Pub Date : 2025-02-01DOI: 10.1016/j.jag.2024.104315
Yun Luo , Shiliang Su
A wide variety of spatially explicit modeling algorithms has recently mushroomed in geoinformation research. These algorithms establish local models with data from spatially confined subsets, thereby offering a new impetus for addressing the issue of spatiotemporal non-stationarity. However, a significant challenge persists in literature that local models are primarily predicated on linear assumptions, limiting their capacity to capture the non-linear relationships prevalent in real-world geographical phenomena. This study remedies this gap through proposing a novel approach that integrates the bagging and stacking approaches of ensemble learning into the spatially explicit modeling framework. We specifically develop the SpatioTemporal Random Forest (STRF) and SpatioTemporal Stacking Tree (STST) algorithms1, which capture and interpret the non-linearity in the spatial and temporal context more effectively. Additionally, we introduce the ‘local importance score’ and ‘spatiotemporally accumulated local effects’ as novel interpretable metrics for visualizing and unraveling the dynamics of non-stationarity in spatial analyses. Simulation and real data experiments validate that the STRF and STST outperform over traditional spatially explicit modeling algorithms to a large content. This study contributes to the methodological innovation of spatially explicit modeling by bringing the nonlinearity in spatiotemporal non-stationarity to the fore.
{"title":"SpatioTemporal Random Forest and SpatioTemporal Stacking Tree: A novel spatially explicit ensemble learning approach to modeling non-linearity in spatiotemporal non-stationarity","authors":"Yun Luo , Shiliang Su","doi":"10.1016/j.jag.2024.104315","DOIUrl":"10.1016/j.jag.2024.104315","url":null,"abstract":"<div><div>A wide variety of spatially explicit modeling algorithms has recently mushroomed in geoinformation research. These algorithms establish local models with data from spatially confined subsets, thereby offering a new impetus for addressing the issue of spatiotemporal non-stationarity. However, a significant challenge persists in literature that local models are primarily predicated on linear assumptions, limiting their capacity to capture the non-linear relationships prevalent in real-world geographical phenomena. This study remedies this gap through proposing a novel approach that integrates the bagging and stacking approaches of ensemble learning into the spatially explicit modeling framework. We specifically develop the SpatioTemporal Random Forest (STRF) and SpatioTemporal Stacking Tree (STST) algorithms<span><span><sup>1</sup></span></span>, which capture and interpret the non-linearity in the spatial and temporal context more effectively. Additionally, we introduce the ‘local importance score’ and ‘spatiotemporally accumulated local effects’ as novel interpretable metrics for visualizing and unraveling the dynamics of non-stationarity in spatial analyses. Simulation and real data experiments validate that the STRF and STST outperform over traditional spatially explicit modeling algorithms to a large content. This study contributes to the methodological innovation of spatially explicit modeling by bringing the nonlinearity in spatiotemporal non-stationarity to the fore.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104315"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142825360","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-01DOI: 10.1016/j.jag.2024.104335
Jiangfan Feng , Hongxin Luo , Zhujun Gu
Remote Sensing Image (RSI) scene classification plays a pivotal role in diverse applications such as land cover mapping, urban planning, and environmental monitoring. Traditional deep learning approaches, however, heavily rely on abundant labeled data, which is often costly and challenging to acquire. Semi-supervised learning emerges as a cost-effective alternative, yet existing methods frequently overlook the intricate characteristics of remote sensing data, such as multi-scale features and complex spatial patterns, thereby limiting their ability to effectively address these challenges. In this study, a novel Scene Semi-Supervised Method (SSSM) is introduced, marking a significant advancement in both network architecture and semi-supervised techniques. At the core of the SSSM framework lies the Multi-Level Feature Fusion Network (MFFN), meticulously designed to extract and integrate complex features from remote sensing data across diverse scales and locations. To optimize the utilization of pseudo-labels and minimize mislabeling, the Pseudo-Label Multi-Level Sampling strategy (PMLS) is proposed, a probabilistic approach that selectively identifies high-quality pseudo-labels to enhance training. Rigorous experiments conducted on three benchmark datasets reveal that the SSSM method significantly improves classification accuracy, achieving an increase of 3%–5% on a specific dataset compared to existing approaches. This accomplishment underscores the effectiveness of the MFFN design and the semi-supervised strategy in tackling the complexities of remote sensing scene classification. In summary, the MFFN-driven pseudo-label framework presented in this research pioneers a cutting-edge and promising new direction for semi-supervised remote sensing scene classification.
