Search

Quantum machine learning (QML) models promise to have some computational (or quantum) advantage for classifying supervised datasets (e.g., satellite images) over some conventional deep learning (DL) techniques due to their expressive power via their local effective dimension. There are, however, two main challenges regardless of the promised quantum advantage. Currently available quantum bits (qubits) are very small in number, whereas real-world datasets are characterized by hundreds of high-dimensional elements (i.e., features). In addition, there is not a single unified approach for embedding real-world high-dimensional datasets in a limited number of qubits. Some real-world datasets are too small for training intricate QML networks. Hence, to tackle these two challenges for benchmarking and validating QML networks on real-world, small, and high-dimensional datasets in one go, we employ quantum transfer learning comprising a classical VGG16 layer and a multiqubit QML layer. We use real-amplitude and strongly-entangling N-layer QML networks with and without data reuploading layers as a multiqubit QML layer, and evaluate their expressive power quantified by using their local effective dimension and the lower local effective dimension of a QML network, the better its performance on unseen data. As datasets, we utilize Eurosat and synthetic datasets (i.e., easy-to-classify datasets), and an UC Merced Land use dataset (i.e., a hard-to-classify dataset). Our numerical results show that the strongly-entangling N-layer QML network has a lower local effective dimension than the real-amplitude QML network and outperforms it on the hard-to-classify datasets. In addition, quantum transfer learning helps tackle the two challenges mentioned above for benchmarking and validating QML networks on real-world, small, and high-dimensional datasets.

Optical remote sensing instruments accumulate abundant data from across all of the earth's land surfaces, making it possible both to understand the effects of climate change and to monitor, investigate, and manage ground-level events in detail. Processing data using resources located near on-board satellite sensors can bring major benefits in terms of minimizing analysis time and quickly initiating active actions in critical situations. In satellite missions, long-term production on-board algorithms may encounter unexplored samples, i.e., abnormal ground-level events, and need to be able to discriminate and take the correct action. In this matter, the authors present a field programmable gate array (FPGA)-based solution for natural anomaly detection in multispectral imagery using deep convolutional neural networks. The effects of weather-induced hazards and natural disasters, considered anomalies in this sense, are discovered by modeling an anomaly detector on a hybrid system that is hardware efficient. The proposed approach is assembled on a Xilinx Zynq UltraScale+ XCZU9EG multiprocessor system-on-chip (MPSoC) device, where a deep convolutional model is scaled into the FPGA logic, followed by a downstream statistical meta-recognition predictor. The proposed anomaly detection accelerator has produced notable results in identifying a contemporary natural hazard, i.e., burned areas, in scenes acquired by Sentinel-2 over Europe, i.e., Spain and France. The implemented algorithm achieved on the FPGA accelerator an equivalent speedup of 4.46× and 4.5× lower power consumption than the equivalent implementation on the Tesla K80 GPU.

Multispectral (MS) remote sensing images are of great interest for various applications, yet, quite often, an MS product exhibits one or more noisy bands, strip lines, or even missing bands, which leads to decreased confidence in the information it contains. Meeting this challenge, this article proposes a UNet-based neural network architecture to reconstruct a spectral band. The worst case scenario is considered, that of a missing band, the reconstruction being performed based on the available bands. Besides the comparison with state-of-the-art methods, both the qualitative and quantitative analyses are fulfilled considering several metrics: root-mean-square error, structural similarity index, signal-to-reconstruction error, peak-signal-to-noise ratio, and spectral angle mapper. The experiments focus on Sentinel-2 open data within the Copernicus program. Various patterns of urban areas, agricultural regions, and regions from North Pole or Kyiv, Ukraine are included in our dataset to prove the efficiency of band reconstruction regardless of land-cover diversity.

Openly available satellite image time series (SITS) are considered an important resource for spatiotemporal change monitoring. However, obtaining semantically annotated datasets for such tasks is an expensive affair. To alleviate this problem, this article presents a novel framework to model and understand the image dynamics by discovering latent information in Sentinel-1 SITS, even with limited ground truth data. The framework suggests how to use visualizations to efficiently integrate domain knowledge both for execution and evaluation of the machine-learning pipeline in the absence from ground truth data in SITS change studies. In a case study at a Polar region, we extend a limited amount of ground truth data and then discover its temporal evolution at image patch level, in an unsupervised manner. The trustworthiness of the framework is ensured by integration of domain knowledge and intelligent visual verification strategies. A visualization tool is also implemented for this purpose. The proposed framework contains two modules: a classifier and a change modeler. Our experiments show that a domain-knowledge-based classifier gives the best accuracy. The classifier semantically labeled the complete dataset of 24 study months, containing 153 600 patches with a size of 256 × 256 pixels by extending the available semantic labels from just three months. The temporal sequence of these sematic labels are then recorded and fed to a Bayesian model called Latent Dirichlet Allocation (LDA) to discover the underlying patterns. LDA generates a change map containing the dominant dynamic patterns to give a consolidated view of the evolution without having to browse the whole dataset. Further, color-coded change signatures explain the change classes.

