Letteratura scientifica selezionata sul tema "Sub-Meter resolution imagery"

High spatial resolution coastal Digital Elevation Models (DEMs) are crucial to assess coastal vulnerability and hazards such as beach erosion, sedimentation, or inundation due to storm surges and sea level rise. This paper explores the possibility to use high spatial-resolution Pleiades (pixel size = 0.7 m) stereoscopic satellite imagery to retrieve a DEM on sandy coastline. A 40-km coastal stretch in the Southwest of France was selected as a pilot-site to compare topographic measurements obtained from Pleiades satellite imagery, Real Time Kinematic GPS (RTK-GPS) and airborne Light Detection and Ranging System (LiDAR). The derived 2-m Pleiades DEM shows an overall good agreement with concurrent methods (RTK-GPS and LiDAR; correlation coefficient of 0.9), with a vertical Root Mean Squared Error (RMS error) that ranges from 0.35 to 0.48 m, after absolute coregistration to the LiDAR dataset. The largest errors (RMS error > 0.5 m) occurred in the steep dune faces, particularly at shadowed areas. This work shows that DEMs derived from sub-meter satellite imagery capture local morphological features (e.g., berm or dune shape) on a sandy beach, over a large spatial domain.

Abstract. GaoFen-7 (GF-7) satellite mission is further expanding the very high resolution 3D mapping application. Carrying the first civilian Chinese sub-meter resolution stereo satellite sensors, GF-7 satellite was launched on November 7, 2019. With 0.65 meter resolution on backward view and 0.8 meter resolution forward view, GF-7 has been designed to meet the demand of natural resource monitoring, land surveying, and other mapping applications in China. The use of GF-7 for 3D city reconstruction is unfortunately restricted by the fixed large stereo view angle of forward and backward cameras with +26 and −5 degrees respectively which is not optimal for dense stereo matching in urban regions. In this paper we intensively evaluate the quality of the GF-7 datasets by performing a series of urban monitoring applications, including road detection, building extraction and 3D reconstruction. In addition, we propose a 3D reconstruction workflow which uses the land cover classification result to refine the stereo matching result. Six sub-urban regions are selected from the available datasets in the middle of Germany. The results show that basic elements in urban scenes like buildings and roads could be detected from GF-7 datasets with high accuracy. With the proposed workflow, a 3D city model with a visually observed good quality can be delivered.

Rapid global warming is catalyzing widespread permafrost degradation in the Arctic, leading to destructive land-surface subsidence that destabilizes and deforms the ground. Consequently, human-built infrastructure constructed upon permafrost is currently at major risk of structural failure. Risk assessment frameworks that attempt to study this issue assume that precise information on the location and extent of infrastructure is known. However, complete, high-quality, uniform geospatial datasets of built infrastructure that are readily available for such scientific studies are lacking. While imagery-enabled mapping can fill this knowledge gap, the small size of individual structures and vast geographical extent of the Arctic necessitate large volumes of very high spatial resolution remote sensing imagery. Transforming this ‘big’ imagery data into ‘science-ready’ information demands highly automated image analysis pipelines driven by advanced computer vision algorithms. Despite this, previous fine resolution studies have been limited to manual digitization of features on locally confined scales. Therefore, this exploratory study serves as the first investigation into fully automated analysis of sub-meter spatial resolution satellite imagery for automated detection of Arctic built infrastructure. We tasked the U-Net, a deep learning-based semantic segmentation model, with classifying different infrastructure types (residential, commercial, public, and industrial buildings, as well as roads) from commercial satellite imagery of Utqiagvik and Prudhoe Bay, Alaska. We also conducted a systematic experiment to understand how image augmentation can impact model performance when labeled training data is limited. When optimal augmentation methods were applied, the U-Net achieved an average F1 score of 0.83. Overall, our experimental findings show that the U-Net-based workflow is a promising method for automated Arctic built infrastructure detection that, combined with existing optimized workflows, such as MAPLE, could be expanded to map a multitude of infrastructure types spanning the pan-Arctic.