{"title":"Improving semi-supervised remote sensing scene classification via Multilevel Feature Fusion and pseudo-labeling","authors":"Jiangfan Feng , Hongxin Luo , Zhujun Gu","doi":"10.1016/j.jag.2024.104335","DOIUrl":"10.1016/j.jag.2024.104335","url":null,"abstract":"<div><div>Remote Sensing Image (RSI) scene classification plays a pivotal role in diverse applications such as land cover mapping, urban planning, and environmental monitoring. Traditional deep learning approaches, however, heavily rely on abundant labeled data, which is often costly and challenging to acquire. Semi-supervised learning emerges as a cost-effective alternative, yet existing methods frequently overlook the intricate characteristics of remote sensing data, such as multi-scale features and complex spatial patterns, thereby limiting their ability to effectively address these challenges. In this study, a novel Scene Semi-Supervised Method (SSSM) is introduced, marking a significant advancement in both network architecture and semi-supervised techniques. At the core of the SSSM framework lies the Multi-Level Feature Fusion Network (MFFN), meticulously designed to extract and integrate complex features from remote sensing data across diverse scales and locations. To optimize the utilization of pseudo-labels and minimize mislabeling, the Pseudo-Label Multi-Level Sampling strategy (PMLS) is proposed, a probabilistic approach that selectively identifies high-quality pseudo-labels to enhance training. Rigorous experiments conducted on three benchmark datasets reveal that the SSSM method significantly improves classification accuracy, achieving an increase of 3%–5% on a specific dataset compared to existing approaches. This accomplishment underscores the effectiveness of the MFFN design and the semi-supervised strategy in tackling the complexities of remote sensing scene classification. In summary, the MFFN-driven pseudo-label framework presented in this research pioneers a cutting-edge and promising new direction for semi-supervised remote sensing scene classification.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104335"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142901790","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-01DOI: 10.1016/j.jag.2025.104364
Yanan Jiang , Qiang Xu , Ran Meng , Chao Zhang , Linfeng Zheng , Zhong Lu
The Mountain Excavation and City Construction project (MECC) in Yan’an New District (YND) on the Chinese Loess Plateau is one of the largest geotechnical works globally. Ground deformation resulting from these extensive earthworks continues to evolve spatially and temporally even after construction is completed. Monitoring this deformation is crucial for understanding uneven post-construction subsidence and ensuring the structural integrity of infrastructure. This study proposes a framework for monitoring and predicting post-construction ground settlement (PCGS) using a dual-polarization Multi-temporal InSAR method (dual-pol MT-InSAR) and Self-Attention Memory Convolutional Long Short-Term Memory (SAM-ConvLSTM) model. Compared to single-polarization (single-pol) MT-InSAR methods, the dual-pol MT-InSAR approach, which utilizes both polarization channels of Sentinel-1 (S1) SAR data, achieves a 24 % increase in Permanent Scatterer (PS) density for PS-InSAR and improves average coherence while reducing coherence standard deviation for Small Baseline Subset (SBAS). The study further examines the factors contributing to uneven ground deformation, including fill and excavation activities (e.g., the thickness and geotechnical properties of loess), construction activities and surface loads, and precipitation. A consolidation settlement model is employed to simulate and assess ground settlement decay due to loess compaction. Based on this analysis, the most affected area in Qiaoergou is selected for spatiotemporal forecasting using MT-InSAR measurements and the SAM-ConvLSTM model. The results indicate that regions with significant subsidence form a characteristic funnel shape, with subsidence increasing over time and the deformation perimeter expanding outward. The model achieved an average absolute error of 1.6 mm, with the majority of errors concentrated within 5 mm.
{"title":"Remote sensing characterizing and deformation predicting of Yan'an New District’s Mountain Excavation and City Construction with dual-polarization MT-InSAR method","authors":"Yanan Jiang , Qiang Xu , Ran Meng , Chao Zhang , Linfeng Zheng , Zhong Lu","doi":"10.1016/j.jag.2025.104364","DOIUrl":"10.1016/j.jag.2025.104364","url":null,"abstract":"<div><div>The Mountain Excavation and City Construction project (MECC) in Yan’an New District (YND) on the Chinese Loess Plateau is one of the largest geotechnical works globally. Ground deformation resulting from these extensive earthworks continues to evolve spatially and temporally even after construction is completed. Monitoring this deformation is crucial for understanding uneven post-construction subsidence and ensuring the structural integrity of infrastructure. This study proposes a framework for monitoring and predicting post-construction ground settlement (PCGS) using a dual-polarization Multi-temporal InSAR method (dual-pol MT-InSAR) and Self-Attention Memory Convolutional Long Short-Term Memory (SAM-ConvLSTM) model. Compared to single-polarization (single-pol) MT-InSAR methods, the dual-pol MT-InSAR approach, which utilizes both polarization channels of Sentinel-1 (S1) SAR data, achieves a 24 % increase in Permanent Scatterer (PS) density for PS-InSAR and improves average coherence while reducing coherence standard deviation for Small Baseline Subset (SBAS). The study further examines the factors contributing to uneven ground deformation, including fill and excavation activities (e.g., the thickness and geotechnical properties of loess), construction activities and surface loads, and precipitation. A consolidation settlement model is employed to simulate and assess ground settlement decay due to loess compaction. Based on this analysis, the most affected area in Qiaoergou is selected for spatiotemporal forecasting using MT-InSAR measurements and the SAM-ConvLSTM model. The results indicate that regions with significant subsidence form a characteristic funnel shape, with subsidence increasing over time and the deformation perimeter expanding outward. The model achieved an average absolute error of 1.6 mm, with the majority of errors concentrated within 5 mm.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104364"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142990327","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-01DOI: 10.1016/j.jag.2025.104392
