Li , Lei Fan","doi":"10.1016/j.isprsjprs.2025.12.021","DOIUrl":"10.1016/j.isprsjprs.2025.12.021","url":null,"abstract":"<div><div>Semi-dense matching of multi-source remote sensing images under noise interference remains a challenging task. Existing detector-free methods often exhibit low efficiency and reduced performance when faced with large viewpoint variations and significant noise disturbances. Due to the inherent noise and modality differences in multi-source remote sensing images, the accuracy and robustness of feature matching are substantially compromised. To address this issue, we propose a hybrid network for multi-source remote sensing image matching based on an efficient and robust Mamba framework, named MARSNet. The network achieves efficient and robust matching through the following innovative designs: First, it leverages the efficient Mamba network to capture long-range dependencies within image sequences, enhancing the modeling capability for complex scenes. Second, a frozen pre-trained DINOv2 foundation model is introduced as a robust feature extractor, effectively improving the model’s noise resistance. Finally, an adaptive fusion strategy is employed to integrate features, and the Mamba-like linear attention mechanism is adopted to refine the Transformer-based linear attention, further enhancing the efficiency and expressive power for long-sequence processing. To validate the effectiveness of the proposed method, extensive experiments were conducted on multi-source remote sensing image datasets, covering various scenarios such as noise-free, additive random noise, and periodic stripe noise. The experimental results demonstrate that the proposed method achieves significant improvements in matching accuracy and robustness compared to state-of-the-art methods. Additionally, by performing pose error evaluation on a large-scale general dataset, the superior performance of the proposed method in 3D reconstruction is validated, complementing the test results from the multi-source remote sensing dataset, thereby providing a more comprehensive assessment of the method’s generalization ability and robustness.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"232 ","pages":"Pages 457-475"},"PeriodicalIF":12.2,"publicationDate":"2026-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145903503","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A Spatially Masked Adaptive Gated Network for multimodal post-flood water extent mapping using SAR and incomplete multispectral data 基于SAR和不完全多光谱数据的多模态洪水后水位映射空间掩膜自适应门控网络

IF 12.2 1区地球科学 Q1 GEOGRAPHY, PHYSICAL

ISPRS Journal of Photogrammetry and Remote Sensing

Pub Date : 2026-01-05 DOI: 10.1016/j.isprsjprs.2025.12.023

Hyunho Lee, Wenwen Li

Mapping water extent during a flood event is essential for effective disaster management throughout all phases: mitigation, preparedness, response, and recovery. In particular, during the response stage, when timely and accurate information is important, Synthetic Aperture Radar (SAR) data are primarily employed to produce water extent maps. This is because SAR sensors can observe through cloud cover and operate both day and night, whereas Multispectral Imaging (MSI) data, despite providing higher mapping accuracy, are only available under cloud-free and daytime conditions. Recently, leveraging the complementary characteristics of SAR and MSI data through a multimodal approach has emerged as a promising strategy for advancing water extent mapping using deep learning models. This approach is particularly beneficial when timely post-flood observations, acquired during or shortly after the flood peak, are limited, as it enables the use of all available imagery for more accurate post-flood water extent mapping. However, the adaptive integration of partially available MSI data into the SAR-based post-flood water extent mapping process remains underexplored. To bridge this research gap, we propose the Spatially Masked Adaptive Gated Network (SMAGNet), a multimodal deep learning model that utilizes SAR data as the primary input for post-flood water extent mapping and integrates complementary MSI data through feature fusion. In experiments on the C2S-MS Floods dataset, SMAGNet consistently outperformed other multimodal deep learning models in prediction performance across varying levels of MSI data availability. Specifically, SMAGNet achieved the highest IoU score of 86.47% using SAR and MSI data and maintained the highest performance with an IoU score of 79.53% even when MSI data were entirely missing. Furthermore, we found that even when MSI data were completely missing, the performance of SMAGNet remained statistically comparable to that of a U-Net model trained solely on SAR data. These findings indicate that SMAGNet enhances the model robustness to missing data as well as the applicability of multimodal deep learning in real-world flood management scenarios. The source code is available at https://github.com/ASUcicilab/SMAGNet.

绘制洪水期间的水位图对于在减灾、备灾、救灾和恢复等各个阶段进行有效的灾害管理至关重要。特别是在响应阶段，需要及时准确的信息，合成孔径雷达（Synthetic Aperture Radar， SAR）数据主要用于生成水域图。这是因为SAR传感器可以透过云层进行观测，并且可以在白天和夜间操作，而多光谱成像（MSI）数据尽管提供了更高的制图精度，但只能在无云和白天的条件下使用。最近，通过多模态方法利用SAR和MSI数据的互补特征已成为利用深度学习模型推进水域制图的一种有前途的策略。当在洪峰期间或洪峰后不久获得的及时的洪水后观测数据有限时，这种方法特别有用，因为它可以使用所有可用的图像来更准确地绘制洪水后的水范围图。然而，将部分可用的MSI数据自适应集成到基于sar的洪水后水位制图过程中仍未得到充分探索。为了弥补这一研究空白，我们提出了空间掩膜自适应门控网络（SMAGNet），这是一种多模态深度学习模型，利用SAR数据作为洪水后水范围映射的主要输入，并通过特征融合整合互补的MSI数据。在C2S-MS洪水数据集的实验中，SMAGNet在不同级别的MSI数据可用性的预测性能方面始终优于其他多模态深度学习模型。具体而言，SMAGNet在使用SAR和MSI数据时获得了最高的IoU得分86.47%，即使在MSI数据完全缺失的情况下也保持了最高的IoU得分79.53%。此外，我们发现，即使在MSI数据完全缺失的情况下，SMAGNet的性能在统计上仍与仅使用SAR数据训练的U-Net模型相当。这些发现表明SMAGNet增强了模型对缺失数据的鲁棒性，以及多模态深度学习在现实世界洪水管理场景中的适用性。源代码可从https://github.com/ASUcicilab/SMAGNet获得。

