PFG-Journal of Photogrammetry Remote Sensing and Geoinformation Science最新文献

Multiple viewpoint 3D reconstruction has been used in recent years to create accurate complete scenes and objects used for various applications. This is to overcome limitations of single viewpoint 3D digital imaging such as occlusion within the scene during the reconstruction process. In this paper, we propose a weighted point cloud fusion process using both local and global spatial information of the point clouds to fuse them together. The process aims to minimize duplication and remove noise while maintaining a consistent level of details using spatial information from point clouds to compute a weight to fuse them. The algorithm improves the overall accuracy of the fused point cloud while maintaining a similar degree of coverage comparable with state-of-the-art point cloud fusion algorithms.

Long-term time series satellite imagery became highly essential for analyzing earth cycles such as global warming, climate change, and urbanization. Landsat‑7 satellite imagery plays a key role in this domain since it provides open-access data with expansive coverage and consistent temporal resolution for more than two decades. This paper addresses the challenge of stripe errors induced by Scan Line Corrector sensor malfunction in Landsat‑7 ETM+ satellite imagery, resulting in data loss and degradation. To overcome this problem, we propose a Generative Adversarial Networks approach to fill the gaps in the Landsat‑7 ETM+ panchromatic images. First, we introduce the YTU_STRIPE dataset, comprising Landsat‑8 OLI panchromatic images with synthetically induced stripe errors, for model training and testing. Our results indicate sufficient performance of the Pix2Pix GAN for this purpose. We demonstrate the efficiency of our approach through systematic experimentation and evaluation using various accuracy metrics, including Peak Signal-to-Noise Ratio, Structural Similarity Index Measurement, Universal Image Quality Index, Correlation Coefficient, and Root Mean Square Error which were calculated as 38.5570, 0.9206, 0.7670, 0.7753 and 3.8212, respectively. Our findings suggest promising prospects for utilizing synthetic imagery from Landsat‑8 OLI to mitigate stripe errors in Landsat‑7 ETM+ SLC-off imagery, thereby enhancing image reconstruction efforts. The datasets and model weights generated in this study are publicly available for further research and development: https://github.com/ynsemrevrl/eliminating-stripe-errors.

Recent developments in deep learning technology have boosted the performance of dense stereo reconstruction. However, the state-of-the-art deep learning-based stereo matching methods are mainly trained using close-range synthetic images. Consequently, the application of these methods in aerial photogrammetry and remote sensing is currently far from straightforward. In this paper, we propose a new disparity estimation network for stereo matching and investigate its generalization abilities in regard to aerial images. First, we propose an end-to-end deep learning network for stereo matching, regularized by disparity gradients, which includes a residual cost volume and a reconstruction error volume in a refinement module, and multiple losses. In order to investigate the influence of the multiple losses, a comprehensive analysis is presented. Second, based on this network trained with synthetic close-range data, we propose a new pipeline for matching high-resolution aerial imagery. The experimental results show that the proposed network improves the disparity accuracy by up to 40% in terms of errors larger than 1 px compared to results when not including the refinement network, especially in areas containing detailed small objects. In addition, in qualitative and quantitative experiments, we are able to show that our model, pre-trained on a synthetic stereo dataset, achieves very competitive sub-pixel geometric accuracy on aerial images. These results confirm that the domain gap between synthetic close-range and real aerial images can be satisfactorily bridged using the proposed new deep learning method for dense image matching.

Increasing the degree of digitization and automation in concrete production can make a decisive contribution to reducing the associated (text{CO}_{2}) emissions. This paper presents a method which predicts the properties of fresh concrete during the mixing process on the basis of stereoscopic image sequences of the moving concrete and mix design information or a variation of these. A Convolutional Neural Network (CNN) is used for the prediction, which receives the images supported by information about the mix design as input. In addition, the network receives temporal information in the form of the time difference between image acquisition and the point in time for which the concrete properties are to be predicted. During training, the times at which the reference values were captured are used for the latter. With this temporal information, the network implicitly learns the time-dependent behavior of the concrete properties. The network predicts the slump flow diameter, the yield stress and the plastic viscosity. The time-dependent prediction opens up the possibility of forecasting the temporal development of the fresh concrete properties during mixing. This is a significant advantage for the concrete industry, as countermeasures can then be taken in a timely manner, if the properties deviate from the desired ones. In various experiments it is shown that both the stereoscopic observations and the mix design information contain valuable information for the time-dependent prediction of the fresh concrete properties.

