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

The intersection of deep learning and photogrammetry unveils a critical need for balancing the power of deep neural networks with interpretability and trustworthiness, especially for safety-critical application like autonomous driving, medical imaging, or machine vision tasks with high demands on reliability. Quantifying the predictive uncertainty is a promising endeavour to open up the use of deep neural networks for such applications. Unfortunately, most current available methods are computationally expensive. In this work, we present a novel approach for efficient and reliable uncertainty estimation for semantic segmentation, which we call Deep Uncertainty Distillation using Ensembles for Segmentation (DUDES). DUDES applies student-teacher distillation with a Deep Ensemble to accurately approximate predictive uncertainties with a single forward pass while maintaining simplicity and adaptability. Experimentally, DUDES accurately captures predictive uncertainties without sacrificing performance on the segmentation task and indicates impressive capabilities of highlighting wrongly classified pixels and out-of-domain samples through high uncertainties on the Cityscapes and Pascal VOC 2012 dataset. With DUDES, we manage to simultaneously simplify and outperform previous work on Deep-Ensemble-based Uncertainty Distillation.

Monitoring of real estate growth is essential with the increasing demand for housing and working space in cities. In this study, a new methodological framework is proposed to map the area under real estate using geospatial techniques. In this framework, the built-up area and open land at successive stages of development are used to map the area under real estate. Three machine learning algorithms were used, namely random forest (RF), support vector machine (SVM), and feedforward neural networks (FFNN), to classify the land use and land cover (LULC) map of Delhi NCR during 1990–2018, which is the basic input for real estate mapping. The results of the study show that optimized RF performed better than SVM and FFNN in LULC classification. The real estate land increased by 279% in Delhi NCR during 1990–2018. The area under real estate increased by 33%, 47%, 29%, 21%, and 22% during 1990–1996, 1996–2003, 2003–2008, 2008–2014, and 2014–2018, respectively. Among the cities surrounding Delhi, Gurgaon, Rohtak, Noida, and Faridabad have witnessed maximum real estate growth. The approach used in this study could be used for real estate mapping in other cities across the world.

The detection and continuous updating of buildings in geodatabases has long been a major research area in geographic information science and is an important theme for national mapping agencies. Advancements in machine learning techniques, particularly state-of-the-art deep learning (DL) models, offer promising solutions for extracting and modeling building rooftops from images. However, tasks such as automatic labelling of learning data and the generalizability of models remain challenging. In this study, we assessed the sensor and geographic area adaptation capabilities of a pretrained DL model implemented in the ArcGIS environment using very-high-resolution (50 cm) SkySat imagery. The model was trained for digitizing building footprints via Mask R‑CNN with a ResNet50 backbone using aerial and satellite images from parts of the USA. Here, we utilized images from three different SkySat satellites with various acquisition dates and off-nadir angles and refined the pretrained model using small numbers of buildings as training data (5–53 buildings) over Ankara. We evaluated the buildings in areas with different characteristics, such as urban transformation, slums, regular, and obtained high accuracies with F‑1 scores of 0.92, 0.94, and 0.96 from SkySat 4, 7, and 17, respectively. The study findings showed that the DL model has high transfer learning capability for Ankara using only a few buildings and that the recent SkySat satellites demonstrate superior image quality.

During flood events near real-time, synthetic aperture radar (SAR) satellite imagery has proven to be an efficient management tool for disaster management authorities. However, one of the challenges is accurate classification and segmentation of flooded water. A common method of SAR-based flood mapping is binary segmentation by thresholding, but this method is limited due to the effects of backscatter, geographical area, and surface characterstics. Recent advancements in deep learning algorithms for image segmentation have demonstrated excellent potential for improving flood detection. In this paper, we present a deep learning approach with a nested UNet architecture based on a backbone of EfficientNet-B7 by leveraging a publicly available Sentinel‑1 dataset provided jointly by NASA and the IEEE GRSS Committee. The performance of the nested UNet model was compared with several other UNet-based convolutional neural network architectures. The models were trained on flood events from Nebraska and North Alabama in the USA, Bangladesh, and Florence, Italy. Finally, the generalization capacity of the trained nested UNet model was compared to the other architectures by testing on Sentinel‑1 data from flood events of varied geographical regions such as Spain, India, and Vietnam. The impact of using different polarization band combinations of input data on the segmentation capabilities of the nested UNet and other models is also evaluated using Shapley scores. The results of these experiments show that the UNet model architectures perform comparably to the UNet++ with EfficientNet-B7 backbone for both the NASA dataset as well as the other test cases. Therefore, it can be inferred that these models can be trained on certain flood events provided in the dataset and used for flood detection in other geographical areas, thus proving the transferability of these models. However, the effect of polarization still varies across different test cases from around the world in terms of performance; the model trained with the combinations of individual bands, VV and VH, and polarization ratios gives the best results.

Abstract

The aim of this study is to conduct a risk analysis of fluvial and pluvial flood disasters, focusing on the vulnerability of those residing in the river basin in coastal regions. However, there are numerous factors and indicators that need to be considered for this type of analysis. Swift and precise acquisition and evaluation of such data is an arduous task, necessitating significant public investment. Remote sensing offers unique data and information flow solutions in areas where access to information is restricted. The Google Earth Engine (GEE), a remote sensing platform, offers strong support to users and researchers in this context. A data-based and informative case study has been conducted to evaluate the disaster risk analysis capacity of the platform. Data on three factors and 17 indicators for assessing disaster risk were determined using coding techniques and web geographic information system (web GIS) applications. The study focused on the Filyos River basin in Turkey. Various satellite images and datasets were utilized to identify indicators, while land use was determined using classification studies employing machine learning algorithms on the GEE platform. Using various applications, we obtained information on ecological vulnerability, fluvial and pluvial flooding analyses, and the value of indicators related to construction and population density. Within the scope of the analysis, it has been determined that the disaster risk index (DRI) value for the basin is 4. This DRI value indicates that an unacceptable risk level exists for the 807,889 individuals residing in the basin.

