This PDF file contains the front matter associated with SPIE Proceedings Volume 10789, including the Title Page, Copyright information, Table of Contents, Author and Conference Committee lists.

In this paper, a simple and totally unsupervised image-based procedure is derived for the alignment of interpolated multispectral (MS) bands over the panchromatic (Pan) image. Key point of the method is the pixel-varying residue of the multivariate regression between interpolated MS bands and lowpass-filtered Pan image that is used for component-substitution (CS) pansharpening to produce a minimum mean squared error (MMSE) intensity component. Such a residue locally measures the misalignment between the datasets and, once it has been properly weighted by the projection coefficient of the MMSE intensity onto the kth band, the overlap of the lowpass and highpass components of the scene is recovered. Experiments performed on simulated Pleiades and true GeoEye-1 images show that local shifts, reasonably around two ´ or three pixels in each direction, can be largely mitigated. The coefficient of determination (CD) of the multivariate regression of interpolated MS images to lowpass filtered Pan is used to globally quantify the alignment of datasets before and after the proposed pre-processing patch. The CD of the multivariate regression of pansharpened MS bands to original high-resolution Pan is a consistent full-scale measure of spatial quality of pansharpened products.

High spatial resolution imaging data is always considered desirable in the field of remote sensing, particularly Earth observation. However, given the physical constraints of the imaging instruments themselves, one needs to be able to trade-off spatial resolution against launch mass as well as telecommunications bandwidth for transmitting data back to the Earth. In this paper, we present a newly developed super-resolution restoration system, called MAGiGAN, based on our original GPT-SRR system combined with deep learning image networks to be able to restore up to 4x higher resolution enhancement using multi-angle repeat images as input.

We present a fast and simple method of the nonuniformity determination of infrared imagers with Focal Plane Arrays (FPA). It is based on the Data Reference Method (DRM) developed at PTB. The benefit of this method is that it can be applied with inhomogeneous radiation sources of arbitrary spatial radiance temperature distribution. It only requires a sufficient temporal stability of the detected radiance distribution during the measurement process.

The nonuniformity in the response of the infrared imager, caused by the pixel-to-pixel responsivity variation, is determined with the help of the DRM. The DRM is based on at least three successively taken images of a radiation source. A first image (primary image) is taken from the source, then a second image is taken from the source with the field of view of the imager purposely shifted in the direction represented by the rows of the FPA and, finally, a third image is taken with the field of view of the imager purposely shifted in the direction represented by the columns of the FPA in relation to the primary image. From these three images a result matrix is generated which yields the respective differences between the radiance temperature responsivity values of all detector pixels of the FPA to a user-selected reference pixel. For the application of this method we distinguish between infrared imagers with actively cooled and/or actively temperaturestabilized FPAs on the one hand and infrared imagers with non-temperature-stabilized FPAs on the other hand, as the later ones generally exhibit larger temporal drifts in their detected radiance temperatures.

The Advanced Land Observing Satellite (ALOS) was launched on January 24, 2006, by a Japan Aerospace Exploration Agency (JAXA) H-IIA launcher. It carries three remote sensing sensors: the Advanced Visible and Near Infrared Radiometer type 2 (AVNIR-2), the Panchromatic Remote-sensing Instrument for Stereo Mapping (PRISM), and the Phased Array type L-band Synthetic Aperture Radar (PALSAR).

Within the framework of ALOS Data European Node (ADEN), as part of the European Space Agency (ESA), has collected 5 years of data observed in Arctic, in Europe and in Africa through the ground stations of Tromsoe (Norway) and Matera (Italy).

Some data has been repatriated directly from JAXA from the on-board recorder (in particular over Africa, outside the visibility of the stations). The data were available to the scientific users via on-request ordering from the stations through the ESA ordering system. In ordering to provide a better and easier access to the data in the framework of the ESA Third Party Missions, in 2015 ESA started a project aimed to repatriate the data from the stations, consolidate them, harmonise the format to the ESA standards.

For the PALSAR data, view the different processing levels available to the users, ESA decided to setup a dissemination system, able to process automatically at the user demand the data to the requested level (on-the-fly processing). For the optical data, instead, the decision was to systematically process the PRISM and AVNIR-2 as orthorectified products (so to a higher level in respect of what available before) with a systematic quality control.

This paper presents the functionalities of the new Level 1 orthorectified products and details the block adjustment algorithms used for refinement of geometric accuracy. A specific quality control strategy has been laid down in order to re-analyse the entire archive. Also, validation methods are explained and the final product accuracy specification are given.

The JPSS-1 (now named NOAA-20) VIIRS instrument has successfully operated since its launch in November 18, 2017. A panchromatic channel onboard NOAA-20 VIIRS is called the day-night band (DNB). With its large dynamic range and high sensitivity, the DNB detectors can make observations during both daytime and nighttime. However, the DNB night image quality is affected by the straylight contamination. In this study, we focused on Earth view data in the mid-to-high latitude of the northern and southern hemispheres when spacecraft is crossing the day/night terminators at the beginning of NOAA-20 mission. Based on on-orbit data analysis from previous VIIRS sensor onboard S-NPP mission, straylight contamination mainly depends on the Earth-Sun-spacecraft geometry, and it is also detector and scan-angle dependent. Inter-comparison investigation of straylight behavior in both SNPP and NOAA-20 instruments will be conducted to better understand straylight characteristics. The preliminary study has been performed in this paper to mitigate straylight contamination for NOAA-20VIIRS DNB night images. The effectiveness of the straylight correction algorithm, directly adapted from the S-NPP DNB, is assessed for night images in the day/night terminators. Further work has been identified to improve current straylight correction methodology and DNB-based environmental data products.NOAA-20.

TROPOMI (the TROPOspheric Monitoring Instrument) is the single instrument aboard the ESA’s Copernicus Sentinel-5 Precursor (S-5P) satellite launched on October 13, 2017 . S-5P is dedicated to monitoring our atmosphere over a period of at least 7 years. Through spectral absorption features within reflected sunlight, TROPOMI measures total column densities of a large number of gases relevant for air quality and climate. This is done at unprecedented spatial resolution of approximately 3.5-7x7 km2. At this resolution, air pollution at scales of midsized cities can be distinguished from its surroundings. In addition, it can be used to estimate other air pollutant quantities such as aerosols. Each gas has a unique spectral fingerprint across the TROPOMI channels, which range from the UV to the infrared. The Short Wavelength Infrared channel, or SWIR, measures reflected sunlight between wavelengths of 2305 to 2385 nanometers. This channel has been designed to accurately determine methane (CH4) and carbon monoxide (CO) columns To obtain scientifically useful values for both gas species, measured spectra require the SWIR channel to be calibrated to a very high precision and accuracy. The requirements are particularly important for the CH4 product to achieve the ambitious S-5P mission goals. A reliable CH4 product will prove invaluable to the atmospheric scientific community in the next decade. The first six months after the launch of S5-P were used a.o. to execute an in-flight calibration campaign. This campaign was performed as a complement to the on-ground calibration campaign. It was designed to (a) take calibration measurements that were unable to be done on-ground, (b) validate on-ground calibration and (c) prepare for the instrument calibration monitoring during regular operations. Here, we report the calibration results obtained by the SRON Level 1 team for the TROPOMI-SWIR channel. This includes the correct method and settings for background corrections (both dark flux and offset), determination of the initial pixel quality map, the evolution of the background correction and quality map during the first 10 months of flight, validation of the on-ground calibration of the instrumental spectral response function (ISRF) and stray light correction.

Remote sensing images acquired by modern multichannel sensors are usually corrupted by the mixed noise that contains both signal-independent and signal-dependent components. Information about the characteristics of this noise is usually unknown a-priori and, to be used in further image processing, it is retrieved from images using special blind methods. One of such methods is considered in the paper and its modification taking into account the high level of inter-channel correlation of remote sensing images is proposed. The method is based on line fitting into a set of cluster centers determined on basis of the fourth-order statistical moment analysis in overlapping image blocks. The modification consists in simultaneous evaluation of noise parameters for a group of channels (at least, three) of a multichannel image. To obtain the cluster noise variance estimates for a channel image, it is needed to solve a system of linear equations. The system contains the noise variance estimates evaluated in the difference images obtained for all possible pairs of channel images without repetitions. The effectiveness of the proposed modification is confirmed by numerical simulation results for TID2008 database images and by approbation results on AVIRIS hyperspectral images. It is shown that the obtained noise parameter estimation results are in good agreement with the results provided by the best existing methods whereas the operating speed of the proposed modified method is considerably higher in comparison to the analog.

