A5032370631- Remote-Sensing Image Classification
- Image Retrieval and Classification Techniques
- Advanced Image and Video Retrieval Techniques
- Synthetic Aperture Radar (SAR) Applications and Techniques
- Advanced SAR Imaging Techniques
- Remote Sensing and Land Use
- Image and Signal Denoising Methods
- Advanced Computational Techniques and Applications
- Geochemistry and Geologic Mapping
- Remote Sensing in Agriculture
- Data Management and Algorithms
- Geographic Information Systems Studies
- Medical Image Segmentation Techniques
- Underwater Acoustics Research
- Time Series Analysis and Forecasting
- Advanced Image Fusion Techniques
- Soil Moisture and Remote Sensing
- Algorithms and Data Compression
- Sparse and Compressive Sensing Techniques
- Arctic and Antarctic ice dynamics
- Anomaly Detection Techniques and Applications
- Remote Sensing and LiDAR Applications
- Neural Networks and Applications
- Microwave Imaging and Scattering Analysis
- Computability, Logic, AI Algorithms
Deutsches Zentrum für Luft- und Raumfahrt e. V. (DLR)
2016-2025
Universitatea Națională de Știință și Tehnologie Politehnica București
2015-2024
École Polytechnique Fédérale de Lausanne
2024
University of Science and Technology
2024
SpaceTech (United States)
2023
Dienstleistungszentrum Ländlicher Raum
2003-2022
Technische Universität Berlin
2022
Centre d'Etudes et De Recherche en Informatique et Communications
2022
Sorbonne Université
2021
Artificial Intelligence in Medicine (Canada)
2021
Object detection is an important and challenging problem in computer vision. Although the past decade has witnessed major advances object natural scenes, such successes have been slow to aerial imagery, not only because of huge variation scale, orientation shape instances on earth's surface, but also due scarcity well-annotated datasets objects scenes. To advance research Earth Vision, known as Observation Remote Sensing, we introduce a large-scale Dataset for deTection Aerial images (DOTA)....
Deep learning methods such as convolutional neural networks (CNNs) can deliver highly accurate classification results when provided with large enough data sets and respective labels. However, using CNNs along limited labeled be problematic, this leads to extensive overfitting. In letter, we propose a novel method by considering pretrained CNN designed for tackling an entirely different problem, namely, the ImageNet challenge, exploit it extract initial set of representations. The derived...
In the past decade, object detection has achieved significant progress in natural images but not aerial images, due to massive variations scale and orientation of objects caused by bird's-eye view images. More importantly, lack large-scale benchmarks become a major obstacle development (ODAI). this paper,we present Dataset Object deTection Aerial (DOTA) comprehensive baselines for ODAI. The proposed DOTA dataset contains 1,793,658 instances 18 categories oriented-bounding-box annotations...
Abstract. This paper describes a deep learning approach to semantic segmentation of very high resolution (aerial) images. Deep neural architectures hold the promise end-to-end from raw images, making heuristic feature design obsolete. Over last decade this idea has seen revival, and in recent years convolutional networks (CNNs) have emerged as method choice for range image interpretation tasks like visual recognition object detection. Still, standard CNNs do not lend themselves per-pixel...
The recognition or understanding of the scenes observed with a synthetic aperture radar (SAR) system requires broader range cues beyond spatial context. These encompass but are not limited to imaging geometry, mode, properties Fourier spectrum images, behavior polarimetric signatures. In this article, we propose change paradigm for explainability in data science case SAR ground explainable artificial intelligence (XAI) SAR. It aims use transformations based on well-established models...
In this paper, we demonstrate the concepts of a prototype knowledge-driven content-based information mining system produced to manage and explore large volumes remote sensing image data. The consists computationally intensive offline part an online interface. aims at extraction primitive features, their compression, data reduction, generation completely unsupervised content-index, ingestion catalogue entry in database management system. Then, user's interests-semantic interpretations...
Basic textures as they appear, especially in high resolution SAR images, are affected by multiplicative speckle noise and should be preserved despeckling algorithms. Sharp edges between different regions strong scatterers also must preserved. To despeckle the authors use a maximum aposteriori (MAP) estimation of cross section, choosing prior models. The proposed approach uses Gauss Markov random field (GMRF) model for textured areas allows an adaptive neighborhood system edge preservation...
This paper describes a deep learning approach to semantic segmentation of very high resolution (aerial) images. Deep neural architectures hold the promise end-to-end from raw images, making heuristic feature design obsolete. Over last decade this idea has seen revival, and in recent years convolutional networks (CNNs) have emerged as method choice for range image interpretation tasks like visual recognition object detection. Still, standard CNNs do not lend themselves per-pixel segmentation,...