Since the end of 2022, two ground motion services that cover the complete area of Germany are available as web services: the German Ground Motion Service (Bodenbewegungsdienst Deutschland, BBD) provided by the Federal Institute for Geosciences and Natural Resources (BGR), and the first release of the European Ground Motion Service (EGMS) as part of the Copernicus Land Monitoring Service. Both services are based on InSAR displacement estimations generated from Sentinel‑1 data. It would seem relevant to compare the products of the two services against one another, assess the data coverage they provide, and investigate how well they perform compared to other geodetic techniques. For a study commissioned by the surveying authority of the state of Baden-Württemberg (Landesamt für Geoinformation und Landentwicklung Baden-Württemberg, LGL), BBD and EGMS data from different locations in Baden-Württemberg, Saarland, and North Rhine-Westphalia (NRW) were investigated and validated against levelling and GNSS data. We found that both services provide good data quality. BBD shows slightly better calibration precision than EGMS. The coverage provided by EGMS is better than that of BBD on motorways, federal roads, and train tracks of the Deutsche Bahn. As an example, where both services have difficulties in determining the correct displacements, as they cannot be described well by the displacement models used for processing, we present the test case of the cavern field at Epe (NRW). Finally, we discuss the implications of our findings for the use of the products of BBD and EGMS for monitoring tasks.

Terrestrische spektrale Verfahren für verschiedene thematische Anwendungen erleben mit der Verbreitung der digitalen Aufnahmetechnik in vielen Fachbereichen eine Renaissance oder es erschließen sich neue Anwendungsmöglichkeiten. Bei den daraus resultierenden Veröffentlichungen, zumeist in den Fachzeitschriften der Anwender, entsteht teilweise der Eindruck, dass ein neues Anwendungsfeld erschlossen worden sei. Häufig wurden bereits vor Jahrzehnten diese Themen als relevant eingestuft und mit den zu dieser Zeit aktuellen Sensoren und Methoden bearbeitet. Vermutlich ist eine der Ursachen dieser Auffassung, dass diese alten analogen Literaturquellen noch nicht im Internet verfügbar sind. Mit dem vorliegenden Artikel soll versucht werden, eine Übersicht über verschiedene Anwendungen terrestrischer analoger spektraler fotografischer Aufnahmemethoden in Deutschland zu geben. Thematisch orientieren die Beispiele vor allem auf die Bereiche Geologie, Bergbau, Böden und Vegetation.

Vielleicht gibt der Artikel auch die Anregung, das gesamte Inhaltsverzeichnis unserer Fachzeitschriften und anderer Veröffentlichungen digital aufzubereiten und damit einen Beitrag zur Wissenschaftsgeschichte zu leisten.

Short-term monitoring of alpine glaciers is crucial to understand their response to climate change. This paper presents a low-cost multi-camera system tailored for 4D glacier monitoring using deep learning stereo-photogrammetry. Our approach integrates multi-temporal 3D reconstruction from stereo cameras and surface velocity estimation from a monoscopic camera through digital image correlation. To address the challenges posed by wide camera baselines in complex environments, we have integrated state-of-the-art deep learning feature matching algorithms into ICEpy4D, a Python toolkit designed for 4D monitoring (https://github.com/franioli/icepy4d). In a pilot study conducted on the debris-covered Belvedere Glacier (Italian Alps), our stereoscopic setup, with a camera base–height ratio close to one, captured daily images from May to November 2022. Our approach utilized SuperPoint and SuperGlue for feature matching, resulting in a daily 3D reconstruction of the glacier terminus, as traditional SIFT-like feature matching fails in this scenario. Using dense point clouds with decimetric accuracy, we estimated daily ice volume loss and glacier retreat at the terminus. The total ice volume loss was (63,000,text{m})({}^{3}) and the retreat was (17.8,text{m}). Surface kinematics revealed three times higher surface velocity during the warm season (May–September) than in the fall (September–November). Daily analyses revealed a significant short-term correlation between air temperature, glacier surface velocity and ice ablation, providing insight into the glacier’s response to external forces. The low cost and ease of deployment of the proposed system facilitates replication at other sites for short-term monitoring of glacier dynamics.

Persistent Scatterers (PS) are points selected by Persistent Scatterer for Synthetic Aperture Radar Interferometry (PS-InSAR) technology. PS density and quality determine the accuracy of deformation monitoring results. A comprehension of PS and its influencing factors could provide suggestions for data selection and parameter setting in the time series of InSAR, and it can also provide the decision basis for radar satellite engineers to select imaging modes for different application requirements. PS characteristics are mainly affected by SAR image resolution, wavelength and land cover type, etc. However, these influencing factors are coupled together, so it is difficult to study the relationship between the single factor and PS characteristics. Therefore, this paper adopted the Split-Spectrum to TerraSAR datasets to construct a series of simulated SAR datasets with different resolutions while keeping the other imaging parameters the same. We found that the PS density presents a declining linear trend as the bandwidth (resolution) decreases, while the deformation patterns of PS obtained from different bandwidth datasets are consistent. In addition, we proposed a simplified model to estimate the PS density obtained from 1/k bandwidth datasets. Then, we compared the PS results obtained from X-band TerraSAR datasets and C-band Sentinel-1A datasets and analyzed the reason for the difference from the perspective of spatiotemporal decorrelation. Finally, combined with the land cover map and Bayesian estimation, we obtained the distribution probability of PS on land cover types.