Image restoration is a necessary stage in the processing of remotely sensed hyperspectral images, when they are severely degraded by blur and noise. We address the semi-blind restoration of such degraded images component-wise, according to a sequential scheme. By semi-blind, we mean introducing a minimum of a priori knowledge on main unknowns in the restoration process. For each degraded component image, main unknowns are the point spread function of the blur, the original component image and the noise level. Then, the sequential component-wise scheme amounts in a first stage to estimating the blur point spread function directly from the considered degraded component image and in a second and final stage, deconvolving the degraded channel by using the PSF previously estimated. Our contribution is to improve further the sequential component-wise semi-blind variants of a recently proposed method. In this work, modifications previously introduced separately are applied all together. All these modifications together are beneficial as they tend to make the newly proposed method as independent as possible of the data content and their degradations. The resulting method is experimentally compared against its original version and the best ADMM-based alternative found experimentally in previous works. The tests are performed on three real Specim-AISA-Eagle hyperspectral images. The component images of these images are degraded synthetically with eight real and arbitrary blurs. Our attention is mainly paid to the objective analysis of the l₁-norm of the estimation errors. Experimental results of this comparative analysis show that the newly proposed method exhibits interesting competitive performances and can outperform the methods involved in the experimental comparison.

This paper presents a novel multi-label active learning (MLAL) technique in the framework of multi-label remote sensing (RS) image scene classification problems. The proposed MLAL technique is developed in the framework of the multi-label SVM classifier (ML-SVM). Unlike the standard AL methods, the proposed MLAL technique redefines active learning by evaluating the informativeness of each image based on its multiple land-cover classes. Accordingly, the proposed MLAL technique is based on the joint evaluation of two criteria for the selection of the most informative images: i) multi-label uncertainty and ii) multi-label diversity. The multi-label uncertainty criterion is associated to the confidence of the multi-label classification algorithm in correctly assigning multi-labels to each image, whereas multi-label diversity criterion aims at selecting a set of un-annotated images that are as more diverse as possible to reduce the redundancy among them. In order to evaluate the multi-label uncertainty of each image, we propose a novel multi-label margin sampling strategy that: 1) considers the functional distances of each image to all ML-SVM hyperplanes; and then 2) estimates the occurrence on how many times each image falls inside the margins of ML-SVMs. If the occurrence is small, the classifiers are confident to correctly classify the considered image, and vice versa. In order to evaluate the multi-label diversity of each image, we propose a novel clustering-based strategy that clusters all the images inside the margins of the ML-SVMs and avoids selecting the uncertain images from the same clusters. The joint use of the two criteria allows one to enrich the training set of images with multi-labels. Experimental results obtained on a benchmark archive with 2100 images with their multi-labels show the effectiveness of the proposed MLAL method compared to the standard AL methods that neglect the evaluation of the uncertainty and diversity on multi-labels.

This paper proposes a ship classification method for synthetic aperture radar (SAR) images, which incorporates SAR geometry information into a convolutional neural network (CNN). Most of the conventional methods for ship classification employ appearance-based features. These features extracted from SAR image are not robust to a geometry change because the geometry difference significantly changes the appearance of target objects in SAR images. CNN has a potential to handle the variations in appearance. However, it requires huge training data, which is rarely available in SAR, to implicitly learn geometry-invariant features. In this paper, we propose a CNN-based ship classification method incorporating SAR geometry information. We focus on the incident angle information that is included in a metadata because incident angle change directly affects the appearance of objects. The proposed method enables a network to learn a relationship between the appearance and SAR geometry by utilizing the incident angle information as a condition. Experimental results show that the proposed method improves the classification accuracy by 1.1% as compared to the conventional CNN, which does not utilize incident angle information. Furthermore, our method requires 25% less training data as compared to the conventional CNN to achieve 70% classification accuracy.

In using the convolutional neural network (CNN) for classification, there is a set of hyperparameters available for the configuration purpose. This study aims to validate the effectiveness of the CNN architecture, i.e. AlexNet on land cover classification, based on four remotely sensed land-use land-cover (LULC) datasets. In addition to the evaluation of the impact of a range of parameters in the evaluation tests, the influence of a set of hyperparameters on the classification performance will be assessed. The parameters include the epoch values, batch size, convolutional filter size and input image size. Thus, a set of experiments were conducted to specify the effectiveness of the selected parameters. We first explore the number of epochs under several selected batch size values. The impact of the first layer kernel size of the convolutional filters also was evaluated. Moreover, testing assorted sizes of the input images provided insight of the influence of the size of the convolutional filters and the image sizes. To generalize the validation, four remote sensing datasets, AID, RDS, UCMerced and RSCCN7, which have different land coverage and are publicly available, were used in the experiments. These datasets have a wide diversity of input data, such as the number of classes, amount of labelled data and texture patterns. A specifically-designed, interactive, deep-learning GPU training platform for image classification (Nvidia Digit) was employed in the experiments. It has shown efficiency in both training, evaluation and testing. These results provide opportunities toward better classification performance in various applications, such as hyperspectral multi-temporal agricultural LULC.

Tree species information is crucial for accurate forest parameter estimation. Small footprint high density multireturn Light Detection and Ranging (LiDAR) data contain a large amount of structural details for modelling and thus distinguishing individual tree species. To fully exploit the potential of these data, we propose a data-driven tree species classification approach based on a volumetric analysis of single-tree-point-cloud that extracts features that are able to characterize both the internal and the external crown structure. The method captures the spatial distribution of the LiDAR points within the crown by generating a feature vector representing the threedimensional (3D) crown information. Each element in the feature vector uniquely corresponds to an Elementary Quantization Volume (EQV) of the crown. Three strategies have been defined to generate unique EQVs that model different representations of the crown components. The classification is performed by using a Support Vector Machines (C-SVM) classifier using the histogram intersection kernel that has the enhanced ability to give maximum preference to the key features in high dimensional feature space. All the experiments were performed on a set of 200 trees belonging to Norway Spruce, European Larch, Swiss Pine, and Silver Fir (i.e., 50 trees per species). The classifier is trained using 120 trees and tested on an independent set of 80 trees. The proposed method outperforms the classification performance of the state-of-the-art method used for comparison.

Recent advances in very high resolution (VHR) earth observation (EO) techniques have led to a massive increase in data volumes to be processed. These remote sensing (RS) processing steps are complex and heterogeneous and an optimized use of these algorithms in both High Processing Computer (HPC) and Cloud platforms are still an important and opened study field for RS data providers with actual and future EO missions. The goal of our study is to identify and develop a new architecture to deal with large and increasing quantity of data and versatile production profiles. In this study, we took the example of the actual and complete processing pipeline used in Pleiades production to deliver perfect sensor image. This pipeline is composed of heterogeneous radiometric and geometric processing steps. In the first part, we study five main big data framework solutions. As result of this study, we identify Apache Spark as the best framework to use due to its performance, great development maturity, and data resilience certification. In the second part, and to develop the new processing pipeline, we redesign the processing pipeline with separation of the metadata management and the core processing. These good practices help us to develop and reuse legacy algorithms to an operational processing pipeline compatible with big data paradigm. As result of this development, we successfully identify a generic way to develop new processes and reuse legacy algorithms with large data paradigm and by keeping great performance and, more importantly, gaining platform flexibility.

Due to the low spatial resolution of the hyperspectral cameras, the acquired spectral pixels are mixtures of present materials in the scene called endmembers. All spectral pixels are assumed to be mixtures of these endmembers with different amounts called abundances. Unmixing of the spectral pixels is a very important task for the analysis of these hyperspectral data cubes. Unsupervised unmixing aims to estimate the endmembers signatures and their abundances in each pixel without any prior knowledge about the given cube. Sparsity is one of the recent approaches used in the unmixing techniques. Solving the basis pursuit problem could be used as a sparsity-based approach to solving this unmixing problem where the endmembers are assumed to be sparse in a domain known as a dictionary. The choice of an appropriate dictionary is important for obtaining sparser representations of the given spectral pixels for better unmixing results. Two main approaches of dictionaries for sparse representation; analytic dictionaries approach and learned custom dictionaries approach. The contribution of this paper is the use of the recent online sparse coding dictionary as a learned custom dictionary for more fitting of the spectral pixels in solving the basis pursuit unmixing problem. While using the online sparse coding algorithm is more computational than any predefined transform family, but it has the advantage of no need for choosing a suitable family for the given spectral pixels. The online dictionary learning was successfully used in solving the basis pursuit unmixing problem to unmix both real AVIRIS data cube obtained from AVIRIS-NASA website and a synthetic cube made up from few materials selected from the given ASTER spectral library.