Efficient and accurate rice identification based on high spatial and temporal resolution remote sensing imagery is essential for achieving precision agriculture and ensuring food security. Semantic segmentation networks in deep learning are an effective solution for crop identification, and they are mainly based on two architectures: the commonly used convolutional neural network (CNN) architecture and the novel Vision Transformer architecture. Research on crop identification from remote sensing imagery using Vision Transformer has only emerged in recent times, mostly in sub-meter resolution or even higher resolution imagery. Sub-meter resolution images are not suitable for large scale crop identification as they are difficult to obtain. Therefore, studying and analyzing the differences between Vision Transformer and CNN in crop identification in the meter resolution images can validate the generalizability of Vision Transformer and provide new ideas for model selection in crop identification research at large scale. This paper compares the performance of two representative CNN networks (U-Net and DeepLab v3) and a novel Vision Transformer network (Swin Transformer) on rice identification in Sentinel-2 of 10 m resolution. The results show that the three networks have different characteristics: (1) Swin Transformer has the highest rice identification accuracy and good farmland boundary segmentation ability. Although Swin Transformer has the largest number of model parameters, the training time is shorter than DeepLab v3, indicating that Swin Transformer has good computational efficiency. (2) DeepLab v3 also has good accuracy in rice identification. However, the boundaries of the rice fields identified by DeepLab v3 tend to shift towards the upper left corner. (3) U-Net takes the shortest time for both training and prediction and is able to segment the farmland boundaries accurately for correctly identified rice fields. However, U-Net’s accuracy of rice identification is lowest, and rice is easily confused with soybean, corn, sweet potato and cotton in the prediction. The results reveal that the Vision Transformer network has great potential for identifying crops at the country or even global scale.

Abstract. The requirement for very high-resolution satellite imagery by different applications has been increasing continuously. Several commercial and government-supported missions provide sub-meter spatial resolutions from optical sensors aboard Earth Observation (EO) satellites. The MAXAR satellite constellation acquires images with up to 30 cm Ground Sampling Distances (GSDs); and the High-Definition (HD) image production technology developed by MAXAR doubles the resolution by using artificial intelligence methods. Although the spatial resolution is one of the most important image quality metrics, several other factors indicated by diverse radiometric and geometric characteristics may circumscribe the usability of data in different projects. As part of mandatory activities of European Space Agency (ESA), Earthnet Programme provides a framework for integrating Third-Party Missions into the overall EO strategy and promotes the international use of the data. The Earthnet Data Assessment Pilot (EDAP) project aims at assessing the quality and the suitability of TPMs, and provides a communication platform between mission providers to ensure the coherence of the systems. In this study, the radiometric quality of the MAXAR HD products was evaluated within the EDAP project framework by using several General Image-Quality Equation (GIQE) metrics, visual inspections, and comparative assessments with orthophotos obtained from an Unmanned Aerial Vehicle (UAV) platform and with the original (non-HD) orthophotos with 30 cm resolutions. The results show that the spatial resolution improvements are observable in urban areas, where sharp edges are present. However, blurring and color noise patterns also occured in the HD images.

Abstract. In recent decades, the geospatial domain has benefitted from technological advances in sensors, methodologies, and processing tools to expand capabilities in mapping applications. Airborne techniques (LiDAR and aerial photogrammetry) generally provide most of the data used for this purpose. However, despite the relevant accuracy of these technologies and the high spatial resolution of airborne data, updates are not sufficiently regular due to significant flight costs and logistics. New possibilities to fill this information gap have emerged with the advent of Very High Resolution (VHR) optical satellite images in the early 2000s. In addition to the high temporal resolution of the cost-effective datasets and their sub-meter geometric resolutions, the synoptic coverage is an unprecedented opportunity for mapping remote areas, multi-temporal analyses, updating datasets and disaster management. For all these reasons, VHR satellite imagery is clearly a relevant study for National Mapping and Cadastral Agencies (NMCAs). This work, supported by EuroSDR, summarises a series of experimental analyses carried out over diverse landscapes to explore the potential of VHR imagery for large-scale mapping.

