Hyperspectral imaging instruments allow data collection in hundreds or even thousands of spectral bands (at different wavelength channels) for the same area on the surface of the Earth. The resulting multidimensional data cube typically comprises several GBs per flight. Due to the extremely large volumes of data collected by imaging spectrometers, hyperspectral data compression, dimensionality reduction and Compressive Sensing (CS) techniques has received considerable interest in recent years. These data are usually acquired by a satellite or an airbone instrument and sent to a ground station on Earth for subsequent processing. Usually the bandwidth connection between the satellite/airborne platform and the ground station is reduced, which limits the amount of data that can be transmitted. As a result, there is a clear need for (either lossless or lossy) hyperspectral data compression techniques that can be applied on-board the imaging instrument. In this paper, a parallel implementation for a spectral compressive acquisition method on a Jetson TX1 platform is evaluated, both in terms of computational performance and energy consumption. Herein we focus on the coder-side of the method because it is the one to be performed onboard, while the decoder side may be performed on Earth station where plenty of hardware may be available. In this implementation, the measurement process is carried out with Gaussian i.i.d. matrices of double precision. The Jetson TX1 used in the experiments is a low power consumption platform that works with a maximum power of 15W with a voltage input of 5.5V-19.6V DC. It incorporates a Nvidia Maxwell GPU with 256 NVIDIA CUDA cores, a quad-core ARM cortex-A57 MPCore processor, 4GB LPDDR4 memory and 16GB eMMC 5.1 flash storage. Our study, based on a hyperspectral image of total size of $600 \times 512$ pixels (275 MB), reveals that the implementation of the proposed method using the Jetson TX1 board has promising results what concerns with real-time response under onboard energy constraints.

Collaborative representation has been a popular classifier for hyperspectral image classification because it can offer excellent classification accuracy with a closed-form solution. Collaborative representation can be implemented using a dictionary with training samples of all classes, or using class-specific sub-dictionaries. In either case, a testing pixel is assigned to the class whose training samples offer the minimum representation residual. The Collaborative Representation Optimized Classifier (CROC) was developed to combine these two types of collaborative representations to achieve the balance for optimized performance. The class-specific collaborative representation involves inverse operation of matrices constructed from class-specific samples, and the all-class version requires inversion operation of the matrix constructed from all samples. In this paper, we propose a low-complexity CROC to avoid redundant operations in all-class and class-specific collaborative representations. It can further reduce the computational cost of CROC while maintaining its excellent classification performance.

In order to increase Signal to Noise Ratio of very high resolution Earth observation satellites images, Time Delay Integration (TDI) sensors are usually used. A TDI device synchronizes the electronic charge transfer with the satellite speed to virtually increase exposure time and thus signal to noise ratio. In this paper, we present an on-board satellite implementation of multiframe registration to improve classical TDI performances. A solution to no longer constrain the number of lines to be accumulated would be to compute, in real-time, the shift between each line and resample them before summation. A motion sensor dedicated to shift measurements would be added in the focal plane. Then a fast real-time algorithm will compute shift between two consecutives images delivered by the motion sensor. The optimization study of the motion sensor and the performances oh the gradient-based algorithm on these images will be presented.

When the electron plasma is blown from the solar wind and enters into the earth atmosphere, a large number of neutral particles are excited to cause the significant event called aurora phenomenon. In this process, there are several sources of excitation including electron impact, dissociative recombination, thermal electron excitation. Particularly, auroral optical radiation produced by electron impact on oxygen atoms is investigated to explore the relationship between secondary electron energy and spectroscopic emission features. Based on the ground observations of aurora spectral images, the emission characteristics reveal the primary electron energy and flux, the basic atmosphere of species concentration and electron temperature, abundant information of the deposited particles. With the consideration that the radiations of atomic oxygen 5577 Å and 6300 Å are representative spectral lines in aurora, we designed a scheme to determine the radiation feature at other spectral bands by using the emission rate ratios of 5577/6300. And the according numerical results are represented with approximated curves varies with the secondary electron energy. Then an inverse transform is proposed to obtain the virtual optical imagery in the auroral ionosphere. It considers atmospheric radiation attenuation to deduce the real-time total flux of bombarding electrons in the atmosphere corresponding to each ground-based image’s hiding radiation condition.