Urbanization, a global phenomenon with profound implications for sustainable development, is a focal point of Sustainable Development Goal 11 (SDG 11). Aimed at fostering inclusive, resilient, and sustainable urbanization by 2030, SDG 11 emphasizes the importance of monitoring land use efficiency (LUE) through indicator 11.3.1. In the Philippines, urbanization has surged over recent decades. Despite its importance, research on urbanization and LUE has predominantly focused on the country’s national capital region (Metro Manila), while little to no attention is given to comprehensive investigations across different regions, provinces, cities, and municipalities of the country. Additionally, challenges in acquiring consistent spatial data, especially due to the Philippines’ archipelagic nature, have hindered comprehensive analysis. To address these gaps, this study conducts a thorough examination of urbanization patterns and LUE dynamics in the Philippines from 1975 to 2020, leveraging Global Human Settlement Layers (GHSL) data and secondary indicators associated with SDG 11.3.1. Our study examines spatial patterns and temporal trends in built-up area expansion, population growth, and LUE characteristics at both city and municipal levels. Among the major findings are the substantial growth in built-up areas and population across the country. We also found a shift in urban growth dynamics, with Metro Manila showing limited expansion in recent years while new urban growth emerges in other regions of the country. Our analysis of the spatiotemporal patterns of Land Consumption Rate (LCR) revealed three distinct evolutional phases: a growth phase between 1975–1990, followed by a decline phase between 1990–2005, and a resurgence phase from 2005–2020. Generally declining trends in LCR and Population Growth Rate (PGR) were evident, demonstrating the country’s direction towards efficient built-up land utilization. However, this efficiency coincides with overcrowding issues as revealed by additional indicators such as the Abstract Achieved Population Density in Expansion Areas (AAPDEA) and Marginal Land Consumption per New Inhabitant (MLCNI). We also analyzed the spatial patterns and temporal trends of LUE across the country and found distinct clusters of transitioning urban centers, densely inhabited metropolises, expanding metropolitan regions, and rapidly growing urban hubs. The study’s findings suggest the need for policy interventions that promote compact and sustainable urban development, equitable regional development, and measures to address overcrowding in urban areas. By aligning policies with the observed spatial and temporal trends, decision-makers can work towards achieving SDG 11, fostering inclusive, resilient, and sustainable urbanization in the Philippines.

In this paper we address the task of pixel-wise land cover (LC) classification using satellite image time series (SITS). For that purpose, we use a supervised deep learning model and focus on combining spatial and temporal features. Our method is based on the Swin Transformer and captures global temporal features by using self-attention and local spatial features by convolutions. We extend the architecture to receive multi-temporal input to generate one output label map for every input image. In our experiments we focus on the application of pixel-wise LC classification from Sentinel‑2 SITS over the whole area of Lower Saxony (Germany). The experiments with our new model show that by using convolutions for spatial feature extraction or a temporal weighting module in the skip connections the performance improves and is more stable. The combined usage of both adaptations results in the overall best performance although this improvement is only minimal. Compared to a fully convolutional neural network without any self-attention layers our model improves the results by 2.1% in the mean F1-Score on a corrected test dataset. Additionally, we investigate different types of temporal position encoding, which do not have a significant impact on the performance.

A Digital Twin (DT) is a virtual replica of a physical object or system, created to monitor, analyze, and optimize its behavior and characteristics. A Spatial Digital Twin (SDT) is a specific type of digital twin that emphasizes the geospatial aspects of the physical entity, incorporating precise location and dimensional attributes for a comprehensive understanding of its spatial environment. With the recent advancement in spatial technologies and breakthroughs in other computing technologies such as Artificial Intelligence (AI) and Machine Learning (ML), the SDTs market is expected to rise to 25 billion, covering a wide range of applications. The majority of existing research focuses on DTs and often fails to address the necessary spatial technologies essential for constructing SDTs. The current body of research on SDTs primarily concentrates on analyzing their potential impact and opportunities within various application domains. As building an SDT is a complex process and requires a variety of spatial computing technologies, it is not straightforward for practitioners and researchers of this multi-disciplinary domain to grasp the underlying details of enabling technologies of the SDT. In this paper, we are the first to systematically analyze different spatial technologies relevant to building an SDT in a layered approach (starting from data acquisition to visualization). More specifically, we present the tech stack of SDTs into five distinct layers of technologies: (i) data acquisition and processing; (ii) data integration, cataloging, and metadata management; (iii) data modeling, database management & big data analytics systems; (iv) Geographic Information System (GIS) software, maps, & APIs; and (v) key functional components such as visualizing, querying, mining, simulation, and prediction. Moreover, we discuss how modern technologies such as AI/ML, blockchains, and cloud computing can be effectively utilized in enabling and enhancing SDTs. Finally, we identify a number of research challenges and opportunities in SDTs. This work serves as an important resource for SDT researchers and practitioners as it explicitly distinguishes SDTs from traditional DTs, identifies unique applications, outlines the essential technological components of SDTs, and presents a vision for their future development along with the challenges that lie ahead.