As deep learning attracts earth observation (EO) community's interest, the challenge to derive explainable, actionable information creates a bottleneck in EO models' development. Computer vision proved that effective results of the DL algorithms imply a considerable amount of training datasets. This is not the case in EO, where images are characterized by a broad variety of sensor data, ranging from multispectral to Synthetic Aperture Radar (SAR) with variable number of spectral bands, polarization, or spatial resolution. In this article, we present effective methodologies for fast training with reduced datasets of simple deep neural networks, while preserving the similar performance of state-of-the-art methods. The hybrid solutions we provide imply the reduction of input data dimension in a convolutional neural network. We replaced dataset's patches with histograms of pixel intensity, Bag-of-Words, or downsampled images. Following the proposed approaches, the training time and the dataset size are significantly reduced, while the performance of classification is preserved. These optimized implementations enable the deployment of lightweight deep learning models for real-time processing tasks able to exhibit accurate results, for instance, a disaster management scenario. We demonstrated the computational efficiency of these approaches on various, complex data, both multispectral and SAR, with different resolutions.

This article studies remote sensing image retrieval using kernel-based support vector data description (SVDD). We exploit deep SVDD, which is a well-known method for one-class classification to recover the most relevant samples from the archive. To this end, a deep neural network (DNN) is jointly trained to map the data into a hypersphere of minimum volume in the latent space. It is expected that similar samples to the query are compressed inside of the hypersphere. The closest embedding to the center of the hypersphere is related to the most relevant sample to query. We enhance deep SVDD by injecting the statistical information of data to the DNN by means of additional terms in the cost function. The first enhancement method takes advantage of covariance regularization of batches of the training set to penalize unnecessary redundancy and minimize the correlation between the different dimensions of the embedding. The second method involves unlocking the hypersphere's predefined center while preventing network divergence during training. Therefore, two parameters are designed to control the importance of the drifting of the center and the importance of a fixed predefined center (convergence), respectively. This has been implemented by considering the average of batches of embedding in each iteration as the updated center. This pushes irrelevant samples away from query samples, making data clustering easier for the DNN. The performance of the proposed methods is evaluated on benchmark datasets.

The pavement distress measurement is a crucial aspect in guaranteeing the safety of transportation infrastructure. In this regard, we introduce a novel and cost-effective multi-sensor approach for pavement segmentation during low-light night conditions. By utilizing the low-cost Azure Kinect multi-sensor system, we generated a multi-sensor dataset that encompasses aligned IR, RGB, and depth images. Then we carried out the data annotation process on the RGB images. A total of 11,343 manual annotations were meticulously made on 791 images, which were randomly selected from a collection of 96,891 frames. Subsequently, four different deep learning-based image segmentation models were analyzed both quantitatively and qualitatively. The results indicated that the segmentation performance on the IR dataset outperformed that of the RGB dataset. The model with the highest mIoU (mean Intersection over Union) of 0.7169 was ConvNext when trained on the IR dataset. Furthermore, we proposed the use of relative height for evaluating the severity of pavement distress. On the aligned depth map, the relative height was calculated using the depth data from the corresponding pavement distress area. Additionally, a quantitative comparison between manual annotations and the results obtained through deep learning revealed that the latter was more effective in identifying more severe forms of pavement distress. Through this study, we established the feasibility of collecting pavement distress data during nighttime using a low-cost multi-sensor system.

In this paper, we proposed to investigate unsupervised anomaly detection in Synthetic Aperture Radar (SAR) images. Our approach considers anomalies as abnormal patterns that deviate from their surroundings without prior knowledge of their characteristics. This method deals with the crucial problems related to the presence of speckle, the spatial correlation structures in SAR images, and the lack of annotated data to train a detection algorithm. Our proposed method aims to address these issues through a self-supervised learning algorithm. First, we propose to mitigate the SAR speckle through the deep learning SAR2SAR algorithm. We then develop an Adversarial Autoencoder (AAE) to reconstruct anomaly-free SAR images from despeckled data taking into account potential spatial correlation structures. Finally, a change detection processing step is applied between the input and the output to detect anomalies. Experiments are performed to show the advantages of our method compared to the conventional Reed-Xiaoli algorithm, highlighting the importance of an efficient despeckling pre-processing step.