In this communication, we address the problem of unsupervised dimensionality reduction (DR) for hyperspectral images (HSIs), using nearest-neighbor density-based (NN-DB) approaches. Dimensionality reduction is an important tool in the HSI processing chain, aimed at reducing the high redundancy among the HSI spectral bands, while preserving the maximum amount of relevant information for further processing. Basically, the idea is to formalize DR as the process of partitioning the spectral bands into coherent band sets. Two DR schemes can be set up directly, one based on band selection, and the other one based on band averaging. Another scheme is proposed here, based on compact band averaging. Experiments are conducted with hyperspectral images composed of an AISA Eagle HSI issued from our acquisition platform, and the AVIRIS Salinas HSI. We evaluate the efficiency of the reduced HSIs for final classification results under the three schemes, and compare them to the classification results without reduction. We show that despite a high dimensionality reduction (< 8% of the bands left), the clustering results provided by NN-DB methods remain comparable to the ones obtained without DR, especially for GWENN in the band averaging case. We also compare the classification results obtained after applying other unsupervised or semi-supervised DR schemes, based either on band selection or band averaging, and show the superiority of the proposed DR scheme.

The upcoming launch of the European Space Agency (ESA) ExoMars 2020 rover signals a need for an analysis tool to be created which can exploit the multi- and hyperspectral data that will be returned by its Panoramic Camera (PanCam), Infrared Spectrometer for Mars (ISEM), and Close-UP Imager (CLUPI) instruments. Data processed by this analysis tool will be invaluable in (i) characterising the geology local to the ExoMars rover, (ii) relating ground-based observations to orbital Compact Reconnaissance Imaging Spectrometer for Mars (CRISM) data, (iii) detecting evidence of past habitability on Mars, and (iv) identifying drilling locations. PanCam, ISEM, and CLUPI offer spectral analysis capabilities in both spatial (140-1310 microns/pixel at 2 m working distance) and spectral (440-3300 nm) dimensions. We have developed the ExoMars Spectral Tool (ExoSpec) which functions as a GUI-based extension to ENVI + IDL and performs steps from image import and compilation into ENVI .dat format, flat-fielding, radiometric correction, radiance-toreflectance (R*) corrections using the in-scene Gretag MacBeth ColorChecker^TM, and calculation of spectral parameters. We demonstrate the functionality of ExoSpec at its current stage of development and illustrate its utility with results from field expeditions to Mars analogue terrains in: (i) geothermally altered basalts in N´amafjall, Iceland, and (ii) layered alluvial plains deposits in Hanksville, USA, using ExoMars PanCam, ISEM, and CLUPI emulator instruments.

The observed radiance at-sensor level generally records the radiation changes caused by multiple factors during the imaging process for Earth observation satellites. Most methods can not retrieve the precise reflectance because of the neglecting of the mutual effects between the atmosphere and the topography. In this study, based on a physical component model of at-surface level radiation, demonstrates that the earth-atmosphere coupling effect is playing a key role in a practical reflectance retrieval method over rugged terrains. The comparison of retrieving reflectance on sunlit areas and shady areas shows how the accuracy is affected after applying earth-atmosphere coupling correction.

Including the estimation of endmembers and fractional abundances in hyperspectral images (HSI), blind hyperspectral unmixing (HU) is one of the most prominent research topics in image and signal processing for hyperspectral remote sensing. In this paper, a method of blind HU based on online dictionary learning and sparse coding is proposed, for the condition of the spectral signatures unknown in the HSI. An online optimization algorithm based on stochastic approximations is used for dictionary learning, which performs the optimization on the sparse coding and dictionary atoms alternately. On the sparse coding, a fully constrained least squares (FCLS) problem is solved because of the physical significance of fractional abundances. To estimate the endmembers in the HSI, a kind of clustering algorithm is used to cluster the atoms in the pruned dictionary obtained via the statistics on the sparse codes. With the estimated endmembers, the final fractional abundances can be obtained by using a variable splitting augmented Lagrangian and total variation algorithm. The experimental results with the synthetic data and the real-world data illustrate the effectiveness of the proposed approach.

Remote sensing image fusion allows the spectral, spatial and temporal enhancement of images. New techniques for image fusion are constantly emerging shifting the focus from pan-sharpening to spatiotemporal fusion of data originating from different sensors and platforms. However, the application of image fusion in the field of Earth observation still remains limited. The number and complexity of the different techniques available today can be overwhelming thus preventing users from fully exploiting the potential of fusion. The aim of this study is to make fusion products more accessible to users by providing them with a simple tool for spatiotemporal fusion in Python. This tool will contribute to the better exploitation of data from available sensors making possible to bring the images to the spectral, spatial and temporal resolution required by the user. The fusion algorithm implemented in the tool is based on the spatial and temporal adaptive reflectance fusion model (STARFM) - a well established fusion technique in the field of remote sensing often used as benchmark by other algorithms. The capabilities of the tool are demonstrated by three case studies using Sentinel-2 and simulated Sentinel-3 data. The first case study is about deforestation in the Amazon forest. The other two case studies concentrate on detecting change in an agricultural site in Southern Germany and urban flooding caused by the hurricane Harvey.

Physical inverse problems found on appropriate forward models, which can have highly systematic errors. As an example, in remote sensing from satellite observations, the forward model depends on spectroscopy of atmospheric gas molecules and radiative transfer modelling, whose accuracy is not perfect. The problem of correctly addressing both error components (instrument and forward model) is one of major concern in retrieval methodology. Until now, the treatment has relied on ad-hoc strategies, which makes the retrieval algorithms sub-optimal or nonoptimal at all. Optimal estimation is based on the Gaussian assumption for noise, which is normally not satisfied in presence of forward model error. In this paper, we will show that a proper Random Projections approach can provide a) an unified and coherent treatment of systematic and random errors; b) a compression tool, which can reduce the dimensionality of the data space; c) a noise model which is truly Gaussian therefore, making it possible to apply rigorously Optimal Estimation and derive the correct retrieval error; d) a simplified treatment of the inverse algebra to get the final solution. The present paper addresses the specific point of how to fully exploit the compression capability of random projections to develop an inverse algorithm able to deal with big data, and minimal loss of information content. The approach will be exemplified for IASI (Infrared Atmospheric Sounder Interfermoter) and we will show the very first physical retrieval scheme, which exploits the full IASI spectral coverage for the simultaneous retrieval of surface and atmospheric parameters. The methodology can be applied to any inverse physical problem dealing with high-dimensionality data space, how normally arises in astrophysical and Earth remote sensing science. The performance of the methodology for the retrieval of temperature and water profiles has been assessed through comparison with radiosonde observations. The retrieval accuracy, for a tropical atmosphere, is better than ± 1.25 K and ± 1.5 g/kg for temperature and water vapour, respectively. We have also performed a retrieval exercise for the Eastern China and we have shown that air quality gases, such as CO, SO₂ and NH₃ can be simultaneously and confidently retrieved, meaning that Random Projections preserve information content of data.

Oil spill can be described as a global issue, which causes serious concern to human life and the environment, therefore early identification and remedial measures taken at an early stage is very important. Spectral Unmixing is the process of identifying the constituent spectra of a mixed pixel referred to as endmembers and computing the corresponding proportions or abundances within each pixel in a given image. Many spectral unmixing methods have being proposed in the literature based on linear or nonlinear models. Deep neural networks allow computational models that are composed of multiple processing layers to learn representations of data with multiple levels of abstractions. Deep neural networks have shown excellent performance on various task on image processing with better accuracy compared to shallow learning networks and are increasingly gaining popularity with Hyperspectral imaging. Here we propose to use deep neural network to quantify different Hydrocarbon (HCs) substances in sandy clay loam soil type. Hyperspectral data sets have been acquired using mixtures of different HCs with the soil type. Vertex Component Analysis (VCA) algorithm was used to identify the endmembers and deep neural network was used to predict the quantity of each endmember. Experimental result shows the effectiveness of the proposed method with high accuracy.