{"title":"A Spatially Masked Adaptive Gated Network for multimodal post-flood water extent mapping using SAR and incomplete multispectral data","authors":"Hyunho Lee, Wenwen Li","doi":"10.1016/j.isprsjprs.2025.12.023","DOIUrl":"10.1016/j.isprsjprs.2025.12.023","url":null,"abstract":"<div><div>Mapping water extent during a flood event is essential for effective disaster management throughout all phases: mitigation, preparedness, response, and recovery. In particular, during the response stage, when timely and accurate information is important, Synthetic Aperture Radar (SAR) data are primarily employed to produce water extent maps. This is because SAR sensors can observe through cloud cover and operate both day and night, whereas Multispectral Imaging (MSI) data, despite providing higher mapping accuracy, are only available under cloud-free and daytime conditions. Recently, leveraging the complementary characteristics of SAR and MSI data through a multimodal approach has emerged as a promising strategy for advancing water extent mapping using deep learning models. This approach is particularly beneficial when timely post-flood observations, acquired during or shortly after the flood peak, are limited, as it enables the use of all available imagery for more accurate post-flood water extent mapping. However, the adaptive integration of partially available MSI data into the SAR-based post-flood water extent mapping process remains underexplored. To bridge this research gap, we propose the Spatially Masked Adaptive Gated Network (SMAGNet), a multimodal deep learning model that utilizes SAR data as the primary input for post-flood water extent mapping and integrates complementary MSI data through feature fusion. In experiments on the C2S-MS Floods dataset, SMAGNet consistently outperformed other multimodal deep learning models in prediction performance across varying levels of MSI data availability. Specifically, SMAGNet achieved the highest IoU score of 86.47% using SAR and MSI data and maintained the highest performance with an IoU score of 79.53% even when MSI data were entirely missing. Furthermore, we found that even when MSI data were completely missing, the performance of SMAGNet remained statistically comparable to that of a U-Net model trained solely on SAR data. These findings indicate that SMAGNet enhances the model robustness to missing data as well as the applicability of multimodal deep learning in real-world flood management scenarios. The source code is available at <span><span>https://github.com/ASUcicilab/SMAGNet</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"232 ","pages":"Pages 492-508"},"PeriodicalIF":12.2,"publicationDate":"2026-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145902910","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Uncovering spatial process heterogeneity from graph-based deep spatial regression 基于图的深度空间回归揭示空间过程异质性

IF 12.2 1区地球科学 Q1 GEOGRAPHY, PHYSICAL

ISPRS Journal of Photogrammetry and Remote Sensing

Pub Date : 2026-01-05 DOI: 10.1016/j.isprsjprs.2025.12.008

Di Zhu , Sheng Wang , Peng Luo

One can never specify the true statistical form of a complex spatial process. Model misspecification, such as linearity and additivity, will lead to fundamentally flawed interpretations in the estimated coefficients, particularly in empirical geographic studies involving a large number of observations and complex data generation processes. Motivated to learn representative patterns of spatial process heterogeneity, we propose a deep explainable spatial regression (XSR) framework based on graph convolutional neural networks (GCN), which bypasses the conventional parametric statistical assumptions in spatial regression modeling and can generate deep spatially varying coefficients that depict the heterogeneity structure of spatial processes. We introduce an analytical framework to (1) perform deep spatial regression modeling in multivariate cross-sectional scenarios, (2) reconstruct spatial heterogeneity patterns from the learned deep coefficients, and (3) explain the effectiveness of heterogeneity through a simple diagnostic test. Experiments on Greater Boston house prices modeling demonstrate better fitting performance over spatial regression baselines. The spatial patterns of deep local coefficients consistently exhibit stronger explanatory power than those derived from geographically weighted regression, indicating a better representation of the true spatial process heterogeneity uncovered by graph-based deep spatial regression.

一个人永远不能确定一个复杂空间过程的真正统计形式。模型规格不当，如线性和可加性，将导致估计系数的解释从根本上存在缺陷，特别是在涉及大量观测和复杂数据生成过程的实证地理研究中。为了学习空间过程异质性的代表性模式，我们提出了一个基于图卷积神经网络（GCN）的深度可解释空间回归（XSR）框架，该框架绕开了空间回归建模中传统的参数统计假设，可以生成描述空间过程异质性结构的深度空间变化系数。我们引入了一个分析框架：(1)在多变量横截面情景下进行深度空间回归建模，(2)根据学习到的深度系数重构空间异质性模式，(3)通过简单的诊断测试解释异质性的有效性。大波士顿地区房价模型的实验表明，空间回归基线具有更好的拟合性能。与地理加权回归相比，深度局部系数的空间格局表现出更强的解释力，表明基于图的深度空间回归更能反映真实的空间过程异质性。

{"title":"Uncovering spatial process heterogeneity from graph-based deep spatial regression","authors":"Di Zhu , Sheng Wang , Peng Luo","doi":"10.1016/j.isprsjprs.2025.12.008","DOIUrl":"10.1016/j.isprsjprs.2025.12.008","url":null,"abstract":"<div><div>One can never specify the true statistical form of a complex spatial process. Model misspecification, such as linearity and additivity, will lead to fundamentally flawed interpretations in the estimated coefficients, particularly in empirical geographic studies involving a large number of observations and complex data generation processes. Motivated to learn representative patterns of spatial process heterogeneity, we propose a deep explainable spatial regression (XSR) framework based on graph convolutional neural networks (GCN), which bypasses the conventional parametric statistical assumptions in spatial regression modeling and can generate deep spatially varying coefficients that depict the heterogeneity structure of spatial processes. We introduce an analytical framework to (1) perform deep spatial regression modeling in multivariate cross-sectional scenarios, (2) reconstruct spatial heterogeneity patterns from the learned deep coefficients, and (3) explain the effectiveness of heterogeneity through a simple diagnostic test. Experiments on Greater Boston house prices modeling demonstrate better fitting performance over spatial regression baselines. The spatial patterns of deep local coefficients consistently exhibit stronger explanatory power than those derived from geographically weighted regression, indicating a better representation of the true spatial process heterogeneity uncovered by graph-based deep spatial regression.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"232 ","pages":"Pages 509-523"},"PeriodicalIF":12.2,"publicationDate":"2026-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145902923","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

AEOS: Active Environment-aware Optimal Scanning Control for UAV LiDAR-Inertial Odometry in Complex Scenes 复杂场景下无人机激光雷达-惯性里程计主动环境感知最优扫描控制