The classification of large-scale high-resolution synthetic aperture radar (SAR) land cover images acquired by satellites is a challenging task, facing several difficulties such as semantic annotation with expertise, changing data characteristics due to varying imaging parameters or regional target area differences, and complex scattering mechanisms being different from optical imaging. Given SAR set collected TerraSAR-X hierarchical three-level 150 categories comprising more than 100 000...
Remote sensing (RS) image retrieval is of great significant for geological information mining. Over the past two decades, a large amount research on this task has been carried out, which mainly focuses following three core issues: feature extraction, similarity metric, and relevance feedback. Due to complexity multiformity ground objects in high-resolution remote (HRRS) images, there still room improvement current approaches. In article, we analyze issues RS provide comprehensive review...
Integrating the special electromagnetic characteristics of Synthetic Aperture Radar (SAR) in deep neural networks is essential order to enhance explainability and physics awareness learning. In this paper, we first propose a novel physically explainable convolutional network for SAR image classification, namely guided injected learning (PGIL). It comprises three parts: (1) models (XM) provide prior knowledge, (2) (PGN) encode knowledge into physics-aware features, (3) (PIN) adaptively...
Earth observation (EO) is a prime instrument for monitoring land and ocean processes, studying the dynamics at work, taking pulse of our planet. This article gives bird's eye view essential scientific tools approaches informing supporting transition from raw EO data to usable EO-based information. The promises, as well current challenges these developments, are highlighted under dedicated sections. Specifically, we cover impact (i) Computer vision; (ii) Machine learning; (iii) Advanced...
This paper brings a solution for bridging the gap between results of state-of-the-art automatic classification algorithms and high semantic human-defined manually created terminology cartographic data. Using recent pure-spectral rule-based fully classifier to define basic 'vocabulary', we provide hybrid method automatically understand describe rules that link existent mapping data according different specifications with end-results unsupervised computer information mining methods. Following...
In this paper, we propose to study the dependence of information extraction technique performance on synthetic aperture radar (SAR) imaging parameters and selected primitive features (PFs). The evaluation is done TerraSAR-X data, interpretation realized automatically. first part paper (use case I), following issues are analyzed: 1) finding optimal products their limits variability 2) retrieving number categories/classes that can be extracted from images using PFs (gray-level co-occurrence...
Advances in the image retrieval (IR) field have contributed to elaboration of tools for interactive exploration and extraction images from huge archives associating content with semantic meaning. This paper presents an Earth-observation (EO) IR system based on enriched metadata, annotations, called EO retrieval. generates EO-data model by using automatic feature extraction, processing product defining semantics, which later is fully exploited supporting complex queries. In order demonstrate...
This letter studies how to program and assess a parameterized quantum circuit (PQC) for classifying Earth observation (EO) satellite images. In this exploratory study, we PQC two-label EO image dataset compare it with classic deep learning classifier. We use the an input space of only 17 bits (qubits) due current limitations technology. As real-world EO, selected Eurosat obtained from multispectral Sentinel-2 images as training Berlin, Germany, test image. However, high dimensionality our is...
The authors present a concept of interactive learning and probabilistic retrieval user-specific cover types in content-based remote sensing image archive. A type is incrementally defined via user-provided positive negative examples. From these examples, the infer probabilities Bayesian network that link user interests to pre-extracted content index. Due stochastic nature definitions, database system not only retrieves images according estimated coverage but also accuracy estimation given...
Automatic interpretation of remote-sensing (RS) images and the growing interest for query by image content from large archives rely on ability robustness information extraction observed data. In Parts I II this article, authors turn attention to modern Bayesian way thinking introduce a pragmatic approach extract structural RS selecting library priori models those which best explain structures within an image. Part introduces defines as two-level procedure: 1) model fitting, is incertitude...
For pt.I see ibid., p.1431-45 (1998). The authors present Gibbs-Markov random field (GMRF) models as a powerful and robust descriptor of spatial information in typical remote-sensing image data. This class stochastic provides an intuitive description the data using parameters energy function. selection among several nested fit model, proceed two steps Bayesian inference. procedure yields most plausible model its likely parameters, which together describe content optimal way. Its additional...
This paper deals with the automatic extraction of road network in dense urban areas using a few-meters-resolution synthetic aperture radar (SAR) images. The first part presents proposed method, which is an adaptation previous work to specific case areas. major modifications are 1) clique potentials Markov random field that extracts adapted and 2) multiscale framework used. Results on shuttle mission aerial SAR images different resolutions presented. second dedicated combining two taken...