Spatial heterogeneities of substrate type, water-surface roughness and also inherent optical properties (IOPs) of the water column can pose substantial challenges to optical remote sensing of fluvial bathymetry. Development of robust techniques with respect to the optical complexities of riverine environments is then central to produce accurate bathymetry maps over large spatial extents. The empirical (regression-based) techniques (e.g., Lyzenga’s model) have widely been applied for estimation of bathymetry from optical imagery in inland/coastal waters. The models in the literature are built upon only magnitude-related predictors derived from spectral radiances/reflectances at different bands. However, optically complicating factors such as variations in bottom type and water column constituents can change not only the magnitude but also the shape of water-leaving spectra. This research incorporates spectral derivatives as shaperelated predictors in order to enhance the description of spectra through the regression-based depth retrieval. A stepwise regression is utilized to select the optimal predictors among all the possible Lyzenga (i.e., magnitude-related) and derivative (i.e., shape-related) predictors. Radiative transfer simulations are used to examine the bathymetry models in optically-complex shallow rivers by considering variable bottom-types and IOPs. The methods are also applied to a WorldView-3 image of the Sarca River located in Italian Alps and resultant bathymetry estimates are assessed using insitu measurements. The results indicate the effectiveness of spectral derivatives in improving the accuracies of depth retrievals particularly for optically-complex waters.

SAR sensors play an important role in different fields of remote sensing. One of these is Automatic Target Recognition (ATR). In this paper, a new dataset for ATR is introduced, consisting of five classes of passenger cars imaged by SmartRadar of Hensoldt Sensors GmbH. The basic characteristics of the dataset and some details of the measurement campaign are provided. The second part of the paper deals with the creation of a sufficiently large database of training samples to train a Convolutional Neural Network (CNN) to classify these cars. Since training data are not as readily available as in the EO case, training data are simulated using the CohRaS SAR simulator of Fraunhofer IOSB, which is also briefly described. The basic setup of the CNN used for the classification task is outlined and some issues arising in the classification of the training data are discussed. The paper also contains some very preliminary classification results using the CNN and the simulated training data, and a discussion of these results.

Reliable vehicle detection and tracking in wide area motion imagery (WAMI), a novel class of imagery captured by airborne sensor arrays and characterized by large ground coverage and low frame rate, are the basis for higher-level image analysis tasks in wide area aerial surveillance. Possible applications include real-time traffic monitoring, driver behavior analysis, and anomaly detection. Most frameworks for detection and tracking in WAMI data rely on motion-based input detections generated by frame differencing or background subtraction. Subsequently employed tracking approaches aim at recovering missing motion detections to enable persistent tracking, i.e. continuous tracking also for vehicles that become stationary. Recently, a moving object detection method based on convolutional neural networks (CNNs) showed promising results on WAMI data. Therefore, in this work we analyze how CNN-based detection methods can improve persistent WAMI tracking compared to detection methods based on difference images. To find detections, we employ a network that uses consecutive frames as input and computes detection heatmaps as output. The high quality of the output heatmaps allows for detection localization by non-maximum suppression without further post processing. For quantitative evaluation, we use several regions of interest defined on the publicly available, annotated WPAFB 2009 dataset. We employ the common metrics precision, recall, and f-score to evaluate detection performance, and additionally consider track identity switches and multiple object tracking accuracy to assess tracking performance. We first evaluate the moving object detection performance of our deep network in comparison to a previous analysis of difference-image based detection methods. Subsequently, we apply a persistent multiple hypothesis tracker with WAMI-specific adaptations to the CNN-based motion detections, and evaluate the tracking results with respect to a persistent tracking ground truth. We yield significant improvement of both the motion-based input detections and the output tracking quality, demonstrating the potential of CNNs in the context of persistent WAMI tracking.

Algorithms for automatic semantic segmentation of the satellite images provide an effective approach for the generation of vector maps. Convolutional neural networks (CNN) have achieved the state-of-the-art quality of the output segmentation on the satellite images-to-semantic labels task. However, the generalization ability of such methods is not sufficient to process the satellite images that were captured in the different area or during the different season. Recently, the Generative Adversarial Networks (GAN) were introduced that can overcome the overfitting using the adversarial loss. This paper is focused on the development of the new GAN model for effective semantic segmentation of multispectral satellite images. The pix2pix¹ model is used as the starting point of the research. It is trained in the semi-supervised setting on the aligned pairs of images. The perceptual validation has demonstrated the high quality of the output labels. The evaluation on the independent test dataset has proved the robustness of GANs on the task of semantic segmentation of multispectral satellite images.

Scene classification is a hot research topic in the geoscience and remote sensing (RS) community. Currently, the investigations conducted in RS domain mainly use single source data (e.g. multispectral imagery (MSI), hyperspectral imagery (HSI), or light detection and ranging (LiDAR), etc.). However, one of the RS data aforementioned merely provides one certain perspective of the complex scenes while the multisource data fusion can provide complementary and robust knowledge about the objects of interest. We aim at fusing the spectral– spatial information of the HSI and the spatial-elevation information of LiDAR data for scene classification. In this work, the densely connected convolutional neural network (DensNet), which connects all preceding layers to later layers in feed-forword manner, is employed to effectively extract and reuse heterogeneous features from HSI and LiDAR data. More specifically, a novel two-stream DensNet architecture is proposed, which builds an identical but separated DensNet stream for each data respectively. Then one of stream is utilized to extract the spectral-spatial features from HSI, the other is exploited to extract the spatial-elevation features of LiDAR data. Subsequently, the spectral–spatial–elevation features extracted in two streams are deeply fused within the fusion network which consists of two fully-connected layers for the final classification. Experimental results conducted on widely-used benchmark datasets show that the proposed architecture provides competitive performance in comparison with the state-of-the-art methods.

This paper presents a novel content-based image search and retrieval (CBIR) system that achieves coarse to fine remote sensing (RS) image description and retrieval in JPEG 2000 compressed domain. The proposed system initially: i) decodes the code-streams associated to the coarse (i.e., the lowest) wavelet resolution, and ii) discards the most irrelevant images to the query image that are selected based on the similarities estimated among the coarse resolution features of the query image and those of the archive images. Then, the code-streams associated to the sub-sequent resolution of the remaining images in the archive are decoded and the most irrelevant images are selected by considering the features associated to both resolutions. This is achieved by estimating the similarities between the query image and remaining images by giving higher weights to the features associated to the finer resolution while assigning lower weights to those related to the coarse resolution. To this end, the pyramid match kernel similarity measure is exploited. These processes are iterated until the code-streams associated to the highest wavelet resolution are decoded only for a very small set of images. By this way, the proposed system exploits a multiresolution and hierarchical feature space and accomplish an adaptive RS CBIR with significantly reduced retrieval time. Experimental results obtained on an archive of aerial images confirm the effectiveness of the proposed system in terms of retrieval accuracy and time when compared to the standard CBIR systems.

Compression is a typical stage in processing of remote sensing images. Here we consider lossy compression of images that contain noise. Then, there are, at least, two approaches – to compress images without pre-filtering and with prefiltering. The second approach is shown to have some advantages but only under certain conditions. Several metrics characterizing quality of compressed images are used. Main conclusions are in good agreement in cases of using different metrics. Recommendations concerning parameter selection for lossy compression with pre-filtering are given.

The modern optical satellite sensors capture images in stereo and tri-stereo acquisition modes. This allows reconstruction of high-resolution (30-70 cm) topography from the satellite data. However, numerous areas on the Earth exhibit complex topography with a lot of “discontinuities”. One case is tectonic fault sites, which form steep topographic escarpments enclosing narrow, deep corridors that mask parts of the ground. Built with common approaches (stereo or tri-stereo), a digital surface model (DSM) would not recover the topography in these masked zones. In this work, we have settled on a new methodology, based on the combination of multiple satellite Pleiades images taken with different geometries of acquisition (pitch and roll angles), with the purpose to generate fully-resolved DSMs at very high-resolution (50 cm). We have explored which configurations of satellites (i.e., number of images and ranges of pitch and roll angles) allow to best measure the topography inside deep and narrow canyons. We have collected seventeen Pleiades Images with different configurations over the Valley of Fire fault zone, USA, where the fault topography is complex. We have also measured sixteen ground control points (GCPs) in the zone. From all possible combinations of 2 to 17 Pleiades images, we have selected 150 combinations and have generated the corresponding DSMs. The calculations are done by solving an energy minimization problem that searches for a disparity map minimizing the energy, which depends on the likelihood for pixels to belong to a unique point in 3D as well as regularization terms. We have statistically studied which combinations of images deliver DSMs with the best surface coverage, as well as the lowest uncertainties on geolocalisation and elevation measures, by using the GCPs. Our first results suggest that an exceeding time between our acquisitions leads to DSM with a low covered area. We conclude that Stereo and Tri-Stereo acquisition in one-single pass of the satellite will systematically generate a better DSM than multidate acquisition. We also conclude that in some cases, multi-date acquisitions with 7-8 images can improve the DSM robustness compared to multi-date acquisitions with fewer images.