Zijian Lu , Xueyan Zhu , Jinfeng Li , Mingyue Li , Jie Wang , Wenqiang Wang , Yili Zheng , Qianggong Zhang
Global warming has accelerated glacier retreat, subsequently leading to the formation of glacial lakes in high-altitude mountainous regions. These lakes represent emerging ecological water systems and could potentially pose significant hazards. Observations of these systems are constrained by their remote locations and the lack of cost-effective monitoring methods, resulting in limited understanding of their dynamics. In this study, we synchronized surveillance monitoring with in-situ water quality measurements at a typical high-altitude glacial lake on the Qinghai-Tibet Plateau. We aim to use images from surveillance cameras to estimate the turbidity parameter, a key indicator of changes in the water environment and the impacts of climate change on high-altitude ecosystems. We segmented RGB images and applied regression modeling with field-measured water turbidity data, and then used deep learning models to accurately estimates turbidity levels and their changes. Our study demonstrates the potential of RGB imagery and deep learning for the long-term, continuous, and high-resolution monitoring of water quality in remote glacial lakes. It presents a novel and cost-effective approach for monitoring these newly emerged and swiftly changing water systems at high altitudes, offering a significant advancement in tracking environmental changes in these critical high mountain regions.
{"title":"Detecting glacial lake water quality indicators from RGB surveillance images via deep learning","authors":"Zijian Lu , Xueyan Zhu , Jinfeng Li , Mingyue Li , Jie Wang , Wenqiang Wang , Yili Zheng , Qianggong Zhang","doi":"10.1016/j.jag.2025.104392","DOIUrl":"10.1016/j.jag.2025.104392","url":null,"abstract":"<div><div>Global warming has accelerated glacier retreat, subsequently leading to the formation of glacial lakes in high-altitude mountainous regions. These lakes represent emerging ecological water systems and could potentially pose significant hazards. Observations of these systems are constrained by their remote locations and the lack of cost-effective monitoring methods, resulting in limited understanding of their dynamics. In this study, we synchronized surveillance monitoring with in-situ water quality measurements at a typical high-altitude glacial lake on the Qinghai-Tibet Plateau. We aim to use images from surveillance cameras to estimate the turbidity parameter, a key indicator of changes in the water environment and the impacts of climate change on high-altitude ecosystems. We segmented RGB images and applied regression modeling with field-measured water turbidity data, and then used deep learning models to accurately estimates turbidity levels and their changes. Our study demonstrates the potential of RGB imagery and deep learning for the long-term, continuous, and high-resolution monitoring of water quality in remote glacial lakes. It presents a novel and cost-effective approach for monitoring these newly emerged and swiftly changing water systems at high altitudes, offering a significant advancement in tracking environmental changes in these critical high mountain regions.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104392"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143083305","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-01DOI: 10.1016/j.jag.2025.104398
Xiaoyong Ming , Yichao Tian , Qiang Zhang , Yali Zhang , Jin Tao , Junliang Lin
Tidal flats represent one of the Earth’s most critical ecosystems characterized by substantial ecological value, but these areas are also among the most fragile ecosystems. A detailed topography survey of tidal flat is essential for exploring how tidal flat ecosystems respond to environmental changes and for predicting morphological shifts, thereby impacting the protection and restoration of mangrove ecosystems. However, there is still a dearth of data available for mangrove tidal flat topography, as the majority of measurements primarily rely on traditional cartographic methods or small-scale surveys. Therefore, we aim to rely entirely on Earth observation satellite platforms, combining satellite-based Light Detection and Ranging (LiDAR) and optical remote sensing to monitor extensive mangrove tidal flat topography. This methodology was rigorously applied and validated on China’s largest and most representative mangrove tidal flats, revealing a Root Mean Square Error (RMSE) not exceeding 7.5 cm and an R-squared value surpassing 0.89 when compared to airborne LiDAR data. We use the inundation frequency derived from the long-term Sentinel-2 image sequences and elevation data extracted from the Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) to establish a specific relationship between inundation frequency and ground elevation using both classical and generalized regression models, a mangrove tidal flat topography covering 76.9 km2 was generated. Our findings delineate suitable distribution areas for mangroves in the Maowei Sea, covering an expansive 18.2 km2.