IF 12.2 1区地球科学 Q1 GEOGRAPHY, PHYSICAL

ISPRS Journal of Photogrammetry and Remote Sensing

Pub Date : 2026-01-05 DOI: 10.1016/j.isprsjprs.2026.01.006

Jianping Li , Xinhang Xu , Zhongyuan Liu , Shenghai Yuan , Muqing Cao , Lihua Xie

LiDAR-based 3D perception and localization on unmanned aerial vehicles (UAVs) are fundamentally limited by the narrow field of view (FoV) of compact LiDAR sensors and the payload constraints that preclude multi-sensor configurations. Traditional motorized scanning systems with fixed-speed rotations lack scene awareness and task-level adaptability, leading to degraded odometry and mapping performance in complex, occluded environments. Inspired by the active sensing behavior of owls, we propose AEOS (Active Environment-aware Optimal Scanning), a biologically inspired and computationally efficient framework for adaptive LiDAR control in UAV-based LiDAR-Inertial Odometry (LIO). AEOS combines model predictive control (MPC) and reinforcement learning (RL) in a hybrid architecture: an analytical uncertainty model predicts future pose observability for exploitation, while a lightweight neural network learns an implicit cost map from panoramic depth representations to guide exploration. To support scalable training and generalization, we develop a point cloud-based simulation environment with real-world LiDAR maps across diverse scenes, enabling sim-to-real transfer. Extensive experiments in both simulation and real-world environments demonstrate that AEOS significantly improves odometry accuracy compared to fixed-rate, optimization-only, and fully learned baselines, while maintaining real-time performance under onboard computational constraints. The project page can be found at https://kafeiyin00.github.io/AEOS/.

基于激光雷达的无人机3D感知和定位从根本上受到紧凑型激光雷达传感器窄视场（FoV）和多传感器配置的有效载荷限制的限制。传统的固定速度旋转的机动扫描系统缺乏场景感知和任务级适应性，导致在复杂、闭塞的环境中里程测量和映射性能下降。受猫头鹰主动感知行为的启发，我们提出了AEOS（主动环境感知最佳扫描），这是一种受生物学启发且计算效率高的框架，用于基于无人机的LIO （LiDAR- inertial Odometry）中的自适应LiDAR控制。AEOS在混合架构中结合了模型预测控制（MPC）和强化学习（RL）：分析不确定性模型预测未来的姿态可观察性以进行开发，而轻量级神经网络从全景深度表示中学习隐式成本图以指导探索。为了支持可扩展的训练和泛化，我们开发了一个基于点云的模拟环境，其中包含不同场景的真实激光雷达地图，实现了从模拟到真实的传输。在仿真和现实环境中进行的大量实验表明，与固定速率、仅优化和完全学习基线相比，AEOS显著提高了里程测量精度，同时在机载计算限制下保持了实时性能。项目页面可以在https://kafeiyin00.github.io/AEOS/上找到。

{"title":"AEOS: Active Environment-aware Optimal Scanning Control for UAV LiDAR-Inertial Odometry in Complex Scenes","authors":"Jianping Li , Xinhang Xu , Zhongyuan Liu , Shenghai Yuan , Muqing Cao , Lihua Xie","doi":"10.1016/j.isprsjprs.2026.01.006","DOIUrl":"10.1016/j.isprsjprs.2026.01.006","url":null,"abstract":"<div><div>LiDAR-based 3D perception and localization on unmanned aerial vehicles (UAVs) are fundamentally limited by the narrow field of view (FoV) of compact LiDAR sensors and the payload constraints that preclude multi-sensor configurations. Traditional motorized scanning systems with fixed-speed rotations lack scene awareness and task-level adaptability, leading to degraded odometry and mapping performance in complex, occluded environments. Inspired by the active sensing behavior of owls, we propose AEOS (Active Environment-aware Optimal Scanning), a biologically inspired and computationally efficient framework for adaptive LiDAR control in UAV-based LiDAR-Inertial Odometry (LIO). AEOS combines model predictive control (MPC) and reinforcement learning (RL) in a hybrid architecture: an analytical uncertainty model predicts future pose observability for exploitation, while a lightweight neural network learns an implicit cost map from panoramic depth representations to guide exploration. To support scalable training and generalization, we develop a point cloud-based simulation environment with real-world LiDAR maps across diverse scenes, enabling sim-to-real transfer. Extensive experiments in both simulation and real-world environments demonstrate that AEOS significantly improves odometry accuracy compared to fixed-rate, optimization-only, and fully learned baselines, while maintaining real-time performance under onboard computational constraints. The project page can be found at <span><span>https://kafeiyin00.github.io/AEOS/</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"232 ","pages":"Pages 476-491"},"PeriodicalIF":12.2,"publicationDate":"2026-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145902908","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

RSMT: Robust stereo matching training with geometric correction, clean pixel selection and loss weighting RSMT：具有几何校正、干净像素选择和损失加权的鲁棒立体匹配训练

IF 12.2 1区地球科学 Q1 GEOGRAPHY, PHYSICAL

ISPRS Journal of Photogrammetry and Remote Sensing

Pub Date : 2025-12-30 DOI: 10.1016/j.isprsjprs.2025.12.014

Haoxuan Sun , Taoyang Wang , Qian Cheng , Jiaxuan Huang

Inaccurate or noisy labels have a huge impact on the training of deep learning models. To date, few studies have focused on the label error problem in satellite image stereo matching. In this paper, we analyzed and found the two open datasets US3D and WHU-Stereo contain label errors that cannot be overlooked. A new task is extremely necessary: learning from inaccurate labels with neural networks. Our motivation is to deal with label errors at the training level. A robust stereo matching training framework (RSMT) with geometric correction, clean pixel selection, and loss weighting modules is proposed. In addition, we also propose a dataset correcting method and provide two inaccurate-label stereo matching datasets US3D(E) and WHU(E) based on raw datasets. The framework can be applied to common stereo methods like IGEV-Stereo and ACVNet to achieve SOTA performance on the corrected datasets. To the best of our knowledge, the study is the first systemic inaccurate-label learning framework dedicated to stereo matching. Datasets are available at https://github.com/endu111/robust-satellite-image-stereo-matching.