This article presents a methodology based on matched-adaptive filters, which is used to mitigate the effect of near-range multipath in multistatic SAR systems and transponders. It is challenging to physically construct a bistatic receiver such that the reference signal is not leaked into the received signal, either via coupling in the circuitry or via reflections off objects in the vicinity of the receiver. Due to its much larger amplitude, the reference signal can easily mask near-range targets with its sidelobes. A similar signal degradation is observed in active transponders that are used for calibrating radar systems, when objects exist in their vicinity. In this article, we address these two issues: the coupling between the reference channel and the imaging channel, and the parasitic echoes present in the transponder response. A novel methodology is proposed that is capable of time-domain filtering the undesired components in real time. The novelty consists in combining matched and adaptive filters as a means of boosting performance and resolution estimation, resulting in an extremely accurate multipath elimination method. The proposed methodology is experimentally evaluated and optimized for each of the two aforementioned problems.

Recent advances in remote sensing technology have provided (very) high spatial resolution Earth Observation data with abundant latent semantic information. Conventional data processing algorithms are not capable of extracting the latent semantic information form these data and harness their full potential. As a result, semantic information discovery methods, based on data mining techniques, such as latent Dirichlet allocation and bag of visual words models, can discover the latent information. Despite their crucial rule, there are only a few studies in the field of semantic data mining for remote sensing applications. This article is focused on this shortage. Three different scenarios are used to evaluate the semantic information discovery in various remote sensing applications, including both optical and synthetic aperture radar (SAR) data with different spatial resolutions. In the first scenario, semantic discovery method correlated the semantic perception of the user and machine to correct and enhance the user defined Ground Truth map in very high-resolution RGB data. The potential of the semantic discovery is evaluated for wildfire affected area detection in Sentinel-2 data in the second scenario. Finally, in the third scenario, the semantic discovery method is utilized to detect the misclassifications as well as the patches with ambiguous or multiple semantic labels in a Sentinel-1 SAR patch-based benchmark dataset to enhance the robustness and accuracy of the annotation in the dataset. Our results in these three scenarios demonstrated the capability of the data-mining-based semantic information discovery methods for various remote sensing.

In this article, we propose a promising approach for the application-oriented content classification of spaceborne radar imagery that presents an interesting alternative to popular current machine learning algorithms. In the following, we consider the problem of unsupervised feature-free satellite image classification with already known classes as an explainable data mining problem for regions with no prior information. Three important issues are addressed here: explainability, feature independence, and unsupervision. There is an increasing demand toward explainable machine learning models as they strive to meet the “right to explanation.” The importance of feature-free classification stems from the problem that different classification outcomes are obtained from using different features and the complexity of computing sophisticated image primitive features. Developing unsupervised discovery techniques helps overcome the limitations in object discovery due to the lack of labeled data and the dependence on features. In this article, we demonstrate the applicability of a latent Dirichlet allocation (LDA) model, one of the most established unsupervised probabilistic methods, in discovering the latent structure of synthetic aperture radar data. The idea is to use LDA as an explainable data mining tool to discover scientifically explainable semantic relations. The suitability of the approach as an explainable model is discussed and interpretable topic representation maps are produced, which practically demonstrate the idea of “interpretability” in an explainable machine learning paradigm. LDA discovers the latent structures in the data as a set of topics. We create the interpretable visualizations of the data utilizing these topics and compute the topic distributions for each land-cover class. Our results show that each class has a distinct topic distribution that represents that particular class. Then these classes can be grouped based on their similarity of topic composition. Both the topic composition and grouping are explainable by domain experts.

This article presents a new scheme called 2-D mixed compressive sensing back-projection (CS-BP-2D), for synthetic aperture radar (SAR) imaging on a geocoded grid, in a single measurement vector frame. The back-projection linear operator is derived in matrix form and a patched-based approach is proposed for reducing the dimensions of the dictionary. Spatial compressibility of the radar image is exploited by constructing the sparsity basis using the back-projection focusing framework and fast solving the reconstruction problem through the orthogonal matching pursuit algorithm. An artifact reduction filter inspired by the synthetic point spread function is used in postprocessing. The results are validated for simulated and real-world SAR data. Sentinel-1 C-band raw data in both monostatic and space-borne transmitter/stationary receiver bistatic configurations are tested. We show that CS-BP-2D can focus both monostatic and bistatic SAR images, using fewer measurements than the classical approach, while preserving the amplitude, the phase, and the position of the targets. Furthermore, the SAR image quality is enhanced and also the storage burden is reduced by storing only the recovered complex-valued points and their corresponding locations.