{"title":"Coupling ICESat-2 and Sentinel-2 data for inversion of mangrove tidal flat to predict future distribution pattern of mangroves","authors":"Xiaoyong Ming , Yichao Tian , Qiang Zhang , Yali Zhang , Jin Tao , Junliang Lin","doi":"10.1016/j.jag.2025.104398","DOIUrl":"10.1016/j.jag.2025.104398","url":null,"abstract":"<div><div>Tidal flats represent one of the Earth’s most critical ecosystems characterized by substantial ecological value, but these areas are also among the most fragile ecosystems. A detailed topography survey of tidal flat is essential for exploring how tidal flat ecosystems respond to environmental changes and for predicting morphological shifts, thereby impacting the protection and restoration of mangrove ecosystems. However, there is still a dearth of data available for mangrove tidal flat topography, as the majority of measurements primarily rely on traditional cartographic methods or small-scale surveys. Therefore, we aim to rely entirely on Earth observation satellite platforms, combining satellite-based Light Detection and Ranging (LiDAR) and optical remote sensing to monitor extensive mangrove tidal flat topography. This methodology was rigorously applied and validated on China’s largest and most representative mangrove tidal flats, revealing a Root Mean Square Error (RMSE) not exceeding 7.5 cm and an R-squared value surpassing 0.89 when compared to airborne LiDAR data. We use the inundation frequency derived from the long-term Sentinel-2 image sequences and elevation data extracted from the Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) to establish a specific relationship between inundation frequency and ground elevation using both classical and generalized regression models, a mangrove tidal flat topography covering 76.9 km<sup>2</sup> was generated. Our findings delineate suitable distribution areas for mangroves in the Maowei Sea, covering an expansive 18.2 km<sup>2</sup>.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104398"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143271175","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-01DOI: 10.1016/j.jag.2025.104402
Zugang Chen , Shaohua Wang , Kai Wu , Guoqing Li , Jing Li , Jian Wang
Encoding geospatial location is a fundamental problem for geospatial artificial intelligence (GeoAI) research. In recent years, some methods (such as Place2Vec, Space2Vec, and Sphere2Vec) were proposed to encode geospatial point as a high-dimensional vector. However, all these geospatial location encoders were designed to generate a real number vector. So, when applied to some of the brain-inspired neural networks, such as Hierarchical Temporal Memory (HTM), which required the input of a binary vector, the existing methods failed. To solve the problem, based on the research from neuroscience about place cell, we proposed a new geospatial location encoding method called PlaceField2BVec. The method used the place field model to encode a location. The place field was represented by the summation of four Gaussian functions, allowing it to be stretched or divided into multiple fields as the geospatial space expanded. Then we created an HTM and devised an experiment that simulated rats moving on tables of varying sizes. The moving trajectories were encoded by PlaceField2BVec and input to the HTM. After training, we found that the artificial neurons of HTM formed a place field similar to those of hippocampal neurons in the rat brain and the distribution patterns of the place field from the two kinds of neurons were consistent. At last, our method was compared with existing Space2BVec and Buffer2BVec in terms of location prediction accuracy and to demonstrate the robustness of the binary vector encoding methods, two brain-inspired artificial neural networks— HTM and BinaryLSTM were used. The result showed that, for HTM, in smaller geospatial space the PlaceField2BVec and Buffer2BVec had about the same accuracy on average but the highest accuracy of PlaceField2BVec is 100 %; when the geospatial space extended, our method had the highest accuracy and the average accuracy of PlaceField2BVec, Space2BVec, and Buffer2BVec is 83.9 %, 25.2 % and 69.7 % after 20 times’ training. For BinaryLSTM, PlaceField2BVec always had the highest accuracy in location prediction although the accuracy decreased as the space extended. Our research can be utilized for machine self-localization, navigation, and location-related GeoAI applications, and it also contributes to the theory of cognitive maps.