Abstract. Regional extent and spatiotemporal dynamics of Arctic permafrost disturbances remain poorly quantified. High spatial resolution commercial satellite imagery enables transformational opportunities to observe, map, and document the micro-topographic transitions occurring in Arctic polygonal tundra at multiple spatial and temporal frequencies. The entire Arctic has been imaged at 0.5 m or finer resolution by commercial satellite sensors. The imagery is still largely underutilized, and value-added Arctic science products are rare. Knowledge discovery through artificial intelligence (AI), big imagery, high performance computing (HPC) resources is just starting to be realized in Arctic science. Large-scale deployment of petabyte-scale imagery resources requires sophisticated computational approaches to automated image interpretation coupled with efficient use of HPC resources. In addition to semantic complexities, multitude factors that are inherent to sub-meter resolution satellite imagery, such as file size, dimensions, spectral channels, overlaps, spatial references, and imaging conditions challenge the direct translation of AI-based approaches from computer vision applications. Memory limitations of Graphical Processing Units necessitates the partitioning of an input satellite imagery into manageable sub-arrays, followed by parallel predictions and post-processing to reconstruct the results corresponding to input image dimensions and spatial reference. We have developed a novel high performance image analysis framework –Mapping application for Arctic Permafrost Land Environment (MAPLE) that enables the integration of operational-scale GeoAI capabilities into Arctic science applications. We have designed the MAPLE workflow to become interoperable across HPC architectures while utilizing the optimal use of computing resources.

The emergence of very high-resolution (VHR) satellite imagery (less than 1 m spatial resolution) is creating new opportunities within the fields of ecology and conservation biology. The advancement of sub-meter resolution imagery has provided greater confidence in the detection and identification of features on the ground, broadening the realm of possible research questions. To date, VHR imagery studies have largely focused on terrestrial environments; however, there has been incremental progress in the last two decades for using this technology to detect cetaceans. With advances in computational power and sensor resolution, the feasibility of broad-scale VHR ocean surveys using VHR satellite imagery with automated detection and classification processes has increased. Initial attempts at automated surveys are showing promising results, but further development is necessary to ensure reliability. Here we discuss the future directions in which VHR satellite imagery might be used to address urgent questions in whale conservation. We highlight the current challenges to automated detection and to extending the use of this technology to all oceans and various whale species. To achieve basin-scale marine surveys, currently not feasible with any traditional surveying methods (including boat-based and aerial surveys), future research requires a collaborative effort between biology, computation science, and engineering to overcome the present challenges to this platform’s use.

Abstract. Sub-meter resolution, stereoscopic satellite images allow for the generation of accurate and high-resolution digital elevation models (DEMs) over glaciers and ice caps. Here, repeated stereo images of Drangajökull ice cap (NW Iceland) from Pléiades and WorldView2 (WV2) are combined with in situ estimates of snow density and densification of firn and fresh snow to provide the first estimates of the glacier-wide geodetic winter mass balance obtained from satellite imagery. Statistics in snow- and ice-free areas reveal similar vertical relative accuracy (< 0.5 m) with and without ground control points (GCPs), demonstrating the capability for measuring seasonal snow accumulation. The calculated winter (14 October 2014 to 22 May 2015) mass balance of Drangajökull was 3.33 ± 0.23 m w.e. (meter water equivalent), with ∼ 60 % of the accumulation occurring by February, which is in good agreement with nearby ground observations. On average, the repeated DEMs yield 22 % less elevation change than the length of eight winter snow cores due to (1) the time difference between in situ and satellite observations, (2) firn densification and (3) elevation changes due to ice dynamics. The contributions of these three factors were of similar magnitude. This study demonstrates that seasonal geodetic mass balance can, in many areas, be estimated from sub-meter resolution satellite stereo images.