While the analysis and understanding of multispectral (i.e., optical) remote sensing images have made considerable progress during the last decades, the automated analysis of synthetic aperture radar (SAR) satellite images still needs some innovative techniques to support nonexpert users in the handling and interpretation of these big and complex data. In this article, we present a survey of existing multispectral and SAR land cover image datasets. To this end, we demonstrate how an advanced SAR image analysis system can be designed, implemented, and verified that is capable of generating semantically annotated classification results (e.g., maps) as well as local and regional statistical analytics such as graphical charts. The initial classification is made based on Gabor features and followed by class assignments (labeling). This is followed by inclusion. This can be accomplished by the inclusion of expert knowledge via active learning with selected examples, and the extraction of additional knowledge from public databases to refine the classification results. Then, based on the generated semantics, we can create new topic models, find typical country-specific phenomena and distributions, visualize them interactively, and present significant examples including confusion matrices. This semiautomated and flexible methodology allows several annotation strategies, the inclusion of dedicated analytics procedures, and can generate broad as well as detailed semantic (multi-)labels for all continents, and statistics or models for selected countries and cities. Here, we employ knowledge graphs and exploit ontologies. These components could already be validated successfully. The proposed methodology can also be adapted to other SAR instruments with different resolutions as well as to multispectral images.

Hyperspectral images (HSIs) showing objects belonging to several distinct target classes are characterized by dozens of spectral bands being available. However, some of these spectral bands are redundant and/or noisy, and hence, selecting highly informative and trustworthy bands for each class is a vital step for classification and for saving internal storage space; then the selected bands are termed a highly informative spectral band subset. We use a mutual information (MI)-based method to select the spectral band subset of a given class and two additional binary quantum classifiers, namely a quantum boost (Qboost) and a quantum boost plus (Qboost-Plus) classifier, to classify a two-label dataset characterized by the selected band subset. We pose both our MI-based band subset selection problem and the binary quantum classifiers as a quadratic unconstrained binary optimization (QUBO) problem. Such a quadratic problem is solvable with the help of conventional optimization techniques. However, the QUBO problem is an NP-hard global optimization problem, and hence, it is worthwhile for applying a quantum annealer. Thus, we adapted our MI-based optimization problem for selecting highly informative bands for each class of a given HSI to be run on a D-Wave quantum annealer. After the selection of these highly informative bands for each class, we employ our binary quantum classifiers to a two-label dataset on the D-Wave quantum annealer. In addition, we provide a novel multilabel classifier exploiting an error-encoding output code when using our binary quantum classifiers. As a real-world dataset in Earth observation, we used the well-known AVIRIS HSI of Indian Pine, north-western Indiana, USA. We can demonstrate that the MI-based band subset selection problem can be run on a D-Wave quantum annealer that selects the highly informative spectral band subset for each target class in the Indian Pine HSI. We can also prove that our binary quantum classifiers and our novel multilabel classifier generate a correct two- and multilabel dataset characterized by their selected bands and with high accuracy such as having been produced by conventional classifiers—and even better in some instances.

In synthetic aperture radar (SAR) image classification applications, there are three categories of data, including training and benchmark data with fixed classes, as well as actual data in practical applications. A real problem comes that there exist unknown classes not included in training and benchmark data, which is defined as the open set condition. However, little work on recognizing unknown classes and analyzing the separability of SAR datasets has been developed. Motivated by this observation, this article demonstrates the difficulty of practical classification and analyzes SAR dataset separability in open set conditions. In this article, the SAR separability analyzer (SAR-SA) is proposed to model each known class as a multivariate Gaussian distribution. SAR-SA can classify the known classes and recognize the samples locating in each known distribution with low probabilities as unknown. Besides, SAR datasetwise separability index (DSI) and classwise separability index (CSI) are defined to quantify the separability in open set conditions at the dataset level and class level. DSI and CSI are effective indicators of the difficulty of SAR classification datasets. Extensive experimental results demonstrate that the DSI in open set conditions is nearly half of that in supervised conditions. Dataset with low DSI is hard to realize accurate classification in open set conditions. At the class level, even though the SAR image classes are semantically different from each other, there exists more or less overlap between the distributions of supervised known classes and unknown classes. Classes with low CSI are harder to be correctly classified and recognized.