{"title":"PlaceField2BVec: A bionic geospatial location encoding method for hierarchical temporal memory model","authors":"Zugang Chen , Shaohua Wang , Kai Wu , Guoqing Li , Jing Li , Jian Wang","doi":"10.1016/j.jag.2025.104402","DOIUrl":"10.1016/j.jag.2025.104402","url":null,"abstract":"<div><div>Encoding geospatial location is a fundamental problem for geospatial artificial intelligence (GeoAI) research. In recent years, some methods (such as Place2Vec, Space2Vec, and Sphere2Vec) were proposed to encode geospatial point as a high-dimensional vector. However, all these geospatial location encoders were designed to generate a real number vector. So, when applied to some of the brain-inspired neural networks, such as Hierarchical Temporal Memory (HTM), which required the input of a binary vector, the existing methods failed. To solve the problem, based on the research from neuroscience about place cell, we proposed a new geospatial location encoding method called PlaceField2BVec. The method used the place field model to encode a location. The place field was represented by the summation of four Gaussian functions, allowing it to be stretched or divided into multiple fields as the geospatial space expanded. Then we created an HTM and devised an experiment that simulated rats moving on tables of varying sizes. The moving trajectories were encoded by PlaceField2BVec and input to the HTM. After training, we found that the artificial neurons of HTM formed a place field similar to those of hippocampal neurons in the rat brain and the distribution patterns of the place field from the two kinds of neurons were consistent. At last, our method was compared with existing Space2BVec and Buffer2BVec in terms of location prediction accuracy and to demonstrate the robustness of the binary vector encoding methods, two brain-inspired artificial neural networks— HTM and BinaryLSTM were used. The result showed that, for HTM, in smaller geospatial space the PlaceField2BVec and Buffer2BVec had about the same accuracy on average but the highest accuracy of PlaceField2BVec is 100 %; when the geospatial space extended, our method had the highest accuracy and the average accuracy of PlaceField2BVec, Space2BVec, and Buffer2BVec is 83.9 %, 25.2 % and 69.7 % after 20 times’ training. For BinaryLSTM, PlaceField2BVec always had the highest accuracy in location prediction although the accuracy decreased as the space extended. Our research can be utilized for machine self-localization, navigation, and location-related GeoAI applications, and it also contributes to the theory of cognitive maps.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104402"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143271178","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-01DOI: 10.1016/j.jag.2024.104321
Huifu Zhuang , Jianlin Guo , Ming Hao , Sen Du , Kefei Zhang , Xuesong Wang
Due to the significant disparities in feature spaces of multi-source images, change detection (CD) of heterogeneous remote sensing images (HRSIs) remains a highly challenging problem. Currently, CD methods based on domain transfer networks (DTNs) have garnered significant attention. However, the computer scientists underutilize knowledge in the field of CD during DTNs design, and the existing CD methods do not fully utilize the heterogeneous complementary features contained in HRSIs. Therefore, this study proposes a novel CD method based on multiple pseudo-homogeneous image pairs. First, a cycle-consistent generative adversarial network with knowledge constraints (named as KCGAN) was designed for obtaining good pseudo-homogeneous images. In detail, both the domain knowledge that there are land cover changes in multi-temporal images and that the objects in an image can be described from different scales were well integrated into the design of KCGAN. Then, a multi-modal difference Siamese fusion network (named as MDSiamF) was proposed to extract change information from the multiple pseudo-homogeneous image pairs generated with KCGAN. Experiments conducted on three datasets showed that: 1) compared to existing domain transfer methods, the unchanged areas in the pseudo-homogeneous images obtained by KCGAN exhibit better feature consistency (with a peak signal-to-noise ratio higher than 20.85 and a PHash value higher than 0.9); 2) compared to state-of-the-art methods for CD of HRSIs, the proposed method shows stable and good CD performance (with an overall accuracy higher than 0.98 and a F1 Score higher than 0.78).
由于多源图像的特征空间差异巨大，异质遥感图像（HRSI）的变化检测（CD）仍然是一个极具挑战性的问题。目前，基于域转移网络（DTN）的变化检测方法已引起广泛关注。然而，计算机科学家在设计 DTNs 时对 CD 领域的知识利用不足，现有的 CD 方法也没有充分利用 HRSIs 中包含的异构互补特征。因此，本研究提出了一种基于多伪同质图像对的新型 CD 方法。首先，为获得良好的伪同质图像，设计了一个具有知识约束的循环一致性生成对抗网络（命名为 KCGAN）。具体而言，在 KCGAN 的设计中很好地融入了多时相图像中存在土地覆盖变化以及图像中的物体可以从不同尺度进行描述这两个领域知识。然后，提出了一种多模态差分连体融合网络（命名为 MDSiamF），用于从 KCGAN 生成的多伪同质图像对中提取变化信息。在三个数据集上进行的实验表明1）与现有的域转移方法相比，KCGAN 获得的伪同质图像中的不变区域表现出更好的特征一致性（峰值信噪比高于 20.85，PHash 值高于 0.9）；2）与最先进的 HRSI CD 方法相比，所提出的方法表现出稳定而良好的 CD 性能（总体准确率高于 0.98，F1 分数高于 0.78）。