Traditional methods of photogrammetric image processing allow reconstructing a digital terrain model with high accuracy needed for producing different geographic information system (GIS) products such as maps, orthophoto etc. The accuracy of correspondence problem solution for good imaging conditions can reach the level of hundredth of a pixel. Recently deep learning techniques have been applied for dense stereo matching of images. They allow receiving depth information directly from stereo images and show impressive results for such tasks as scene segmentation, object detection and classification. The aim of the study was to evaluate the performance of deep learning techniques for accurate digital terrain models generation. Some of the state-of-the-art deep learning stereo matching models were evaluated along with original convolutional network. To get reliable accuracy estimation of the considered techniques the results of 3D reconstruction were compared with the reference surface obtained by 3D scanning. Also the imagery acquired from unmanned aerial vehicle during the mission on digitizing an archaeological site was used for performance evaluation by comparing with digital terrain model generated by photogrammetric technique.

Coastal environment is worldwide recognized as an important asset for mankind. Relevant threats, such as erosion and changes in the territory caused by anthropogenic activities, should be addressed appropriately to support authorities and environmental organizations. Coastline extraction procedure is a fundamental task in relation to the monitoring of coastal surroundings, public security and study on potential climate change effects. In this work a new method is proposed, which aims at improving the coastline extraction procedure by harnessing Full-Pol SAR imagery and a processing chain constituted by cascading an Autoassociative Neural Network (AANN) and a Pulse-Coupled Neural Network (PCNN). The AANNs, also known as autoencoders, have been widely used in the literature for nonlinear features extraction and component analysis. This kind of neural network is designed to replicate the input into the output layer. When this task is considered as fulfilled, a good compressed input representation must be present in the bottleneck layer, enabling the extraction of significant features. Conversely the PCNNs don’t need training stages, and are proven effective in the image processing and segmentation tasks. Describing the proposed method in a nutshell, during the first stage the AANN aims at extracting features that would help the land-sea separation process; in the next stage, the PCNN aims at producing the final segmentation and helps to perform the coastline extraction task subsequently executed. Major features of the method mainly consist of the complete processing automation and the novel architecture design which chains different neural networks to accomplish the coastline extraction task.

Quick identification of post-earthquake destroyed buildings is critical for disaster management. It can be performed in unsupervised way by comparing pre-disaster and post-disaster Very-High-Resolution (VHR) SAR images. Spatial context needs to be modeled for effective change detection (CD) in VHR SAR images as they are complex and characterized by spatial correlation among pixels. We propose a unsupervised context-sensitive method for CD in multi-temporal VHR SAR images using pre-trained Convolutional-Neural-Network (CNN) based feature extraction. The sub-optimal CNN, pre-trained on an aerial optical image dataset and further optimized for using on SAR images by tuning the batch normalization layer of the CNN, enables us to obtain multi-temporal deep features that are pixelwise compared to identify the changed pixels. Detected changed pixels are further analyzed based on the double bounce property of the buildings in SAR images to detect the pixels corresponding to destroyed buildings. Experimental results on a dataset made up of a pair of multi-temporal VHR SAR images acquired by COSMO-SkyMed constellation on the city of L’Aquila (Italy) demonstrates effectiveness of the proposed approach.

Very-High-Resolution Synthetic Aperture Radar (VHR-SAR) images are of particular interest to characterize and monitor urban areas at a global scale. While automatic urban footprint extraction from SAR images can be theoretically performed at very high spatial and temporal resolutions, in practice it requires a huge amount of processing time and memory resources for a global coverage. The paper presents a fast procedure for 1mresolution Spotlight mode scene analysis which shows limited memory requirements and robustness to one-look speckle disturbance. The proposed method adopts a multiscale approach, with a radiometrically scalable scheme. Experimental results with objective assessment on Spotlight Cosmo-SkyMED images are presented and validated on reference urban classes provided by the Copernicus Urban Atlas 2012 of the European Space Agency.

The aim of this paper is to provide a better understanding of potentialities of the new Sentinel-1 radar images for mapping the different crops in the Camargue region in the South France. The originality relies on deep learning techniques. The analysis is carried out on multitemporal Sentinel-1 data over an area in Camargue,France.50 Sentinel-1 images processed in order to produce an intensity radar data stack from May 2017 to September 2017. We revealed that even with classical machine learning approaches (K nearest neighbors, random forest, and support vector machine), good performance classification could be achieved with F-measure/Accuracy greater than 86 % and Kappa coefficient better than 0.82. We found that the results of the two deep recurrent neural network (RNN)-based classifiers clearly outperformed the classical approaches. Finally, our analyses of Camargue area results show that the same performance was obtained with two different RNN-based classifiers on the Rice class, which is the most dominant crop of this region, with a F-measure metric of 96 %. These results thus highlight that in the near future, these RNN-based techniques will play an important role in the analysis of remote sensing time series.

Synthetic Aperture RADAR (SAR) is a radar imaging technique in which the relative motion of the sensor is used to synthesize a very long antenna and obtain high spatial resolution. The increasing interest of the scientific community to simplify SAR sensors and develop automatic system to quickly obtain a sufficiently good precision image is fostered by the will of developing low-cost/light-weight SAR systems to be carried by drones. Standard SAR raw data processing techniques assume uniform motion of the satellite (or aerial vehicle) and a fixed antenna beam pointing sideway orthogonally to the motion path, assumed rectilinear. In the same hypothesis, a novel blind data focusing technique is presented, able to obtain good quality images of the inspected area, in the presence of a strong point scatterer, without the use of ancillary data information. Despite SAR data processing is a well established imaging technology that has become fundamental in several fields and applications, in this paper a novel approach has been used to exploit coherent illumination, demonstrating the possibility of extracting a large part of the ancillary data information from the raw data itself, to be used in the focusing procedure. Preliminary results – still prone to enhancements – are presented for ERS raw data focusing. The proposed Matlab software is distributed under the Noncommercial – Share Alike 4.0 – International Creative Common license by the authors.

Synthetic aperture radar (SAR) automatic target recognition (ATR) has been an interesting topic of research for decades. Existing methods perform the ATR task after image formation. However, in principle, image formation does not provide any new information regarding the classification task and it may even cause some information loss. Motivated by this, in this paper, we examine two SAR ATR frameworks that work in the phase history domain. In the first framework, we feed the complex-valued phase histories to a deep convolutional neural network (CNN) directly, and in the second one, we perform image formation, phase removal, and phase history generation before feeding the data to the CNN. CNNs are known for their superior performance on image classification tasks. The effectiveness of CNNs is based on dependency patterns in a given input. Thus, the input of CNNs is not limited to images but any input exhibiting such dependencies. Since complex-valued phase histories also have such a structure, they can be the input of a CNN. We perform ATR experiments on the Moving and Stationary Target Acquisition and Recognition (MSTAR) database and compare the results of image-based and phase history-based classification.

Change detection is an important synthetic aperture radar (SAR) application, usually used to detect changes on the ground scene measurements in different moments in time. Traditionally, change detection algorithm (CDA) is mainly designed for two synthetic aperture radar (SAR) images retrieved at different instants. However, more images can be used to improve the algorithms performance, witch emerges as a research topic on SAR change detection. Image stack information can be treated as a data series over time and can be modeled by autoregressive (AR) models. Thus, we present some initial findings on SAR change detection based on image stack considering AR models. Applying AR model for each pixel position in the image stack, we obtained an estimated image of the ground scene which can be used as a reference image for CDA. The experimental results reveal that ground scene estimates by the AR models is accurate and can be used for change detection applications.