Conventional remote sensing data analysistechniques have a significant bottleneck of operating on a selectively chosen small-scale dataset. Availability of an enormous volume of data demands handling large-scale, diverse data, which have been made possible with neural network-based architectures. This article exploits the contextual information capturing ability of deep neural networks, particularly investigating multispectral band properties from Sentinel-2 image patches. Besides, an increase in the spatial resolution often leads to nonlinear mixing of land-cover types within a target resolution cell. We recognize this fact and group the bands according to their spatial resolutions, and propose a classification and retrieval framework. We design a representation learning framework for classifying the multispectral data by first utilizing all the bands and then using the grouped bands according to their spatial resolutions. We also propose a novel triplet-loss function for multilabeled images and use it to design an interband group retrieval framework. We demonstrate its effectiveness over the conventional triplet-loss function. Finally, we present a comprehensive discussion of the obtained results. We thoroughly analyze the performance of the band groups on various land-cover and land-use areas from agro-forestry regions, water bodies, and human-made structures. Experimental results for the classification and retrieval framework on the benchmarked BigEarthNet dataset exhibit marked improvements over existing studies.

Visualization of multispectral images through band selection methods determines an information loss that in utmost cases proves to be critical for the adequate understanding of the represented scene. The R–G–B representation obtained by mapping the visual bands to the R, G, and B channels is highly used due to its great resemblance with the natural color one and aspects perceivable by the human eye. However, despite the similarity in terms of color code, ambiguities between classes such as water and vegetation or atmospheric phenomena like fog, clouds, and smoke that have been penetrated by other bands, remain visible and hinder the process of visualization of the Earth surface. This article presents a set of five different methods to offset the effects caused by ambiguities, fog, light clouds, and smoke by transferring relevant information between bands in order to visually reconstitute those parts of the image affected by atmospheric phenomena. The general concept shared by these methods implies a stacked autoencoder that successfully encompasses the information from all spectral bands into a latent representation used for visualization. Each proposed method is defined by different combination of input and error function formula. Spectral and polar coordinates features represent the possible options for the input, while formulas based on mean squared error or angular spectral distances determine the potential choices in terms of error function definition. The property of angular spectral distance and polar coordinates transformation to obtain illuminant invariant features determined their use in three out of five methods. We evaluate the methods through spectral signature graphical comparison and visual comparison related to the R–G–B representation. We conduct experiments on multiple Sentinel 2 full images.

The increased number of free and open Sentinel satellite images has led to new applications of these data. Among them is the systematic classification of land cover/use types based on patterns of settlements or agriculture recorded by these images, in particular, the identification and quantification of their temporal changes. In this paper, we will present guidelines and practical examples of how to obtain rapid and reliable image patch labelling results and their validation based on data mining techniques for detecting these temporal changes, and presenting these as classification maps and/or statistical analytics. This represents a new systematic validation approach for semantic image content verification. We will focus on a number of different scenarios proposed by the user community using Sentinel data. From a large number of potential use cases, we selected three main cases, namely forest monitoring, flood monitoring, and macro-economics/urban monitoring.

Throughout the years, various Earth Observation (EO) satellites have generated huge amounts of data. The extraction of latent information in the data repositories is not a trivial task. New methodologies and tools, being capable of handling the size, complexity and variety of data, are required. Data scientists require support for the data manipulation, labeling and information extraction processes. This paper presents our Earth Observation Image Librarian (EOLib), a modular software framework which offers innovative image data mining capabilities for TerraSAR-X and EO image data, in general. The main goal of EOLib is to reduce the time needed to bring information to end-users from Payload Ground Segments (PGS). EOLib is composed of several modules which offer functionalities such as data ingestion, feature extraction from SAR (Synthetic Aperture Radar) data, meta-data extraction, semantic definition of the image content through machine learning and data mining methods, advanced querying of the image archives based on content, meta-data and semantic categories, as well as 3-D visualization of the processed images. EOLib is operated by DLR’s (German Aerospace Center’s) Multi-Mission Payload Ground Segment of its Remote Sensing Data Center at Oberpfaffenhofen, Germany.

The Sentinel-1 mission provides a freely accessible opportunity for urban image interpretation based on synthetic aperture radar (SAR) data with a specific resolution, which is of paramount importance for Earth observation. In parallel, with the rapid development of advanced technologies, especially deep learning, we urgently need a large-scale SAR dataset supporting urban image interpretation. This article presents OpenSARUrban: a Sentinel-1 dataset dedicated to the content-related interpretation of urban SAR images, including a well-defined hierarchical annotation scheme, data collection, well-established procedures for dataset compilation and organization as well as properties, visualizations, and applications of this dataset. Particularly, our OpenSARUrban collection provides 33 358 image patches of urban SAR scenes, covering 21 major cities of China, including 10 different target area categories, 4 kinds of data formats, 2 kinds of polarization modes, and owning 5 essential properties: large-scale coverage, diversity, specificity, reliability, and sustainability. These properties guarantee the achievement of several goals for OpenSARUrban. The first one is to support urban target characterization. The second one is to help develop well-applicable and advanced algorithms for Sentinel-1 urban target classification. The third one is to explore content-based image retrieval for these kinds of data. In addition, dataset visualization is implemented from the perspective of manifolds to give an intuitive understanding. Besides a detailed description and visualization of the dataset, we present results of some benchmarking algorithms, demonstrating that this dataset is practical and challenging. Notably, developing algorithms to enhance the classification performance on the whole dataset and considering the data imbalance are especially demanding.