Variations in the tropospheric phase delay pose a primary challenge to achieving precise displacement measurements in Interferometric Synthetic Aperture Radar (InSAR) analysis. This study presents a cluster-based empirical tropospheric phase correction approach to analyze land subsidence rates from large-scale Sentinel‑1 data stacks. Our method identifies the optimum number of clusters in individual interferograms for K‑means clustering, and segments extensive interferograms into areas with consistent tropospheric phase delay behaviors. It then performs tropospheric phase correction based on empirical topography-phase correlation, addressing stratified and broad-scale tropospheric phase delays. Applied to a six-year data stack along a 1000-km track in Iran, we demonstrate that this approach enhances interferogram quality by reducing the standard deviation by 50% and lowering the semivariance of the interferograms to 20 cm² at distances up to 800 km in 97% of the interferograms. Additionally, the corrected time series of deformation shows a 40% reduction in the root mean square of residuals at the most severely deformed points. By analyzing the corrected interferograms, we show that our method improves the efficiency of country-scale InSAR surveys to detect and quantify present-day land subsidence in Iran, which is essential for groundwater management and sustainable water resource planning.

Using images to supplement classical navigation solutions purely based on global navigation satellite systems (GNSSs) has the potential to overcome problems in densely built-up areas. These approaches usually assume a static environment; however, this assumption is not necessarily valid in urban areas. Therefore, many approaches delete information stemming from moving objects in a first processing step, but this results in information being lost. In this paper, we present an approach that detects and models so-called dynamic objects based on image sequences and includes these object models into a bundle adjustment. We distinguish dynamic objects that provide information about their position to others (cooperating objects) and those that do not (non-cooperating objects). Dynamic objects that observe the environment with the help of sensors in order to determine their position are called observing objects. In the experiments discussed here, the observing object is equipped with a stereo camera and a GNSS receiver. We show that cooperating objects can have a positive effect on the exterior orientation of the observing object after the bundle adjustment, both in terms of precision and accuracy. However, we found that introducing non-cooperating objects did not result in further improvements, probably because in our case the photogrammetric block was already stable without them due to the large number and good distribution of static tie points.

Accurate and up-to-date building and road data are crucial for informed spatial planning. In developing regions in particular, major challenges arise due to the limited availability of these data, primarily as a result of the inherent inefficiency of traditional field-based surveys and manual data generation methods. Importantly, this limitation has prompted the exploration of alternative solutions, including the use of remote sensing machine learning-generated (RSML) datasets. Within the field of RSML datasets, a plethora of models have been proposed. However, these methods, evaluated in a research setting, may not translate perfectly to massive real-world applications, attributable to potential inaccuracies in unknown geographic spaces. The scepticism surrounding the usefulness of datasets generated by global models, owing to unguaranteed local accuracy, appears to be particularly concerning. As a consequence, rigorous evaluations of these datasets in local scenarios are essential for gaining insights into their usability. To address this concern, this study investigates the local accuracy of large RSML datasets. For this evaluation, we employed a dataset generated using models pre-trained on a variety of samples drawn from across the world and accessible from public repositories of open benchmark datasets. Subsequently, these models were fine-tuned with a limited set of local samples specific to Rwanda. In addition, the evaluation included Microsoft’s and Google’s global datasets. Using ResNet and Mask R‑CNN, we explored the performance variations of different building detection approaches: bottom-up, end-to-end, and their combination. For road extraction, we explored the approach of training multiple models on subsets representing different road types. Our testing dataset was carefully designed to be diverse, incorporating both easy and challenging scenes. It includes areas purposefully chosen for their high level of clutter, making it difficult to detect structures like buildings. This inclusion of complex scenarios alongside simpler ones allows us to thoroughly assess the robustness of DL-based detection models for handling diverse real-world conditions. In addition, buildings were evaluated using a polygon-wise comparison, while roads were assessed using network length-derived metrics.

Our results showed a precision (P) of around 75% and a recall (R) of around 60% for the locally fine-tuned building model. This performance was achieved in three out of six testing sites and is considered the lowest limit needed for practical utility of RSML datasets, according to the literature. In contrast, comparable results were obtained in only one out of six sites for the Google and Microsoft datasets. Our locally fine-tuned road model achieved moderate success, meeting the minimum usability threshold in four out of six sites. In contrast, the Microsoft dataset performed well on all sites. In summary, our findings suggest improved performance