COSMO-SkyMed di Seconda Generazione (CSG) will ensure operational continuity to the currently operating "first generation" CSK constellation. The CSG constellation will consist of two satellites in Low Earth Orbit equipped with an X-band Synthetic Aperture Radar (SAR). The CSG program is managed and co-financed by the Italian Space Agency (ASI) and the Italian Ministry of Defence. A wider portfolio of Spotlight imaging modes is offered with respect to CSK, among which there are new Spotlight civilian sub-metric modes. Furthermore, products acquired with squinted attitude (i.e. mean pitch angle during the acquisition different by zero) or with multi-swath approach could be used to serve more requests in a defined area, that otherwise could not be served because of conflict for time gap violation. Upgraded Spotlight focusing algorithms have been developed in order to correctly manage side-effects on the SAR Impulse Response Function (IRF) image quality depending on the long integration time and squinted geometry. The paper presents the new Spotlight processing algorithm enhancements providing details about design, implementation and verification results. Image quality and processor time performances have been assessed by simulated data representative of the various operational Spotlight CSG acquisition modes.

The recent advance of satellite technology has led to explosive growth of high-resolution remote sensing images in both quantity and quality. To address the challenges of high-resolution remote sensing images retrieval in both efficiency and accuracy, a distributed system architecture for satellite images retrieval by combining deep and traditional hand-crafted features is proposed in this paper. On one hand, to solve the problem of higher computational complexity and storage capacity, Hadoop framework is applied to manage satellite image data and to extract image features in parallel environment. On the other hand, deep features based on convolutional neural networks (CNNs) are extracted and combined with traditional features to overcome the limitations of hand-crafted features. Besides, object detection are integrated in the proposed system to realize accurate object locating at the time of retrieval. Experiments are carried on several challenging datasets to evaluate the performance of the proposed distributed system. Standard metrics like retrieval precision, recall and computing time under different configurations are compared and analyzed. Experimental results demonstrate that our system architecture is practical and feasible, both efficiency and accuracy can meet realistic demands.

The anomaly detector (AD) is an important topic in the exploitation of hyperspectral image (HSI). Because unsupervised AD can detect the objects whose spectra are different from their surroundings under the conditions of no spectral libraries and atmospheric correction, it becomes increasingly important in hyperspectral military and civilian applications. Although there have been many ADs proposed, the Reed-Xialoi detector (RXD) is the most typical one, which has been successfully applied to many practical applications in terms of HSI anomaly detection. However, RXD suffers from “small sample size” problem. This paper focuses on the investigation of the three issues related to anomaly detection in real HSIs. That is, how many pixels of a target will be detected as an anomaly? Then, is an anomaly responded to its proximity? Finally, what kinds of anomalies will be detected? Some well-known RX-based detectors are programmed, such as global RX, local RX, subspace RX, weight Rx and uniform target detector. And those detectors are applied to three hyperspectral datasets acquired by both the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) and Hyperspectral Digital Imagery Collection Experiment (HYDICE) sensors. Experimental results show that an anomaly pixel is detected, if at least one similar pixel is found at 8-surrounding pixels. The detected anomaly pixels respond to the selected background rather than to spatial neighborhood relationship. Moreover, the RX-based detectors can’t recognize detected anomaly attributes, and no one RX-based detector performance is statistically superior to the others.

Synthetic aperture radar (SAR) ship image classification is of great significance in the field of marine ship monitoring. Extracting effective feature representation and constructing suitable classifier can fundamentally improve the accuracy of ship classification. At present, using distance metric learning (DML) algorithm to learn effective distance metrics for classifiers has been widely used in information retrieval and face recognition, but its ability to implement SAR ship image classification is still unknown. In this paper, we show the performance of 4 feature representations and 20 DML algorithms in SAR ship classification. Experimental results show that extracting effective feature representation is essential, and the DML algorithm has the ability to learn better distance metrics.

The precisely extraction of construction areas in remote sensing images can play an important role in territorial planning, land use management, urban environments and disaster reduction. In this article, we propose a method for extracting construction areas using Gaofen-1 panchromatic remote sensing images by adopting the improved Pantex^[1] (a procedure for the calculation of texture-derived built-up presence index) and unsupervised classification. First of all, texture cooccurrence measures of 10 different directions and displacements are calculated. In this step, we improve the built-up presence index that we use the windows size of 21*21 to calculate the GLCM contrast measure instead of 9*9 according to the spatial resolution of Gaofen-1 panchromatic image. Then we use the intersection operator “MIN” to combine the 10 different anisotropic GLCM contrast measure to generate the final built-up presence index result. At last, we use the unsupervised classification method to classify the Pantex result into two classes and the one with larger cluster center is the construction area class. Confusion matrix of Beijing-Tianjin-Hebei region experiment shows that this method can effectively and accurately extract the construction areas in Gaofen-1 panchromatic images with the overall accuracy of more than 92%.

The presented study is an attempt to determine the object grouping propensity of human visual system for further human-like analysis of High-Resolution Satellite (HRS) image by using an eye tracker. Perception, which is a basic characteristic of the human visual system, enables to part the visual stimuli into meaningful parts. This parting is either based on perceptual or functional grouping. However, these perceptual groups are often different from functional categories. All objects from one functional category are perceived equivalent, irrespective of their physical appearance variation. The recent advancements in eye tracking technology have provided an excellent modality to articulate the eye movement data with inner psychological factors. In this way, the presented study has exploited the fixation data to understand the perceptual organization for HRS images. The outcomes of this study has provided the analytical solution for grouping tendency of humans along with the heat maps as evidence.

In the paper is described fast 3-staged algorithm for segmentation of low-information areas on high spatial resolution panchromatic images of Earth surface. At the first stage, an image is divided into the fragments of 128×128 pixels in size and analysis of pixels “Top of the atmosphere” reflectance is performed. At the second stage the analysis of frequency characteristic of fragments is performed. At the third stage the objects of an artificial origin such as strongly pronounced straight lines is assessed using the Hough transformation. The developed algorithm got high expert appraisal. In the case of high percent of low-informative fragments on the image the proposed algorithm provides significant decrease of GCP finding time.

The present paper has been devoted to the selection of an optimal cartographic projection for standard products of the new hydrometeorological observing system from the high elliptical orbit “Arktika-M”. Conditions of imaginary from the high orbit and the issue how to preserve spatial resolution have been analyzed. The paper has shown that usage of widely used standard projections leads to the excessive number of images. The projections representing imaginary materials with better quality with minimal amount of data have been suggested and researched.

In this presentation, we deal with the design of a filter based on the geodesic distance affinity which can able to suppress the low frequency artifacts as a post-processing step after the main Winner filter restoration. We consider the multispectral signal model in the case of the Winner filter restoration independently in each image channel. The impulse response of the obtained filter is a linear combination of generalized filters optimized with respect to different criteria. The main idea of the proposed algorithm is based on assumption that low frequency outliers and the additional noise in the different image channels are not correlate to each other, thus the affinity space formed by of the opposite channels can effectively suppress the main restoration artifacts. The performance of the proposed filter analyzed and compared in terms of the PSNR accuracy. The proposed method demonstrates the ability to suppress distortion due to the low frequency artifacts.

In consideration of the difficulty for ship detection in SAR images when the ship targets are blurred in speckle noise and clutter, a novel method for ship detection is proposed in this paper. In this approach, the discrete Shearlet features are adopted to capture the intrinsic geometrical features of ship target with discontinuities points and threshold detection method is used to get the ship targets. The SAR image is decomposed by Discrete Shearlet Transform (DST) in multiple scales to get different sub-bands, and the Shearlet coefficients of images are obtained in different sub-bands with different directions. As Shearlet coefficients of the target and the background have completely different performance properties in the high-frequency sub-bands in different directions. The Shearlet coefficients of the ship targets exhibit local maxima characteristics in high-frequency subbands in different directions, while the extreme values of Shearlet coefficients in the background are difficult to simultaneously appear in different directions. Experiments on SAR images with sea backgrounds and multiple ship targets situation have been performed. Comparison with wavelet and CFAR detection methods, the results demonstrate that the proposed method is competitive in detection rate and shape preservation.