不准确或有噪声的标签对深度学习模型的训练有巨大的影响。迄今为止，很少有研究关注卫星图像立体匹配中的标签误差问题。在本文中，我们分析了US3D和WHU-Stereo两个开放数据集，发现它们存在不可忽视的标签错误。一个新的任务是非常必要的：用神经网络从不准确的标签中学习。我们的动机是在训练层面处理标签错误。提出了一种具有几何校正、干净像素选择和损失加权模块的鲁棒立体匹配训练框架（RSMT）。此外，我们还提出了一种数据集校正方法，并在原始数据集的基础上提供了两个不精确标签立体匹配数据集US3D(E)和WHU(E)。该框架可应用于IGEV-Stereo和ACVNet等常用立体方法，在校正后的数据集上实现SOTA性能。据我们所知，这项研究是第一个致力于立体匹配的系统的不准确标签学习框架。数据集可在https://github.com/endu111/robust-satellite-image-stereo-matching上获得。

{"title":"RSMT: Robust stereo matching training with geometric correction, clean pixel selection and loss weighting","authors":"Haoxuan Sun , Taoyang Wang , Qian Cheng , Jiaxuan Huang","doi":"10.1016/j.isprsjprs.2025.12.014","DOIUrl":"10.1016/j.isprsjprs.2025.12.014","url":null,"abstract":"<div><div>Inaccurate or noisy labels have a huge impact on the training of deep learning models. To date, few studies have focused on the label error problem in satellite image stereo matching. In this paper, we analyzed and found the two open datasets US3D and WHU-Stereo contain label errors that cannot be overlooked. A new task is extremely necessary: learning from inaccurate labels with neural networks. Our motivation is to deal with label errors at the training level. A robust stereo matching training framework (RSMT) with geometric correction, clean pixel selection, and loss weighting modules is proposed. In addition, we also propose a dataset correcting method and provide two inaccurate-label stereo matching datasets US3D(E) and WHU(E) based on raw datasets. The framework can be applied to common stereo methods like IGEV-Stereo and ACVNet to achieve SOTA performance on the corrected datasets. To the best of our knowledge, the study is the first systemic inaccurate-label learning framework dedicated to stereo matching. Datasets are available at <span><span>https://github.com/endu111/robust-satellite-image-stereo-matching</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"232 ","pages":"Pages 421-436"},"PeriodicalIF":12.2,"publicationDate":"2025-12-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145884095","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Phase gradient rate constrained minimum cost flow: A robust unwrapping method for landslides with large deformation gradients 相位梯度率约束的最小代价流：大变形梯度滑坡的鲁棒解包裹方法

IF 12.2 1区地球科学 Q1 GEOGRAPHY, PHYSICAL

ISPRS Journal of Photogrammetry and Remote Sensing

Pub Date : 2025-12-30 DOI: 10.1016/j.isprsjprs.2025.12.017

Xiaosong Feng , Lianhuan Wei , Xu Ren , Chaoying Zhao , Meng Ao , Yian Wang , Jiayin Luo , Christian Bignami

Phase unwrapping (PU) is a key step in synthetic aperture radar interferometry (InSAR) techniques, as its accuracy directly determines the precision of deformation estimation. Despite the widespread use of InSAR, many existing PU techniques, such as minimum cost flow (MCF), Snaphu, and recently developed deep-learning models, struggle to maintain high accuracy when faced with large deformation gradients. To address this issue, this study proposes a phase gradient rate constrained minimum cost flow (PGR-MCF) method. It uses the phase gradient rate (PGR) calculated from time series differential interferograms through stacking and fusion, to constrain weighting of the arcs during MCF unwrapping process. By optimizing the unwrapping path to prioritize low-gradient areas, the PGR-MCF method significantly improves unwrapping accuracy in high-gradient regions. In simulated experiments, the PGR-MCF method correctly unwrapped 99.5 % of the pixels in large deformation zones. In the application to Guobu landslide, the PGR-MCF method reduces the root mean square errors (RMSEs) between InSAR and global navigation satellite system (GNSS) deformation time series by more than 69 %, compared to other tested methods. This method does not require external datasets or prior models, ensuring its broad applicability. Additionally, it reliably unwraps interferograms with large spatiotemporal baselines, thus increasing the number of reliable unwrapped interferograms for time-series deformation inversion and improving deformation monitoring accuracy. Moreover, it has been proven to be an effective PU method for deformation estimation in regions with large deformation gradients.

相位展开（PU）是合成孔径雷达干涉测量（InSAR）技术中的关键步骤，其精度直接决定了变形估计的精度。尽管InSAR被广泛使用，但许多现有的PU技术，如最小成本流（MCF）、Snaphu和最近开发的深度学习模型，在面对大变形梯度时难以保持高精度。为了解决这一问题，本研究提出了相位梯度率约束最小成本流（PGR-MCF）方法。该算法利用时间序列差分干涉图通过叠加和融合计算得到的相位梯度率（PGR）来约束MCF解包裹过程中弧的权重。PGR-MCF方法通过优化展开路径，优先考虑低梯度区域，显著提高了高梯度区域的展开精度。在模拟实验中，PGR-MCF方法在大变形区域中能正确地解开99.5%的像元。在果布滑坡的应用中，与其他测试方法相比，PGR-MCF方法将InSAR与全球导航卫星系统（GNSS）变形时间序列的均方根误差（rmse）降低了69%以上。该方法不需要外部数据集或先验模型，保证了其广泛的适用性。对大时空基线的干涉图进行可靠解包裹，增加了时间序列变形反演的可靠解包裹干涉图数量，提高了变形监测精度。此外，它还被证明是一种有效的PU方法，用于大变形梯度区域的变形估计。