The article proposes a methodology to perform azimuth focusing of spaceborne transmitter-stationary receiver bistatic synthetic aperture radar data across multiple along-track apertures to increase azimuth resolution. The procedure uses as input several azimuth apertures (continuous groups of range compressed pulses) from one or more satellite bursts and comprises the following stages: antenna pattern compensation, slow time resampling, reconstruction of missing azimuth samples between neighboring sets of pulses using an autoregressive (AR) model and back-projection focusing of the resulting multiaperture range image. A novel, highly efficient method is proposed to estimate the optimal order for the AR model. It differs from the traditional approach that uses the Akaike Information Criterion to directly estimate the order because the proposed method estimates the order indirectly by detecting the number of targets using principle component analysis. Spatial smoothing is used to obtain a full rank covariance matrix, whose eigenvalues are then analyzed using minimum description length. The optimal order is an integer multiple of the number of targets, which depends on signal-to-noise ratio. The approach is evaluated with real bistatic data acquired over an area of Bucharest city, Romania.

Sentinel-2 satellites provide systematic global coverage of land surfaces, measuring physical properties within 13 spectral intervals at a temporal resolution of five days. Computer-based data analysis is highly required to extract similarity by processing and to assist human understanding and semantic annotation in support of mapping Earth's surface. This article proposes a data mining concept that uses advanced data visualization and explainable features to enhance relevant aspects in the Sentinel-2 data and enable semantic analysis. There is a two-stage process. At first, spectral, texture, and physical parameters related features are extracted from the data and included in a learning process that models the data content according to statistical similarities. In parallel, the second processing stage maximizes the data impact on the human visual system to help image understanding and interpretation. Target classes are subject to exploratory visual analysis, such that both visual and latent characteristics are revealed to the user. The concept is further implemented as Sentinel-2 dedicated data analysis (DAS-Tool) plugin for the Sentinel Application Platform and deployed as an open-source tool empowering the Earth observation community with fast and reliable results. Accommodating multiple solutions for each processing phase, the plugin enables flexibility in information extraction and knowledge discovery that will bring the best accuracy in mapping applications. For demonstration purposes, the authors focus on a detailed benchmark against reference data (ground truth) for the Southern region of Romania, then use the selected algorithms in a forest fires scenario analysis for the Sydney region in Australia. The processing involves full-size Sentinel-2 images.

This paper presents a novel framework for multilabel classification of multispectral remote sensing images using error-correcting output codes. Starting with a set of primary class labels, the proposed framework consists in transforming the multiclass problem into multiple binary learning subtasks. The distributed output representations of these binary learners are then transformed into primary class labels. In order to train robust binary classifiers on a reduced annotated dataset, the learning process is iterative and involves determining most ambiguous examples, which are included in the training set at each iteration. As part of the semantic image recognition process, two categories of high-level image representations are proposed for the feature extraction part. First, deep convolutional neural networks are used to form high-level representations of the images. Second, we test our classification framework with a bag-of-visual words model based on the scale invariant feature transform, used in combination with color descriptors. In the first case, we propose the usage of pretrained state-of-the-art deep learning models that cancel the need to estimate model parameters of complex architectures, whereas, in the second case, a dictionary of visual words must be determined from the training set. Experiments are conducted on GeoEye-1 and Sentinel-2 images and the results show the effectiveness of the proposed approach toward a multilabel classification, when compared to other methods.

Land cover changes may have very different nature, e.g., vegetation development, soil erosion, variation of humidity, or damage of buildings, only to enumerate few cases. In addition, synthetic aperture radar (SAR) observations are a doppelganger of the scene, imaging the scene signature rather than the scene itself. To overcome these challenges, SAR change detection methods generally adapt to the particular situations. We present seamless methods based on normalized compression distance (NCD) estimation. NCD is a similarity metric applied directly to the data, thus with no biases induced by feature estimators or classifiers. Since the diversity of changes is huge and extremely hard to derive typical classes, we introduce paradigm based both on an unsupervised and a supervised method. The change detection procedure mainly consists in dividing image dataset in patches, computing a collection of similarities for pairs of tiles formed differently in each case, and usage of this collection in unsupervised and supervised forms to generate a change map. Both the threshold based histogram, unsupervised method, and the k-NN classifier algorithm, supervised method, have a distinct flow to obtain the change map. To use the NCD operator according to our proposed methods, a speckle resistance test is involved. The experimental results for the two methodologies are computed using two TerraSAR-X images over Sendai and surrounding areas, Japan.