Resurs-P satellite system is one of the recent Earth remote sensing systems deployed by Russia. Its payload consists of the high resolution multispectral imager, the average resolution imager with wide swath and the hyperspectral imaging system. Hyperspectral system consists of two imagers each registering radiation in roughly half of instruments spectral range. So the output from the hyperspectral system are two hyperspectral images representing same area of the Earth but in different spectral ranges with a slight spectral overlap. For further explanation purposes these two images are named as image ‘A’ and image ‘B’. During the on-ground processing stage images ‘A’ and ‘B’ are combined into a single hyperspectral image, covering whole instrument spectral range. During evaluation of quality of hyperspectral data it was found that modular transfer function (MTF) obtained from images ‘A’ and ‘B’ is different, resulting in better spatial resolution of image ‘A’ compared to ‘B’. This fact could pose problems in the following analysis of hyperspectral data as the obtained spectral signatures actually represent slightly different parts of the ground in two halves of an instrument spectral range. The present work describes an algorithm of MTF compensation which purpose is to mitigate difference in spatial resolution of the data, obtained from the hyperspectral imaging system of Resurs-P satellite. The proposed algorithm is based on spatial linear filtering and is applied on the data that was previously transformed to spectral radiances and spatially co-registered. The algorithm consists of two steps. On the first step the coefficients of correction linear filter defined as a window kernel are estimated. For filter estimation we choose one spectral band from image ‘A’ as a reference image with the ‘best’ MTF and one spectral band from image ‘B’. We select spectral bands from within spectral overlap range of images ‘A’ and ‘B’ so they have same spectral ranges. Then linear filter coefficients are calculated using the least square errors method, so that when applying calculated filter to image ‘B’ an image that is closest to ‘A’ is obtained. On the second step correction filter is applied to all bands in image ‘B’ to compensate its difference in MTF compared to image ‘A’. Based on the selection of reference image it is possible to estimate the correction filter that blurs higher resolution image to lower resolution (which also reduces noise) or vice versa, i.e. the filter that increases resolution (but at the cost of increased noise). Effectiveness of the proposed algorithm is evaluated on the images obtained from Resurs-P satellites. The relative difference of resolutions of ‘A’ and ‘B’ images is reduced by more than 3 times.

ATHENA is an on-going Horizon 2020 Twinning project aiming to promote remote sensing technologies for cultural heritage (CH) applications in Cyprus. ATHENA project brings together the Eratosthenes Research Center (ERC) of the Cyprus University of Technology (CUT) with two internationally leading institutions of Europe, namely the National Research Council of Italy (CNR) and the German Aerospace Centre (DLR). The project’s scope is to position the ERC regionally and stimulate future cooperation through placements at partner institutions and enhance the research and academic profile of all participants. The scientific strengthening and networking achieved through the ATHENA project could be of great benefit not only for Cyprus but for the entire Eastern Mediterranean, bearing a plethora of archaeological sites and monuments urgently calling for monitoring and safeguarding.

The preservation of CH and landscape comprises a strategic priority not only to guarantee cultural treasures and evidence of the human past to future generations, but also to exploit them as a strategic and valuable economic asset. The objective of this paper is to present knowledge transfer examples achieved from the ATHENA project through intense training activities. These activities were also designed to enhance the scientific profile of the research staff and to accelerate the development of research capabilities of the ERC. At the same time the results from the training activities were also exploited to promote earth observation knowledge and best practices intended for CH. The activities included active and passive remote sensing data used for archaeological applications, Synthetic Aperture Radar (SAR) image analysis for change and deformation detection, monitoring of risk factors related to cultural heritage sites including archaeological looting etc.

Recently, the new Geographic object-based image analysis (GEOBIA) was proposed as an alternative classification approach to pixel based ones. In GEOBIA, image segments can be depicted with various attributes such as spectral, texture, shape, deep features and context, and hence final classification can produce better land cover/use map. The presence of such a large number of features poses significant challenges to standard machine learning methods and has rendered many existing classification techniques impractical. In this work, we are interested to feature selection techniques, which are employed to reduce the dimensionality of the data while keeping the most of its expressive power. Inspired by recent works in remote sensing using Convolutional Neural Networks (CNNs), especially for hyperspectral band selection, a feature selection approach based on One-Dimensional Convolutional Neural Networks (1-D CNN) is proposed in this study. All object-based features are used to train the 1-D CNN to obtain well trained model. After testing different feature combinations, we use the well trained model to obtain their test classification accuracies, and finally we select the subset of features with the highest precision. In our experiments, we evaluate our feature selection approach on 30-cm resolution colour infrared (CIR) aerial orthoimagery. A multi-resolution segmentation is performed to segment the images into regions, which are characterized later using various spectral, textural and spatial attributes to form the final object-based feature dataset. The obtained experimental results show that the proposed method can achieve satisfactory results when compared with traditional feature selection approaches.

A high spectral resolution of remote sensing images is crucial for the precise environmental research purpose. Therefore, it is essential to preserve the high spectral resolution of satellite imagery in the pan-sharpening process. Unfortunately, the use of original panchromatic (PAN) images as high-spatial data in the pan-sharpening process is ambiguous with maintaining the spectral information about photographed objects as high as possible. Therefore, there is a reasonable need to simulate the high-spatial band. The research has involved the integration of two satellite multispectral (MS) images: Landsat-8 imagery (30 m) used as high-spectral data and Sentinel-2 imagery (10 m) used as high-spatial data. The new panchromatic bands based on Sentinel-2 data was performed. The channel combinations used in spectral indices such as: Difference Vegetation Index (DVI), Green Normalized Difference Vegetation Index (GNDVI), Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), Simple Ratio (SR), and Ratio Vegetation Index (RVI) were utilized for the simulation purpose. Either the spectral quality of sharpened images and their suitability to a classification process were verified. In the assessment stage, a visual analysis, spectral quality indices, a comparison of spectral reflectance characteristics of natural land cover, and a supervised classification were applied. The research indicates the necessity to simulate the high-spatial channel to pan-sharpening process depending on the spectral resolution of the high-spectral imagery as well as depending on the type of object of interest (OOI). The application of appropriate modifications of the original high-spatial images makes it possible to keep the spectral information about photographed objects at a higher level than using original data. Thus, it is essential for the object identification process, change detection, and land management.

The applicability of the returned spatial intensity distribution and its variation in time is discussed for the investigation of nonequilibrium and nonstationary properties of the optical path, in particular, for determining the degree of turbulent distortions, thermal nonequilibrium, transverse crosswind. The results of an experimental analysis of the spatial structures obtained in the focal plane of the registrator are considered for the case of collimated laser beams reflected from a retroreflector for the 1000 m inclined atmospheric path.

Remote sensing aerial photography could produce 3 dimensional object images for archaeological objects. Exposure time was important in getting optimal information. There were two important points of concern in this exposure time, the first was the lighting conditions as a source of energy in remote sensing, including the position of the existence of objects that exist in the external environment that was much influenced by the existence and shooting time. The second point was vehicle conditions of unmanned air vehicle to environmental air condition. In 3 dimensional modeling, It was important to got the detail of archaeological information from the observed facade. Passive systems in remote sensing rely heavily on sources of energy derived from the sun and the intensity of light. This study aims to determine when the most optimal time in a day to produce aerial photographs with the best results by relying on the results of information obtained with a fundamental model of 3 dimensional point cloud for archeological objects of the temple that requires information detail of the object. Multirotor Unmanned Air Vehicle aircraft as device to took aerial photographs, structure in motion modeling and bundling adjustment method to produce the recorded object display. Archeological objects need detailed information for carving reliefs and architecture of the buildings. The results obtained depend solely on the variety of recording objects to obtain the most optimal point cloud data. Hopefully with the more detailed data obtained based on the position of energy sources and weather conditions expected results obtained could be better and efficient.

The application of drill core hyperspectral data in exploration campaigns is receiving great interest to obtain a general overview of a mineral deposit. However, the main approach to the investigation of such data is by visual interpretation, which is subjective and time consuming. To address this issue, recently, the use of machine learning techniques is proposed for the analysis of this data. For drill core samples that for which only very little prior knowledge is often available, applying classification algorithms which are supervised learning methods is very challenging. In this paper, we suggest to use clustering (unsupervised) methods for mineral mapping, which are similar to classification but no predefined class labels are needed. To handle mapping of the very highly mixed pixels in drill core hyperspectral data, we propose to use advance subspace clustering methods, in which pixels are assumed to lie in a union of low-dimensional subspaces. We conduct a comparative study and evaluate the performance of two well-known subspace clustering methods, i.e., sparse subspace clustering (SSC) and low rank representation (LRR). For the experiments, we acquired VNIR-SWIR hyperspectral data and applied scanning electron microscopy based Mineral Liberation Analysis (MLA) for two drill core samples. MLA is a high resolution imaging technique that allows detailed mineral charactrisition. We use the high-resolution MLA image as a reference to analyse the clustering results. Qualitative analysis of the obtained clustering maps indicate that the subspace clustering methods can accurately map the available minerals in the drill core hyperspectral data, especially in comparison to the traditional k-means clustering method.