{"title":"Phase gradient rate constrained minimum cost flow: A robust unwrapping method for landslides with large deformation gradients","authors":"Xiaosong Feng , Lianhuan Wei , Xu Ren , Chaoying Zhao , Meng Ao , Yian Wang , Jiayin Luo , Christian Bignami","doi":"10.1016/j.isprsjprs.2025.12.017","DOIUrl":"10.1016/j.isprsjprs.2025.12.017","url":null,"abstract":"<div><div>Phase unwrapping (PU) is a key step in synthetic aperture radar interferometry (InSAR) techniques, as its accuracy directly determines the precision of deformation estimation. Despite the widespread use of InSAR, many existing PU techniques, such as minimum cost flow (MCF), Snaphu, and recently developed deep-learning models, struggle to maintain high accuracy when faced with large deformation gradients. To address this issue, this study proposes a phase gradient rate constrained minimum cost flow (PGR-MCF) method. It uses the phase gradient rate (PGR) calculated from time series differential interferograms through stacking and fusion, to constrain weighting of the arcs during MCF unwrapping process. By optimizing the unwrapping path to prioritize low-gradient areas, the PGR-MCF method significantly improves unwrapping accuracy in high-gradient regions. In simulated experiments, the PGR-MCF method correctly unwrapped 99.5 % of the pixels in large deformation zones. In the application to Guobu landslide, the PGR-MCF method reduces the root mean square errors (RMSEs) between InSAR and global navigation satellite system (GNSS) deformation time series by more than 69 %, compared to other tested methods. This method does not require external datasets or prior models, ensuring its broad applicability. Additionally, it reliably unwraps interferograms with large spatiotemporal baselines, thus increasing the number of reliable unwrapped interferograms for time-series deformation inversion and improving deformation monitoring accuracy. Moreover, it has been proven to be an effective PU method for deformation estimation in regions with large deformation gradients.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"232 ","pages":"Pages 437-456"},"PeriodicalIF":12.2,"publicationDate":"2025-12-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145884002","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Roof-aware indoor BIM reconstruction from LiDAR via graph-attention for residential buildings 基于图形关注的住宅建筑激光雷达屋顶感知室内BIM重建

IF 12.2 1区地球科学 Q1 GEOGRAPHY, PHYSICAL

ISPRS Journal of Photogrammetry and Remote Sensing

Pub Date : 2025-12-29 DOI: 10.1016/j.isprsjprs.2025.12.024

Biao Xiong , Bohan Wang , Yiyi Liu , Liangliang Wang , Yanchao Yang , Liang Zhou , Qiegen Liu

Building Information Models (BIMs) provide structured, parametric representations that are fundamental for simulation, facility management, and digital twin applications. However, reconstructing BIMs from terrestrial LiDAR scans remains challenging due to clutter, occlusions, and the geometric complexity of roof structures. This paper presents a roof-aware scan-to-BIM pipeline tailored for residential buildings, which processes indoor LiDAR data through four geometric abstractions, raw points, superpoints, triangle meshes, and volumetric polyhedra, each represented by task-specific graphs. The pipeline integrates three modules: LGNet for semantic segmentation, QTNet for floor plan reconstruction, and PPO for roof–floor fusion. It demonstrates strong cross-dataset generalization, being trained on Structured3D and fine-tuned on the real-world WHUTS dataset. The method produces watertight, Revit-compatible BIMs with an average surface deviation of 9 mm RMS on WHUTS scenes featuring slanted roofs. Compared with state-of-the-art scan-to-BIM and floor plan reconstruction methods, the proposed approach achieves higher geometric accuracy on scenes with slanted roofs, reducing surface reconstruction error by over 12–18% and improving layout reconstruction F1-scores by up to 6–8%. The proposed framework provides a robust, accurate, and fully automated solution for roof-aware BIM reconstruction of residential buildings from terrestrial LiDAR data, offering comprehensive support for slanted roof modeling. The source code and datasets are publicly available at https://github.com/Wangbohan-x/roof-aware-scan2bim.git.

建筑信息模型（bim）为仿真、设施管理和数字孪生应用提供结构化、参数化的表示。然而，由于杂波、遮挡和屋顶结构的几何复杂性，从地面激光雷达扫描重建bim仍然具有挑战性。本文介绍了为住宅建筑量身定制的屋顶感知扫描到bim的管道，该管道通过四个几何抽象处理室内激光雷达数据，原始点、叠加点、三角形网格和体积多面体，每个都由特定任务的图形表示。该管道集成了三个模块：用于语义分割的LGNet、用于平面图重建的QTNet和用于屋顶-地板融合的PPO。它展示了强大的跨数据集泛化，在Structured3D上进行了训练，并在现实世界的WHUTS数据集上进行了微调。该方法可以在WHUTS倾斜屋顶场景中产生水密的、revit兼容的bim，平均表面偏差RMS为9毫米。与目前最先进的扫描到bim和平面图重建方法相比，该方法在倾斜屋顶场景中具有更高的几何精度，将表面重建误差降低了12-18%以上，将布局重建f1分数提高了6-8%。提出的框架为基于地面激光雷达数据的住宅建筑的屋顶感知BIM重建提供了一个强大、准确和全自动的解决方案，为倾斜屋顶建模提供了全面的支持。源代码和数据集可在https://github.com/Wangbohan-x/roof-aware-scan2bim.git上公开获得。

{"title":"Roof-aware indoor BIM reconstruction from LiDAR via graph-attention for residential buildings","authors":"Biao Xiong , Bohan Wang , Yiyi Liu , Liangliang Wang , Yanchao Yang , Liang Zhou , Qiegen Liu","doi":"10.1016/j.isprsjprs.2025.12.024","DOIUrl":"10.1016/j.isprsjprs.2025.12.024","url":null,"abstract":"<div><div>Building Information Models (BIMs) provide structured, parametric representations that are fundamental for simulation, facility management, and digital twin applications. However, reconstructing BIMs from terrestrial LiDAR scans remains challenging due to clutter, occlusions, and the geometric complexity of roof structures. This paper presents a <strong>roof-aware scan-to-BIM pipeline</strong> tailored for residential buildings, which processes indoor LiDAR data through four geometric abstractions, raw points, superpoints, triangle meshes, and volumetric polyhedra, each represented by task-specific graphs. The pipeline integrates three modules: <strong>LGNet</strong> for semantic segmentation, <strong>QTNet</strong> for floor plan reconstruction, and <strong>PPO</strong> for roof–floor fusion. It demonstrates strong cross-dataset generalization, being trained on Structured3D and fine-tuned on the real-world WHUTS dataset. The method produces watertight, Revit-compatible BIMs with an average surface deviation of <strong>9 mm RMS</strong> on WHUTS scenes featuring slanted roofs. Compared with state-of-the-art scan-to-BIM and floor plan reconstruction methods, the proposed approach achieves higher geometric accuracy on scenes with slanted roofs, reducing surface reconstruction error by over <strong>12–18%</strong> and improving layout reconstruction F1-scores by up to <strong>6–8%</strong>. The proposed framework provides a robust, accurate, and fully automated solution for roof-aware BIM reconstruction of residential buildings from terrestrial LiDAR data, offering comprehensive support for slanted roof modeling. The source code and datasets are publicly available at <span><span>https://github.com/Wangbohan-x/roof-aware-scan2bim.git</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"232 ","pages":"Pages 408-420"},"PeriodicalIF":12.2,"publicationDate":"2025-12-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145884003","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