{"title":"Change detection in heterogeneous images based on multiple pseudo-homogeneous image pairs","authors":"Huifu Zhuang , Jianlin Guo , Ming Hao , Sen Du , Kefei Zhang , Xuesong Wang","doi":"10.1016/j.jag.2024.104321","DOIUrl":"10.1016/j.jag.2024.104321","url":null,"abstract":"<div><div>Due to the significant disparities in feature spaces of multi-source images, change detection (CD) of heterogeneous remote sensing images (HRSIs) remains a highly challenging problem. Currently, CD methods based on domain transfer networks (DTNs) have garnered significant attention. However, the computer scientists underutilize knowledge in the field of CD during DTNs design, and the existing CD methods do not fully utilize the heterogeneous complementary features contained in HRSIs. Therefore, this study proposes a novel CD method based on multiple pseudo-homogeneous image pairs. First, a cycle-consistent generative adversarial network with knowledge constraints (named as KCGAN) was designed for obtaining good pseudo-homogeneous images. In detail, both the domain knowledge that there are land cover changes in multi-temporal images and that the objects in an image can be described from different scales were well integrated into the design of KCGAN. Then, a multi-modal difference Siamese fusion network (named as MDSiamF) was proposed to extract change information from the multiple pseudo-homogeneous image pairs generated with KCGAN. Experiments conducted on three datasets showed that: 1) compared to existing domain transfer methods, the unchanged areas in the pseudo-homogeneous images obtained by KCGAN exhibit better feature consistency (with a peak signal-to-noise ratio higher than 20.85 and a PHash value higher than 0.9); 2) compared to state-of-the-art methods for CD of HRSIs, the proposed method shows stable and good CD performance (with an overall accuracy higher than 0.98 and a F1 Score higher than 0.78).</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104321"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142825342","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-01DOI: 10.1016/j.jag.2024.104333
Liang Liang , Jian Yang , William C. Wittenbraker , Ellen V. Crocker , Monika A. Tomaszewska , Geoffrey M. Henebry
Many invasive shrubs in the eastern deciduous forests of the United States use the temporal niche before and after the native tree canopy leaf-on period (leafing out prior to most native species and retaining leaves after most natives senesce) to establish in the light-limited environment of the understory. To support an increased understanding of invasive shrub species’ ecology and distribution patterns and inform better management plans, this key phenological difference needs to be characterized in detail. Here we leveraged the high-resolution observations from the French-Israel VENµS mission to examine the phenological characteristics of a widespread invasive shrub species—Amur honeysuckle (AH; Lonicera maackii (Rupr.) Herder)—compared to native deciduous trees in Robinson Forest, Kentucky. VENµS offered daily superspectral (12 narrow bands) observations at 4 m resolution in a limited number of global sites, providing us with crucial data for the analysis. We identified three forest communities with respect to AH presence through field surveys (i.e., uninvaded forest stands, forest stands with AH understory, and AH shrub thickets) and compared their VENµS-derived spectral signatures and time series of vegetation indices. In 2023, AH shrub thickets greened up one month earlier than uninvaded forest stands (mid-March vs. mid-April). AH leaf growth advanced into full green before the canopy tree greenup started in early April, marking an optimal window for isolating areas with AH understory from the uninvaded forest using remote sensing. Based on the phenological differences identified, we predicted the distribution of AH in the study area using a two-date differencing model and a spectral mixture analysis. Our detailed findings using VENµS data offer insights into the temporal dynamics of invasive shrubs and native trees in a typical eastern deciduous forest. While our prediction of the AH distribution was confounded by the presence of native early greening and/or evergreen understory plants at a few locations, it was still moderately accurate (overall accuracy ∼ 70 %) and its abundance estimates agreed with observations in forest stands with minimal native understory growth. Moving forward, high-resolution remote sensing observations combined with a phenology-based approach will likely support more precise monitoring and management of invasive understory plants in native forest ecosystems.