Earth observation (EO) has become a valuable source of comprehensive, reliable, and persistent information for a wide number of applications. However, dealing with the complexity of land cover is sometimes difficult, as the variety of EO sensors reflects in the multitude of details recorded in several types of image data. Their properties dictate the category and nature of the perceptible land structures. The data heterogeneity hampers proper understanding, preventing the definition of universal procedures for content exploitation. The main shortcomings are due to the different human and sensor perception on objects, as well as to the lack of coincidence between visual elements and similarities obtained by computation. In order to bridge these sensory and semantic gaps, the paper presents a compound framework for EO image information extraction. The proposed approach acts like a common ground between the user's understanding, who is visually shortsighted to the visible domain, and the machines numerical interpretation of a much wider information. A hierarchical data representation is considered. At first, basic elements are automatically computed. Then, users can enforce their judgement on the data processing results until semantic structures are revealed. This procedure completes a user-machine knowledge transfer. The interaction is formalized as a dialogue, where communication is determined by a set of parameters guiding the computational process at each level of representation. The purpose is to maintain the data-driven observable connected to the level of semantics and to human awareness. The proposed concept offers flexibility and interoperability to users, allowing them to generate those results that best fit their application scenario. The experiments performed on different satellite images demonstrate the ability to increase the performances in case of semantic annotation by adjusting a set of parameters to the particularities of the analyzed data.

Data visualization guides the process of indexing and retrieval, strengthening the link between low-level image features and high-level human understanding of image content. In this regard, we have described the semantic content of a multidimensional dataset using its descriptors to derive high-dimensional feature spaces. The dimensionality of these spaces is further reduced to three in order to provide a three-dimensional (3-D) representation of the dataset items. Our main challenge was to identify the transformation that projects the high-dimensional feature set into a 3-D space preserving its semantic content. To overcome this issue, we have compared the efficiency of 11 feature space transformations: one feature selection algorithm and ten dimensionality reduction methods. As long as the dataset properties, during mapping, may differ depending on the chosen algorithm, the performance comparison of multiple algorithms is a difficult task. Therefore, three quantitative measures have been used: “Trustworthiness,” “Continuity,” and “QNX”-the number of points preserved in data neighborhoods over projection. The mapping algorithms have been applied to three remote sensing datasets achieved from different sensors: LANDSAT 7 ETM+ and WorldView-3.

This paper aims to present the basic functionality of a radar platform for real time monitoring of displacement and vibration. The real time capabilities make the radar platform useful when live monitoring of targets is required. The system is based on the RF analog front-end of a USRP, and the range compression (time-domain cross-correlation) is implemented on the FPGA included in the USRP. Further processing is performed on the host computer to plot real time range profiles, displacements, vibration frequencies spectra and spectrograms (waterfall plots) for long term monitoring. The system is currently in experimental form and the present paper aims to prove its functionality. The precision of this system is estimated (using the 3σ approximation) at 0.6 mm for displacement measurements and 1.8 mm for vibration amplitude measurements.

Recent years have witnessed an increased interest towards compression-based methods and their applications to remote sensing, as these have a data-driven and parameter-free approach and can be thus succesfully employed in several applications, especially in image information mining. This paper expands the algorithmic information theory frame, on which these methods are based. On the one hand, algorithms originally defined in the pattern matching domain are reformulated, allowing a better understanding of the available compression-based tools for remote sensing applications. On the other hand, the use of existing compression algorithms is proposed to store satellite images with added semantic value.

Information content and compression are tightly related concepts that can be addressed through both classical and algorithmic information theories, on the basis of Shannon entropy and Kolmogorov complexity, respectively. The definition of several entities in Kolmogorov’s framework relies upon ideas from classical information theory, and these two approaches share many common traits. In this work, we expand the relations between these two frameworks by introducing algorithmic cross-complexity and relative complexity, counterparts of the cross-entropy and relative entropy (or Kullback-Leibler divergence) found in Shannon’s framework. We define the cross-complexity of an object x with respect to another object y as the amount of computational resources needed to specify x in terms of y, and the complexity of x related to y as the compression power which is lost when adopting such a description for x, compared to the shortest representation of x. Properties of analogous quantities in classical information theory hold for these new concepts. As these notions are incomputable, a suitable approximation based upon data compression is derived to enable the application to real data, yielding a divergence measure applicable to any pair of strings. Example applications are outlined, involving authorship attribution and satellite image classification, as well as a comparison to similar established techniques.