Le deep learning a transformé de nombreux domaines jusqu'à présent, propulsé par des algorithmes améliorés et une accessibilité accrue aux données. La télédétection, en particulier, offre d'importantes opportunités pour les applications de vision avec des implications socio-écologiques directes. Les forêts représentent un composant environnemental majeur, offrant des fonctions essentielles telles que la régulation du climat, la préservation de la biodiversité et l'interaction avec les êtres vivants. Partant d'études révélant les motifs forestiers à l'aide d'images de résolution grossière ou de capteurs structurels tels que LiDAR, cette thèse explore des images de résolution sub-métrique - des données visuelles humainement interprétables et hautement détaillées couvrant la Terre. Les attributs forestiers de différents types sont étudiés, allant des caractéristiques des arbres à la hauteur et à la biomasse des forêts. Techniquement, cette thèse commence avec une configuration de segmentation sémantique in-domain, plonge dans la régression des attributs structuraux, et se termine par l'adaptation inter-domaine. Cela apporte des éclairages sur la capacité d'apprentissage des modèles de vision dans le contexte de la compréhension d'images de scènes naturelles. La première partie de la thèse introduit un cadre d'apprentissage en profondeur pour compter, localiser et estimer la hauteur des arbres individuels à partir d'images aériennes à l'échelle nationale. Un réseau UNet avec attention est utilisé pour délimiter les couronnes des arbres individuels et compter les arbres avec une supervision ponctuelle. La hauteur des arbres est estimée à partir d'images optiques en apprenant une cartographie des indices visuels aux hauteurs de la canopée projetées à partir de nuages de points LiDAR. Les résultats sont comparés aux données sur le terrain pour évaluer les valeurs pratiques du cadre en tant que données supplémentaires pour soutenir la gestion forestière nationale numérisée. Cette étude met en avant l'efficacité de l'apprentissage profond pour caractériser les structures forestières de manière visuellement interprétable. Cependant, des défis persistent dans l'apprentissage de motifs complexes à partir des données optiques, ce qui motive la seconde étude sur la biomasse forestière au niveau des peuplements, généralement collectée sur le terrain. À une échelle plus large, les méthodes prédominantes appliquent des modèles statistiques ou d'apprentissage automatique sur des images multispectrales, souvent complétées par des données de hauteur et calibrées avec des données de terrain. Nous sommes les premiers à démontrer que la biomasse des peuplements peut être apprise directement à partir d'images RGB de résolution sub-métrique, utilisant des réseaux neuronaux convolutionnels et des données de terrain. Cela ouvre des voies pour une quantification efficace et précise de la biomasse forestière, critique pour la préservation de la nature et les engagements neutres en carbone. Les deux premières études utilisent des systèmes d'apprentissage profond pour quantifier les attributs forestiers avec un jeu de données spécifique, mais une performance dégradée est souvent observée lorsqu'elle est appliquée à des données hors distribution. La troisième étude vise à aborder le problème de décalage de domaine, explorant si les modèles d'apprentissage profond peuvent être rapidement adaptés à de nouveaux jeux de données avec des efforts marginaux. Nous mettons à disposition un nouveau jeu de données composé d'images optiques de résolution sub-métrique collectées dans cinq pays et évaluons l'adaptabilité inter-domaines de diverses tâches de régression au niveau de l'image. Dans l'ensemble, cette thèse présente une collection de systèmes d'apprentissage profond adaptés à la cartographie des ressources forestières, contribuant ainsi aux efforts de gestion durable pour un avenir plus vert
Deep learning has transformed numerous fields so far, propelled by improved algorithms and increased data accessibility. Remote sensing, in particular, offers significant opportunities for vision applications with direct socio-ecological implications. Forests represent a major environmental component, offering essential functions including climate regulation, biodiversity preservation, and interaction with living creatures. Departing from studies that reveal forest patterns using coarse resolution imagery or structural sensors like LiDAR, this thesis explores sub-meter resolution imagery - human-interpretable and highly detailed visual data covering the Earth. Forest attributes of different types are investigated, varying from tree characteristics to forest height and biomass. Technically, this thesis starts with a vanilla setup of in-domain semantic segmentation, delves into the regression of structural attributes, and ends with cross-domain adaptation. This brings insights into the learning capacity of vision models in the context of natural scene image understanding. The first part of the thesis introduces a deep learning framework to count, locate, and estimate the height of individual trees from aerial images at the national scale. An attention UNet is utilized to delineate individual tree crowns and count trees with point supervision. Tree height is estimated from optical imagery by learning a mapping from visual cues to canopy heights projected from LiDAR point clouds. Results are compared against field data to assess the practical values of the framework as supplementary data to support digitized national forest management. This study highlights the capacity of deep