{"title":"Characterizing phenological differences of invasive shrubs in a forest matrix using high resolution VENµS time series","authors":"Liang Liang , Jian Yang , William C. Wittenbraker , Ellen V. Crocker , Monika A. Tomaszewska , Geoffrey M. Henebry","doi":"10.1016/j.jag.2024.104333","DOIUrl":"10.1016/j.jag.2024.104333","url":null,"abstract":"<div><div>Many invasive shrubs in the eastern deciduous forests of the United States use the temporal niche before and after the native tree canopy leaf-on period (leafing out prior to most native species and retaining leaves after most natives senesce) to establish in the light-limited environment of the understory. To support an increased understanding of invasive shrub species’ ecology and distribution patterns and inform better management plans, this key phenological difference needs to be characterized in detail. Here we leveraged the high-resolution observations from the French-Israel VENµS mission to examine the phenological characteristics of a widespread invasive shrub species—Amur honeysuckle (AH; <em>Lonicera maackii</em> (Rupr.) Herder)—compared to native deciduous trees in Robinson Forest, Kentucky. VENµS offered daily superspectral (12 narrow bands) observations at 4 m resolution in a limited number of global sites, providing us with crucial data for the analysis. We identified three forest communities with respect to AH presence through field surveys (<em>i.e.,</em> uninvaded forest stands, forest stands with AH understory, and AH shrub thickets) and compared their VENµS-derived spectral signatures and time series of vegetation indices. In 2023, AH shrub thickets greened up one month earlier than uninvaded forest stands (mid-March vs. mid-April). AH leaf growth advanced into full green before the canopy tree greenup started in early April, marking an optimal window for isolating areas with AH understory from the uninvaded forest using remote sensing. Based on the phenological differences identified, we predicted the distribution of AH in the study area using a two-date differencing model and a spectral mixture analysis. Our detailed findings using VENµS data offer insights into the temporal dynamics of invasive shrubs and native trees in a typical eastern deciduous forest. While our prediction of the AH distribution was confounded by the presence of native early greening and/or evergreen understory plants at a few locations, it was still moderately accurate (overall accuracy ∼ 70 %) and its abundance estimates agreed with observations in forest stands with minimal native understory growth. Moving forward, high-resolution remote sensing observations combined with a phenology-based approach will likely support more precise monitoring and management of invasive understory plants in native forest ecosystems.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104333"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142874809","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Mangroves, critical for ecological sustainability, are challenging to map accurately due to their fragmented nature and difficult accessibility. Existing datasets, often constrained to 10 m or above resolutions, could misrepresent fragmented mangrove regions and suffer from sampling biases, limiting their regional applicability. Furthermore, scale conversion’s spatial and statistical implications on mangrove mapping accuracy and area estimation remain largely unexplored. This study proposes a novel framework that leverages UHR (0.2 m) aerial photos and the DeepLabV3+ model for fine-scale mapping and systematically simulates and quantifies scale-induced effects. The resultant 20 cm-resolution mangrove map of Hong Kong achieved an overall accuracy (OA) of 92.1 %, with up to 53 % improvement compared to various existing datasets. It delineates complex boundaries in diverse coastal settings while preserving the structural integrity of fragmented patches. The total mangrove area in Hong Kong is estimated at ∼720 ha, with Deep Bay comprising 77.5 %. The scale effects analysis revealed pronounced sensitivity in fragmented habitats, where each 1 m increase in resolution could result in an average area underestimation of 5000 m2 and up to 25 % OA degradation when transitioning from 0.2 m to 30 m. Moreover, integrating patch geometry and scale responses indicated that 6 m is the optimal scale for monitoring. Beyond this, OA could sharply decline to below 82 % at the commonly used 10 m resolution and drop as low as 66 % at 30 m. These findings highlight the critical importance of fine-scale mapping using UHR images for effective mangrove conservation and management.
{"title":"Scale effects in mangrove mapping from ultra-high-resolution remote sensing imagery","authors":"Hanwen Zhang , Shan Wei , Xindan Liang , Yiping Chen , Hongsheng Zhang","doi":"10.1016/j.jag.2024.104310","DOIUrl":"10.1016/j.jag.2024.104310","url":null,"abstract":"<div><div>Mangroves, critical for ecological sustainability, are challenging to map accurately due to their fragmented nature and difficult accessibility. Existing datasets, often constrained to 10 m or above resolutions, could misrepresent fragmented mangrove regions and suffer from sampling biases, limiting their regional applicability. Furthermore, scale conversion’s spatial and statistical implications on mangrove mapping accuracy and area estimation remain largely unexplored. This study proposes a novel framework that leverages UHR (0.2 m) aerial photos and the DeepLabV3+ model for fine-scale mapping and systematically simulates and quantifies scale-induced effects. The resultant 20 cm-resolution mangrove map of Hong Kong achieved an overall accuracy (OA) of 92.1 %, with up to 53 % improvement compared to various existing datasets. It delineates complex boundaries in diverse coastal settings while preserving the structural integrity of fragmented patches. The total mangrove area in Hong Kong is estimated at ∼720 ha, with Deep Bay comprising 77.5 %. The scale effects analysis revealed pronounced sensitivity in fragmented habitats, where each 1 m increase in resolution could result in an average area underestimation of 5000 m<sup>2</sup> and up to 25 % OA degradation when transitioning from 0.2 m to 30 m. Moreover, integrating patch geometry and scale responses indicated that 6 m is the optimal scale for monitoring. Beyond this, OA could sharply decline to below 82 % at the commonly used 10 m resolution and drop as low as 66 % at 30 m. These findings highlight the critical importance of fine-scale mapping using UHR images for effective mangrove conservation and management.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104310"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142874810","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-01DOI: 10.1016/j.jag.2024.104329
Renzhe Wu , Guoxiang Liu , Xin Bao , Jichao Lv , Age Shama , Bo Zhang , Wenfei Mao , Jie Chen , Zhihan Yang , Rui Zhang
Glacial lakes (GLs), which serve as natural reservoirs, are also prospective sources of risk, and their risk levels are continuously increasing as a result of global climate warming. Nevertheless, GLs are situated in mountainous and valley regions, which are distinguished by their complex terrain and unpredictable weather conditions. This leads to restricted availability of optical imagery as a consequence of the frequent cloud cover. Synthetic Aperture Radar (SAR), however, encounters issues with geometric distortion. This paper introduces an unsupervised method based on geometric distortion detection (without orbit state information) and historical positioning using dual-orbit SAR imagery to research GL extraction effectively. This method detects low-quality pixels from dual-orbit SAR imagery through geometric distortion. It extracts GLs using a majority voting integration of unsupervised classification algorithms constrained by historical GL center points. The Southeastern Tibetan Plateau (SETP) was chosen as a representative region for the study, and experiments were conducted from July to August 2018 using dual-orbit Sentinel-1 imagery. A total of 600 refined samples were used for comparative verification. The results demonstrate that this method is capable of reliably identifying the active and passive geometric distortions in SAR imagery. The fusion of dual-orbit SAR based on geometric distortion can effectively enhance the classification performance of remote sensing imagery and achieve the acquisition of GL water storage area during the flood season. The geometric distortion rate was reduced from 29.9% to 7.9% after fusion correction, and the accuracy, recall rate, precision, Intersection over Union (IoU), and F1-Score were 0.989, 0.900, 0.908, 0.825, and 0.904, respectively. This serves as a reference for research that investigates the mechanisms of glacier-GL-climate change.