Synthetic Aperture Radar (SAR) Tomography (TomoSAR) allows extending the 2-D focusing capabilities of SAR to the elevation direction, orthogonal to the azimuth and range. The multi-dimensional extension (along the time) also enables the monitoring of possible scatterer displacements. A key aspect of TomoSAR is the identification, in the presence of noise, of multiple persistent scatterers interfering within the same 2-D (azimuth range plane) pixel. To this aim, the use of multi-look has been shown to provide tangible improvements in the detection of single and double interfering persistent scatterers at the expense of a minor spatial resolution loss. Depending on the system acquisition characteristics, this operation may require also the detection of multiple scatterers interfering at distances lower than the Rayleigh resolution (super-resolution). In this work we further investigated the use of multi-look in TomoSAR for the detection of multiple scatterers located also below the Rayleigh resolution. A solution relying on the Capon filtering was first analyzed, due to its improved capabilities in the separation of the responses of multiple scatterers and sidelobe suppression. Moreover, in the framework of the Generalized Likelihood Ratio Test (GLRT), the single-look support based detection strategy recently proposed in the literature was extended to the multi-look case. Experimental results of tests carried out on two datasets acquired by TerraSAR-X and COSMO-SkyMED sensors are provided to show the performances of the proposed solution as well as the effects of the baseline span of the dataset for the detection capabilities of interfering scatterers.

With more and more SAR applications, the demand for enhanced high-quality SAR images has increased considerably. However, high-quality SAR images entail high costs, due to the limitations of current SAR devices and their image processing resources. To improve the quality of SAR images and to reduce the costs of their generation, we propose a Dialectical Generative Adversarial Network (Dialectical GAN) to generate high-quality SAR images. This method is based on the analysis of hierarchical SAR information and the “dialectical” structure of GAN frameworks. As a demonstration, a typical example will be shown, where a low-resolution SAR image (e.g., a Sentinel-1 image) with large ground coverage is translated into a high-resolution SAR image (e.g., a TerraSAR-X image). A new algorithm is proposed based on a network framework by combining conditional WGAN-GP (Wasserstein Generative Adversarial Network—Gradient Penalty) loss functions and Spatial Gram matrices under the rule of dialectics. Experimental results show that the SAR image translation works very well when we compare the results of our proposed method with the selected traditional methods.

Due to the constant increase in Earth Observation (EO) data collections, the monitoring of land cover is facilitated by the temporal diversity of the satellite images datasets. Due to the capacity of Synthetic Aperture Radar (SAR) sensors to operate independently of sunlight and weather conditions, SAR image time series offer the possibility to form a dataset with almost regular temporal sampling. This paper aims at mining the SAR image time series for an analysis of target’s behavior from the perspective of both temporal evolution and coherence. The authors present a two-level analytical approach envisaging the assessment of global (related to perceivable structures on the ground) and local (related to changes occurred within a perceivable structure on the ground) evolution inside the scene. The Latent Dirichlet Allocation (LDA) model is implemented to identify the categories of evolution present in the analyzed scene, while the statistical and coherent proprieties of the dataset’s images are exploited in order to identify the structures with stable electromagnetic response, the so-called Persistent Scatterers (PS). A comparative study of the two algorithms’ classification results is conducted on ERS and Sentinel-1 data. At global scale, the results fit human perception, as most of the points which can be exploited for Persistent Scatterers Interferometry (PS-InSAR) are classified within the same class, referring to stable structures. At local scale, the LDA classification demands for an extended number of classes (defined through a perplexity-based analysis), enabling further differentiation inside the evolutional character of those stable structures. The comparison against the map of detected PS reveals which classes present higher temporal correlation, determining a stable evolutionary character, opening new perspectives for validation of both PS detection and SITS analysis algorithms.

This paper investigates the usefulness of the normalized compression distance (NCD) for image similarity detection. Instead of the direct NCD between images, the paper considers the correlation between NCD based feature vectors extracted for each image. The vectors are derived by computing the NCD between the original image and sequences of translated (rotated) versions. Feature vectors for simple transforms (circular translations on horizontal, vertical, diagonal directions and rotations around image center) and several standard compressors are generated and tested in a very simple experiment of similarity detection between the original image and two filtered versions (median and moving average). The promising vector configurations (geometric transform, lossless compressor) are further tested for similarity detection on the 24 images of the Kodak set subject to some common image processing. While the direct computation of NCD fails to detect image similarity even in the case of simple median and moving average filtering in 3 × 3 windows, for certain transforms and compressors, the proposed approach appears to provide robustness at similarity detection against smoothing, lossy compression, contrast enhancement, noise addition and some robustness against geometrical transforms (scaling, cropping and rotation).