TFRSUB: A terrain-feature retention and spatial uniformity balancing method for simplifying LiDAR ground point clouds TFRSUB：一种简化激光雷达地面点云的地形特征保持和空间均匀性平衡方法

IF 12.2 1区地球科学 Q1 GEOGRAPHY, PHYSICAL

ISPRS Journal of Photogrammetry and Remote Sensing

Pub Date : 2025-12-29 DOI: 10.1016/j.isprsjprs.2025.12.015

Chuanfa Chen, Ziming Yang, Hongming Pan, Yanyan Li, Jinda Hao

The processing of high-density LiDAR point clouds presents significant computational challenges for DEM generation due to data redundancy and topographic feature degradation during simplification. To overcome this problem, this study proposes a Terrain-Feature Retention and Spatial Uniformity Balancing (TFRSUB) method that integrates three key innovations: (i) a Distance-Geometric Synergy Index (DGSI) combining orthogonal deviation distance and point sampling interval to mitigate boundary contraction artifacts; (ii) a Composite Terrain Factor (CTF) synthesizing multiple terrain parameters to characterize diverse topographic features; and (iii) a cluster-driven Gaussian Process Regression (GPR) framework using CTF for iterative feature point selection, optimizing the trade-off between topographic fidelity and point distribution homogeneity. Evaluated on eight high-resolution LiDAR terrain point clouds across six retention ratios, TFRSUB demonstrates significant accuracy improvements over seven state-of-the-art methods, achieving reductions of 9.22%–64.70% in DEM root mean square error, 7.80%–61.34% in mean absolute error, 16.43%–76.88% in slope error, and 28.12%–81.35% in mean curvature error. These results establish TFRSUB as an alternative solution for LiDAR point cloud simplification that maintains topographic fidelity while addressing computational storage challenges.

高密度LiDAR点云的处理由于数据冗余和简化过程中的地形特征退化，给DEM生成带来了巨大的计算挑战。为了克服这一问题，本研究提出了一种地形特征保持和空间均匀性平衡（TFRSUB）方法，该方法集成了三个关键创新：(i)结合正交偏差距离和点采样间隔的距离几何协同指数（DGSI）来减轻边界收缩伪影；（ii）综合多个地形参数以表征不同地形特征的复合地形因子（CTF）；（iii）使用CTF进行迭代特征点选择的聚类驱动高斯过程回归（GPR）框架，优化了地形保真度和点分布均匀性之间的权衡。通过对8个高分辨率LiDAR地形点云的6种保持比进行评估，TFRSUB比7种最先进的方法具有显著的精度提高，DEM均方根误差降低9.22% ~ 64.70%，平均绝对误差降低7.80% ~ 61.34%，坡度误差降低16.43% ~ 76.88%，平均曲率误差降低28.12% ~ 81.35%。这些结果表明，TFRSUB是激光雷达点云简化的替代解决方案，在解决计算存储挑战的同时保持地形保真度。

{"title":"TFRSUB: A terrain-feature retention and spatial uniformity balancing method for simplifying LiDAR ground point clouds","authors":"Chuanfa Chen, Ziming Yang, Hongming Pan, Yanyan Li, Jinda Hao","doi":"10.1016/j.isprsjprs.2025.12.015","DOIUrl":"10.1016/j.isprsjprs.2025.12.015","url":null,"abstract":"<div><div>The processing of high-density LiDAR point clouds presents significant computational challenges for DEM generation due to data redundancy and topographic feature degradation during simplification. To overcome this problem, this study proposes a Terrain-Feature Retention and Spatial Uniformity Balancing (TFRSUB) method that integrates three key innovations: (i) a Distance-Geometric Synergy Index (DGSI) combining orthogonal deviation distance and point sampling interval to mitigate boundary contraction artifacts; (ii) a Composite Terrain Factor (CTF) synthesizing multiple terrain parameters to characterize diverse topographic features; and (iii) a cluster-driven Gaussian Process Regression (GPR) framework using CTF for iterative feature point selection, optimizing the trade-off between topographic fidelity and point distribution homogeneity. Evaluated on eight high-resolution LiDAR terrain point clouds across six retention ratios, TFRSUB demonstrates significant accuracy improvements over seven state-of-the-art methods, achieving reductions of 9.22%–64.70% in DEM root mean square error, 7.80%–61.34% in mean absolute error, 16.43%–76.88% in slope error, and 28.12%–81.35% in mean curvature error. These results establish TFRSUB as an alternative solution for LiDAR point cloud simplification that maintains topographic fidelity while addressing computational storage challenges.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"232 ","pages":"Pages 389-407"},"PeriodicalIF":12.2,"publicationDate":"2025-12-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145884004","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Improvement of the consistency among long-term global land surface phenology products derived from AVHRR, MODIS, and VIIRS observations AVHRR、MODIS和VIIRS长期全球地表物候产品一致性的改进

IF 12.2 1区地球科学 Q1 GEOGRAPHY, PHYSICAL

ISPRS Journal of Photogrammetry and Remote Sensing

Pub Date : 2025-12-28 DOI: 10.1016/j.isprsjprs.2025.12.010

Yongchang Ye , Xiaoyang Zhang , Yu Shen , Khuong H. Tran , Shuai Gao , Yuxia Liu , Shuai An