作为天然水库的冰湖也是潜在的风险源，而且由于全球气候变暖，冰湖的风险源水平正在不断提高。然而，GLs位于山区和山谷地区，其特点是地形复杂，天气条件不可预测。由于频繁的云层覆盖，这导致光学图像的可用性受到限制。然而，合成孔径雷达（SAR）会遇到几何畸变的问题。本文提出了一种基于几何畸变检测（无轨道状态信息）和历史定位的无监督方法，利用双轨SAR图像有效地研究了GL提取。该方法通过几何畸变检测双轨SAR图像中的低质量像元。它使用受历史GL中心点约束的无监督分类算法的多数投票集成来提取GL。选择青藏高原东南部（SETP）作为研究的代表区域，于2018年7月至8月利用Sentinel-1双轨图像进行了实验。共使用600个精制样品进行对比验证。结果表明，该方法能够可靠地识别SAR图像中的主动和被动几何畸变。基于几何畸变的双轨SAR融合可以有效提高遥感影像的分类性能，实现汛期GL储水面积的获取。融合校正后的几何畸变率由29.9%降至7.9%，准确率为0.989，查全率为0.900，准确率为0.908，交叉比联合（Intersection over Union, IoU）为0.825，F1-Score为0.904。这为冰川- gl -气候变化机制的研究提供了参考。
{"title":"Eliminating geometric distortion with dual-orbit Sentinel-1 SAR fusion for accurate glacial lake extraction in Southeast Tibet Plateau","authors":"Renzhe Wu , Guoxiang Liu , Xin Bao , Jichao Lv , Age Shama , Bo Zhang , Wenfei Mao , Jie Chen , Zhihan Yang , Rui Zhang","doi":"10.1016/j.jag.2024.104329","DOIUrl":"10.1016/j.jag.2024.104329","url":null,"abstract":"<div><div>Glacial lakes (GLs), which serve as natural reservoirs, are also prospective sources of risk, and their risk levels are continuously increasing as a result of global climate warming. Nevertheless, GLs are situated in mountainous and valley regions, which are distinguished by their complex terrain and unpredictable weather conditions. This leads to restricted availability of optical imagery as a consequence of the frequent cloud cover. Synthetic Aperture Radar (SAR), however, encounters issues with geometric distortion. This paper introduces an unsupervised method based on geometric distortion detection (without orbit state information) and historical positioning using dual-orbit SAR imagery to research GL extraction effectively. This method detects low-quality pixels from dual-orbit SAR imagery through geometric distortion. It extracts GLs using a majority voting integration of unsupervised classification algorithms constrained by historical GL center points. The Southeastern Tibetan Plateau (SETP) was chosen as a representative region for the study, and experiments were conducted from July to August 2018 using dual-orbit Sentinel-1 imagery. A total of 600 refined samples were used for comparative verification. The results demonstrate that this method is capable of reliably identifying the active and passive geometric distortions in SAR imagery. The fusion of dual-orbit SAR based on geometric distortion can effectively enhance the classification performance of remote sensing imagery and achieve the acquisition of GL water storage area during the flood season. The geometric distortion rate was reduced from 29.9% to 7.9% after fusion correction, and the accuracy, recall rate, precision, Intersection over Union (IoU), and F1-Score were 0.989, 0.900, 0.908, 0.825, and 0.904, respectively. This serves as a reference for research that investigates the mechanisms of glacier-GL-climate change.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104329"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142874815","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号