learning in characterizing visually interpretable forest and tree structures. Yet, challenges persist in learning more complex patterns from the optical data. This motivates the second study on forest biomass at stand level, a structural measure of forests typically collected on the ground. At larger scales, predominating methods apply statistical or machine learning models on multispectral imagery, often complemented by height data and calibrated with field data. To our knowledge, we are the first to demonstrate that stand-level biomass can be directly learned from sub-meter resolution RGB imagery, rich in forest and tree details, using convolutional neural networks and field data. This provides avenues for efficient and accurate quantification of forest biomass, a critical indicator of forest resources that supports nature preservation and carbon-neutral commitments. The first two studies employ deep learning systems to quantify forest attributes given a specific dataset, which follows the principal in-domain assumption of machine learning. Yet, degraded performance is often observed when applied to out-of-distribution data, a common scenario in practice. The third study aims to address the domain shift issue, exploring whether deep learning models trained on one dataset can be quickly adapted to new datasets with marginal efforts. We release a new dataset consisting of sub-meter resolution optical imagery collected in five countries and assess the cross-domain adaptability of various image-level regression tasks, including tree cover, total tree count, and average canopy height. By enforcing ordered embedding space during training, models are effectively prepared for later adaptation in source-free low-shot setups. Overall, this thesis introduces a collection of deep learning systems tailored for forest resource mapping with depth into technical and applied insights, contributing to sustainable management efforts for a greener future
Deep learning har hidtil transformeret talrige områder, drevet af forbedrede algoritmer og øgettilgængelighed af data. Fjernregistrering tilbyder betydelige muligheder for visionsapplikationermed direkte socioøkologiske implikationer. Skoveudgør en vigtig miljømæssig komponent og tilbyder essentielle funktioner, herunder klimaregulering, bevarelse af biodiversitet og interaktion medlevende væsener. Afgang fra studier, der afslørerskovmønstre ved hjælp af grovopløsningsbilledereller struktursensorer som LiDAR, udforsker denneafhandling sub-meteropløsningsbilleder - menneskefortolkelige og detaljerede visuelle data, derdækker Jorden. Forskellige skovattributter undersøges, lige fra trækarakteristika til skovhøjdeog biomasse. Teknisk set starter denne afhandling med en grundlæggende opsætning af semantisk segmentering inden for domænet, går videretil regression af strukturelle attributter og sluttermed tværfaglig tilpasning. Dette giver indsigt i visionmodellers indlæringskapacitet i konteksten afforståelse af naturscener. Afhandlingens førstedel introducerer et dybtlæringsrammeværk til attælle, lokalisere og estimere højden af individuelle træer fra luftfotos på nationalt plan. En opmærksomhed UNet bruges til at afgrænse individuelle trækroner og tælle træer med punktsupervision. Træhøjde estimeres fra optisk billedmaterialeved at lære en afbildning fra visuelle ledetråde tilkronhøjder projiceret fra LiDAR-punktskyer. Resultater sammenlignes med feltdatabaser for at vurdere rammeværkets praktiske værdier som supplement til understøttelse af digitaliseret nationalskovforvaltning. Denne undersøgelse fremhæverdybtlæringens evne til at karakterisere visuelt fortolkede skov- og træstrukturer. Dog vedvarerudfordringer i at lære mere komplekse mønstrefra det optiske data. Dette motiverer den anden undersøgelse af skovbiomasse på bestandsniveau, en strukturel måling af skove, der typiskindsamles på jorden. På større skalaer anvender dominerende metoder statistiske eller maskinlæringsmodeller på multispektralt billedmateriale,ofte suppleret med højdedata og kalibreret medfeltdatabaser. Efter vores viden er vi de førstetil at demonstrere, at bestandsniveauets biomassedirekte kan læres fra sub-meteropløsnings-RGBbilledmateriale, rigt på skov- og trædetaljer, vedhjælp af konvolutionelle neurale netværk og feltdatabaser. Dette giver muligheder for effektiv og præcis kvantificering af skovbiomasse, enkritisk indikator for skovressourcer, der støtternaturbevarelse og klimaneutrale forpligtelser. Deførste to studier benytter dybtlæringssystemer tilat kvantificere skovattributter ud fra et specifikt datasæt, hvilket følger den grundlæggendeantagelse om maskinlæring inden for domænet.Dog observeres der ofte nedsat præstation, nårdet anvendes på data uden for distributionen,en almindelig situation i praksis. Det tredjestudie sigter mod at adressere domæneskiftproblemet ved at undersøge, om dybtlæringsmodeller trænet på ét datasæt hurtigt kan tilpassestil nye datasæt med marginale bestræbelser.Vi frigiver et nyt datasæt bestående af submeteropløsnings optisk billedmateriale indsamleti fem lande og vurderer krydsdomænet tilpasningsevne for forskellige billedniveau-regressionstasks,herunder trædække, samlet trætælling og gennemsnitlig krones højde. Ved at håndhæve en ordnetindlejret rum under træning forberedes modellereffektivt til senere tilpasning i kildefrie lavskudsopsætninger. Overordnet introducerer denne afhandling en samling dybtlæringssystemer skræddersyet til kortlægning af skovressourcer med dybdei tekniske og anvendte indsigter, hvilket bidrager tilbæredygtige forvaltningsindsatser for en grønnerefremtid