Land surface phenology (LSP) has been widely derived from observations of different satellite sensors, including the Advanced Very High-Resolution Radiometer (AVHRR), the Moderate Resolution Imaging Spectroradiometer (MODIS), and the Visible Infrared Imaging Radiometer Suite (VIIRS). However, the consistency of long-term LSP products is a major concern because the time series data quality may vary greatly, particularly due to temporal gaps caused by cloud contamination, instrumental degradations (e.g., orbital drift), and other factors. Therefore, this study investigated the reconstruction of the high-quality time series of vegetation indices using a Spatiotemporal Shape-Matching Model (SSMM) and the reduction of the temporal gap impacts on LSP detections globally at the climate modeling grid (0.05°). Specifically, we first generated the climatology of a 3-day two-band Enhanced Vegetation Index (EVI2) using the MODIS Nadir Bidirectional Reflectance Distribution Function (BRDF)-Adjusted Reflectance (NBAR) dataset from 2003 to 2022. The temporal climatology EVI2 was used in the SSMM algorithm to fuse the 3-day time series of EVI2 data derived separately from five different surface reflectance products: AVHRR reflectance data (1981–2019), MODIS standard surface reflectance (SSR) and NBAR data (2000–2023), and VIIRS SSR and NBAR data (2012–2023). These five sets of EVI2 time series were further applied to detect LSP metrics. The result indicates that the coefficient of determination (R²) increased by up to 0.2 among the fused EVI2 time series from AVHRR, MODIS SSR, VIIRS SSR, MODIS NBAR, and VIIRS NBAR compared to that among the raw EVI2 time series. Although the AVHRR EVI2 dataset was more consistent with MODIS SSR or VIIRS SSR observations than with MODIS NBAR or VIIRS NBAR datasets, the highest R² was found between MODIS and VIIRS NBAR EVI2, especially between their fused EVI2 time series. Consequently, the mean absolute difference (MAD) of LSP metrics was reduced by one to three days in comparing fused EVI2 with raw EVI2 time series between two different sensors. Overall, the highest LSP consistency was found between fused MODIS NBAR and fused VIIRS NBAR, which was followed by LSP detections between raw MODIS NBAR and raw VIIRS NBAR, fused MODIS SSR and fused AVHRR, raw MODIS SSR and raw AVHRR, fused MODIS NBAR and fused AVHRR, and raw MODIS NBAR and raw AVHRR. The result suggests that long-term LSP products from 1980 forward should be generated using the fused EVI2 time series from AVHRR, MODIS SSR, and VIIRS SSR, while the product from 2000 forward should be produced using the fused time series from MODIS NBAR and VIIRS NBAR observations.

陆地表面物候学（LSP）广泛来源于不同卫星传感器的观测，包括先进高分辨率辐射计（AVHRR）、中分辨率成像光谱仪（MODIS）和可见红外成像辐射计套件（VIIRS）。然而，长期LSP产品的一致性是一个主要问题，因为时间序列数据质量可能会有很大变化，特别是由于云污染、仪器退化（例如轨道漂移）和其他因素造成的时间间隙。因此，本研究利用时空形状匹配模型（spatial - temporal Shape-Matching Model， SSMM）重建植被指数的高质量时间序列，并在气候模拟网格（0.05°）下降低时间间隙对LSP检测的影响。具体而言，我们首先使用MODIS Nadir双向反射率分布函数(BRDF)-调整反射率（NBAR）数据集生成了2003 - 2022年3天两波段增强型植被指数（EVI2）的气候。利用时间气候学数据EVI2在SSMM算法中融合了5种不同地表反射率产品分别获得的3 d时间序列数据：AVHRR反射率数据（1981-2019）、MODIS标准地表反射率（SSR）和NBAR数据（2000-2023）、VIIRS SSR和NBAR数据（2012-2023）。这五组EVI2时间序列进一步用于检测LSP度量。结果表明，与原始EVI2时间序列相比，AVHRR、MODIS SSR、VIIRS SSR、MODIS NBAR和VIIRS NBAR融合的EVI2时间序列的决定系数（R2）提高了0.2。虽然AVHRR EVI2数据集与MODIS SSR或VIIRS SSR观测值的一致性高于MODIS NBAR或VIIRS NBAR数据集，但MODIS和VIIRS NBAR EVI2之间的R2最高，尤其是两者融合的EVI2时间序列之间的R2最高。因此，在比较两个不同传感器之间融合的EVI2和原始EVI2时间序列时，LSP度量的平均绝对差（MAD）减少了一到三天。总体而言，融合MODIS NBAR与融合virs NBAR之间的LSP一致性最高，其次是原始MODIS NBAR与原始virs NBAR、融合MODIS SSR与融合AVHRR、原始MODIS SSR与原始AVHRR、融合MODIS NBAR与融合AVHRR、原始MODIS NBAR与融合AVHRR以及原始MODIS NBAR与原始AVHRR之间的LSP一致性。结果表明，1980年以后的长期LSP产品应采用AVHRR、MODIS SSR和VIIRS SSR数据融合的EVI2时间序列生成，2000年以后的产品应采用MODIS NBAR和VIIRS NBAR数据融合的时间序列生成。