The work in this paper deals with moving object detection (MOD) for single/multiple moving objects from unmanned aerial vehicles (UAV). The proposed technique aims to overcome limitations of traditional pairwise image registrationbased MOD approaches. The first limitation relates to how potential objects are detected by discovering corresponding regions between two consecutive frames. The commonly used gray level distance-based similarity measures might not cater well for the dynamic spatio-temporal differences of the camera and moving objects. The second limitation relates to object occlusion. Traditionally, when only frame-pairs are considered, some objects might disappear between two frames. However, such objects were actually occluded and reappear in a later frame and are not detected. This work attempts to address both issues by firstly converting each frame into a graph representation with nodes being segmented superpixel regions. Through this, object detection can be treated as a multi-graph matching task. This allows correspondences to be tracked more reliably across frames, which does not necessarily have to be limited to frame pairs. Building upon this, all detected objects and candidate objects are reanalyzed where a graph-coloring algorithm performs occlusion detection by considering multiple frames. The proposed framework was evaluated against a public dataset and a self-captured dataset. Precision and recall are calculated to evaluate and validate overall MOD performance. The proposed approach is also compared with Support vector machine (SVM), linear SVM classifier, and Canny edge detector detection algorithms. Experimental results show promising results with precision and recall at 94% and 89%, respectively.

Based on the laser range profiles (LRPs), the information of target geometry and movement coefficients can be detected and deduced. In this paper, the theoretical expression is given as well as the convolution model of the target’s LRPs. Primarily, a simulation method for obtaining complex target’s LRPs is introduced and illustrated, which is based on the 3-D modeling and OpenGL. The projection and shading process and the procedure of acquiring the normal of model facet are explained. By using this method the LRPs of complex target in different conditions are obtained conveniently and quickly, then the influences of target’s attitudes, shape, scattering characteristics of surface and emitted pulse width on target’s LRPs are simulated and analyzed. This work is useful to produce abundant LRPs data for target recognition, classification and reflection tomography imaging.

In this paper the target detection based on deep convolution neural network (DCNN) and transfer learning has been developed for synthetic aperture radar (SAR) images inspired by recent successful deep learning methods. DCNN has excellent performance in optical images, while its application for SAR images is restricted by the limited quantity of SAR imagery training data. Transfer learning has been introduced into the target detection of a small quantity of SAR images. Firstly, by some contrast experiments to transfer convolution weights layer by layer and analyze its impact, the combination of fine-tuned and frozen weights is used to improve the generalization and stability of the network. Then, the network model is improved according to the target detection task, it increases the network detection speed and reduces the network parameters. Finally, combining with the complicated scene clutter slices to train the network, the false alarm targets number of background clutter is reduced. The detection results of complex multi-target scenes show that the proposed method has good generality while ensuring good detection performance.

Deep learning technology such as convolutional neural networks (CNN) has achieved outstanding results in the field of crops classification for remote sensing images. The way of land cover or crop types remote sensing classification using CNN is mainly pixel-based classification which is often affected by the phenomenon of “salt and pepper”. In order to reduce this effect, an object-oriented crops classification method based on CNN is proposed in this paper. By combining image segmentation technology and CNN model, we use this method to obtain the results of crops classification from Sentinel-2A multi-spectral remote sensing images in Yuanyang County, Henan Province, China. The experiment show that, compared with the pixel level classification based on CNN which only consider the spectral and temporal characteristics of the crops, the method we proposed comprehensively utilizes more detailed information such as spectral feature, texture feature, spatial relationship, and color space. Thus, it gains a better discriminability for some specific crop and achieves higher classification accuracy.

Stricter requirements for new satellite systems for remote sensing of the Earth (ERS) require the use of video cameras of different frequency ranges with high resolution, which leads to a significant increase in the volume of images. Applied in modern remote sensing satellites SPOT-5, IRS-P5, Beijing-1 and Cartosat-1,2 JPEG method, and on satellites IMS-1, x-SAT, PLEIADES-HR, Radarsat-2 and TerraSAR method JPEG-2000 have a number of limitations due to the fact that they were developed in the era of television standards NTSC, PAL and SECAM. It is known that the procedures of compression and decompression, protective coding and decoding are symmetrical. This is the basis for using orthogonal transformations in these procedures. The choice of DCT in these algorithms in their development was due, including, small size of personnel. However, modern practice shows that not always DCT allows you to successfully compress the image and keep it informative at a resolution of 4K, 8K or more. The explanation is simple-the size of the small parts in the high-resolution image is commensurate with the size of the DCT matrix. One of the ways to solve the problem of high-quality image processing with high resolution is the creation and use of new compression filters, new algorithms of noise-resistant coding, based on the use of orthogonal and quasi-orthogonal matrices of large sizes.

Speckle noise exerts a noticeable impact on the quality of synthetic aperture radar (SAR) images. It could harm their applications in the remote sensing field, for example, the land cover classification. This paper presents a SAR image despeckling method by using the variance constrained convolutional neural network (CNN). We exploit the significant distinction between speckle noise and ground truth from the viewpoint of their statistical characteristics. The estimated noise variance as well as a weighting factor is introduced into the loss function. It can drive the learning of network to produce the result with more dispersion. After the model training, the variance constrained CNN could generate the despeckled SAR image by means of noise matrix estimation from an input contaminated by strong speckle. Finally, experiments on synthetic SAR images are conducted to demonstrate its effectiveness. It indicates that the proposed method is not only independent of image background in training, but also outperforms the classical SAR despeckling CNN.

Archival aerial photographs are a valuable source of information on the land cover. Unfortunately, these are singlechannel, monochromatic data, which means that the interpretation possibilities of these data are significantly limited. Therefore, a research was conducted in order to increase the possibility of detecting and identifying photographed elements of both natural and anthropogenic land cover. In this paper, a semi-automatic method of coloring archival photographs that uses image processing tools used in popular remote sensing software is presented. It is based on the segmentation, classification, pseudocoloring and pansharpening process. The tests were carried out on a set of aerial photos acquired in the 1950s, where mainly agricultural and forest areas with single rural buildings were photographed. Evaluation of the developed method was done through a visual analysis of the generated color images. The visual assessment was supplemented with a calculation of the value of the color accuracy index for each land cover class tested, i.e., forests, low vegetation, bare soils, water and anthropogenic objects. The presented method gives the opportunity to increase the visual quality of aerial images by giving them colors similar to natural ones while maintaining the level of detail. The visual enhancement of archival images, on the other hand, enables the automation of the identification process and analysis of photographed objects that have so far been performed manually only based on the interpreter's experience.

The Himawari-8 geostationary meteorological satellite managed by the Japan Meteorological Agency (JMA) began operations on 7 July 2015, replacing the previous MTSAT-2 operational satellite. Himawari-8 is leading the way to a new generation of satellite meteorology. Himawari-8 has a new optical payload called the Advanced Himawari Imager (AHI). AHI has 16 radiometric channels in the visible, near-infrared and infrared, from 0.47 m to 13.3 m. Better spectral, spatial, and temporal resolution improves quality and number of critical data products. The instrument will also be flown on GOES-R and COMS-2, with customized changes to radiometric coverage. An independent analysis of the INR processing chain was conducted, which characterizes the geometric error not accounted for by JMAs operational processing. The innovative approach incorporates four stages: (1) landmark measurement generation; (2) estimation of co-registration and spacecraft/instrument attitude; (3) Geometric Error Characterization (GEC) analysis; (4) Verification. Starting with images that have been resampled to a fixed grid, landmark measurements are generated in 10 of the 16 channels (except for WV channels) using a sub-pixel accurate auto-correlation technique. Version 2.0 of the Shuttle Radar Topography Mission Water Body Data was used for the coastline map. Co-registration estimates are derived using the 10.4 m channel as reference and compared with results from JMA. Spacecraft and instrument attitude are estimated, in terms of three Euler parameters, using a simple and intuitive measurement model. Systematic errors were removed from the instrument attitude time series by parsing the errors into repeatable and non-repeatable components. Verification was performed by applying the results of the GEC to the resampling process and the four stages were executed again. Initial findings show that GEC captured an error with a diurnal periodicity and peak-peak magnitude of 14 rad. The source of this error appears to be thermal deformation that is not being compensated for in JMAs INR processing. Its not possible at this time to determine if the error originates within the optics or the AOCS-instrument interface.