{"title":"Improvement of the consistency among long-term global land surface phenology products derived from AVHRR, MODIS, and VIIRS observations","authors":"Yongchang Ye , Xiaoyang Zhang , Yu Shen , Khuong H. Tran , Shuai Gao , Yuxia Liu , Shuai An","doi":"10.1016/j.isprsjprs.2025.12.010","DOIUrl":"10.1016/j.isprsjprs.2025.12.010","url":null,"abstract":"<div><div>Land surface phenology (LSP) has been widely derived from observations of different satellite sensors, including the Advanced Very High-Resolution Radiometer (AVHRR), the Moderate Resolution Imaging Spectroradiometer (MODIS), and the Visible Infrared Imaging Radiometer Suite (VIIRS). However, the consistency of long-term LSP products is a major concern because the time series data quality may vary greatly, particularly due to temporal gaps caused by cloud contamination, instrumental degradations (e.g., orbital drift), and other factors. Therefore, this study investigated the reconstruction of the high-quality time series of vegetation indices using a Spatiotemporal Shape-Matching Model (SSMM) and the reduction of the temporal gap impacts on LSP detections globally at the climate modeling grid (0.05°). Specifically, we first generated the climatology of a 3-day two-band Enhanced Vegetation Index (EVI2) using the MODIS Nadir Bidirectional Reflectance Distribution Function (BRDF)-Adjusted Reflectance (NBAR) dataset from 2003 to 2022. The temporal climatology EVI2 was used in the SSMM algorithm to fuse the 3-day time series of EVI2 data derived separately from five different surface reflectance products: AVHRR reflectance data (1981–2019), MODIS standard surface reflectance (SSR) and NBAR data (2000–2023), and VIIRS SSR and NBAR data (2012–2023). These five sets of EVI2 time series were further applied to detect LSP metrics. The result indicates that the coefficient of determination (R<sup>2</sup>) increased by up to 0.2 among the fused EVI2 time series from AVHRR, MODIS SSR, VIIRS SSR, MODIS NBAR, and VIIRS NBAR compared to that among the raw EVI2 time series. Although the AVHRR EVI2 dataset was more consistent with MODIS SSR or VIIRS SSR observations than with MODIS NBAR or VIIRS NBAR datasets, the highest R<sup>2</sup> was found between MODIS and VIIRS NBAR EVI2, especially between their fused EVI2 time series. Consequently, the mean absolute difference (MAD) of LSP metrics was reduced by one to three days in comparing fused EVI2 with raw EVI2 time series between two different sensors. Overall, the highest LSP consistency was found between fused MODIS NBAR and fused VIIRS NBAR, which was followed by LSP detections between raw MODIS NBAR and raw VIIRS NBAR, fused MODIS SSR and fused AVHRR, raw MODIS SSR and raw AVHRR, fused MODIS NBAR and fused AVHRR, and raw MODIS NBAR and raw AVHRR. The result suggests that long-term LSP products from 1980 forward should be generated using the fused EVI2 time series from AVHRR, MODIS SSR, and VIIRS SSR, while the product from 2000 forward should be produced using the fused time series from MODIS NBAR and VIIRS NBAR observations.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"232 ","pages":"Pages 369-388"},"PeriodicalIF":12.2,"publicationDate":"2025-12-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145845309","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Synthetic learning for primitive-based building model reconstruction from point clouds 基于点云的建筑模型重建的综合学习

IF 12.2 1区地球科学 Q1 GEOGRAPHY, PHYSICAL

ISPRS Journal of Photogrammetry and Remote Sensing

Pub Date : 2025-12-27 DOI: 10.1016/j.isprsjprs.2025.12.012

Zhixin Li, Jie Shan

The rapid advancement of digital 3D environments has significantly increased the demand for geometrically accurate and semantically rich parametric building models. However, existing primitive- or model-based building reconstruction approaches often struggle with limited availability of labeled datasets and insufficient reconstruction accuracy. To address these challenges, we propose a novel learning-based method for building reconstruction from point clouds that leverages roof primitives and relies exclusively on synthetic data for supervision. Our approach begins with the generation of a large synthetic dataset comprising 100,000 buildings of varying scales based on a predefined library of 10 roof primitive classes. The synthetic point clouds are created by randomly sampling not only the interiors but also the edges and corners of the roof primitives. Two lightweight transformer-based neural networks are then trained to classify roof primitive classes and estimate their corresponding parameters. Compared to conventional learning-free fitting methods, our learning-based approach achieves higher parameter estimation accuracy and greater robustness when applied to six real-world point cloud datasets collected from drone, airborne, and spaceborne platforms. Notably, the synthetic learning approach reduces primitive parameter estimation errors from approximately 50% to 6% of the point ground spacing — demonstrating a distinctive advantage when trained effectively on synthetic data. Future work may explore generating synthetic data for irregular, complex buildings, expanding the library with additional roof primitive classes, and applying the proposed training strategy to such synthetic datasets.

数字三维环境的快速发展大大增加了对几何精度和语义丰富的参数化建筑模型的需求。然而，现有的基于原语或模型的建筑重建方法经常受到标记数据集可用性有限和重建精度不足的困扰。为了应对这些挑战，我们提出了一种基于学习的点云重建方法，该方法利用屋顶原语并完全依赖合成数据进行监督。我们的方法从生成一个大型合成数据集开始，该数据集包含10个屋顶原始类的预定义库，其中包含100,000个不同规模的建筑物。合成点云是通过随机采样创建的，不仅是内部，还有屋顶原语的边缘和角落。然后训练两个基于轻量级变压器的神经网络对屋顶原语类进行分类并估计其相应的参数。与传统的无学习拟合方法相比，我们的基于学习的方法在应用于从无人机、机载和星载平台收集的六个现实世界点云数据集时，实现了更高的参数估计精度和更强的鲁棒性。值得注意的是，合成学习方法将原始参数估计误差从大约50%减少到6%的点地间距-在合成数据上有效训练时显示出独特的优势。未来的工作可能会探索为不规则、复杂的建筑生成合成数据，用额外的屋顶原语类扩展库，并将提出的训练策略应用于这些合成数据集。

{"title":"Synthetic learning for primitive-based building model reconstruction from point clouds","authors":"Zhixin Li, Jie Shan","doi":"10.1016/j.isprsjprs.2025.12.012","DOIUrl":"10.1016/j.isprsjprs.2025.12.012","url":null,"abstract":"<div><div>The rapid advancement of digital 3D environments has significantly increased the demand for geometrically accurate and semantically rich parametric building models. However, existing primitive- or model-based building reconstruction approaches often struggle with limited availability of labeled datasets and insufficient reconstruction accuracy. To address these challenges, we propose a novel learning-based method for building reconstruction from point clouds that leverages roof primitives and relies exclusively on synthetic data for supervision. Our approach begins with the generation of a large synthetic dataset comprising 100,000 buildings of varying scales based on a predefined library of 10 roof primitive classes. The synthetic point clouds are created by randomly sampling not only the interiors but also the edges and corners of the roof primitives. Two lightweight transformer-based neural networks are then trained to classify roof primitive classes and estimate their corresponding parameters. Compared to conventional learning-free fitting methods, our learning-based approach achieves higher parameter estimation accuracy and greater robustness when applied to six real-world point cloud datasets collected from drone, airborne, and spaceborne platforms. Notably, the synthetic learning approach reduces primitive parameter estimation errors from approximately 50% to 6% of the point ground spacing — demonstrating a distinctive advantage when trained effectively on synthetic data. Future work may explore generating synthetic data for irregular, complex buildings, expanding the library with additional roof primitive classes, and applying the proposed training strategy to such synthetic datasets.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"232 ","pages":"Pages 336-352"},"PeriodicalIF":12.2,"publicationDate":"2025-12-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145845311","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0