The object recognition is one of most complicated tasks that can be solved using image processing. In this work we understand the aerial object recognition as a determination of object type. The set of aerial object types includes various kinds of airplanes, UAVs and helicopters. The common approaches for recognition task solving are the descriptor based approaches and convolution neural networks. The descriptor based approaches include a number of object description techniques based on features (SURF, ORB), Fourier descriptors, Hough moments etc. All these approaches are invariant to object image shift and rotation. However, the most of the listed approaches have some significant problems. Contour descriptor based algorithm is less computationally complicated. It is invariant to the object shift, rotation and scaling, insensitive to the additive noise and precise enough. Thereby, the algorithm, based on contour description calculation is most suitable for object recognition in real time image processing systems. The proposed orientation estimation algorithm consists of 2 stages: learning and recognition. Learning stage is devoted to the exploring of studied objects. The recognition stage is focusing on matching process between an observed image descriptor and the reference image descriptors. The result of the matching produces the measure of the difference between the captured object and the n-th reference object from the database. The object recognition is processed by minimization of this measure. The hardware implementation of this algorithm is divided into two parts. The learning is performing on PC. The estimation is performing on-board using the result object descriptor set loaded to on-board computation system memory. The on-board system model was designed on Virtex 7 evaluation kit. The proposed algorithm was modeled using Matlab. The experimental examinations of the algorithm were performed. We use a number of natural and synthesized videos to determine the algorithm performance. The reference database includes various kinds of airplanes, helicopters and UAVs. The examination result shows that the proposed algorithm reaches the relatively high precision and insensitive to the additive noise.

In modern hard conditions for the whole worldwide oil production industry the problem of increasing volumes of produced oil has recently become vital. This problem concerns the existing oil fields, because due to low crude oil prices the possibilities to drill new ones has almost disappeared. In this paper we describe a novel approach of oil production enhancement, based on the online procedures of all oil field data processing. The essence is that we have a dynamic oil field model that allows to simultaneously handle the information, stored tNavigator, Schlumberger ECLIPSE 100/300 formats in parallel. The model is developed on the basis of modern deep neural networks and evolutionary calculation in hybrid combination with the ‘classical’ methods. An example of successful industrial experiment is depicted.

SAR imagery requires image formation through successive signal processing of received data before browsing images and acquiring information. It is hard to distinguish which part of the data includes valuable information before image formation. The received signal data records of ALOS-2/PALSAR-2 are kept in the onboard mission data storage and transmitted to the ground. In order to compensate the storage usage and the volume of transmittable data through the mission date communication networks, the operation duty of PALSAR-2 is limited. In addition to this limitation, 3m resolution imagery is mostly planned for domestic land observations and global areas are observed with lower resolution modes. This balance strongly relies on the network availability. The observation operations of the present spaceborne SAR systems are planned by simulating the balance, given conflicting user demands, which results in compromised operations and the potential of the spaceborne SAR systems. This problem should be solved such that we do not have to compromise the operations and the potential of the next-generation spaceborne SAR systems. One of the solutions is to further compress the SAR data through onboard signal processing and information extraction from the processed images. This is also beneficial for event-driven observations and fast delivery of information products. JAXA is developing evaluation models of FPGA-based signal processing system for onboard SAR image formation. The evaluation model developed in 2016 can process a 10m resolution single look complex image from ALOS/PALSAR data. The processing speed of the present model resulted in promising. Single image formation of PALSAR will complete in 20 seconds even if the clock speed limited to 100MHz that is equal to the clock speed of Xilinx’s space grade FPGA such as Virtex-5QV. The image processed by the model is in no way inferior to the image processed with single precision computing in MATLAB.

Hyperspectral Imaging (HI) assembles high resolution spectral information from hundreds of narrow bands across the electromagnetic spectrum, thus generating 3D data cubes in which each pixel gathers the spectral information of the reflectance of every spatial pixel. As a result, each image is composed of large volumes of data, which turns its processing into a challenge, as performance requirements have been continuously tightened. For instance, new HI applications demand real-time responses. Hence, parallel processing becomes a necessity to achieve this requirement, so the intrinsic parallelism of the processing algorithms must be exploited. In this paper, a spatial-spectral classification approach has been implemented using a dataflow language known as RVC-CAL. This language represents a system as a set of functional units, and its main advantage is that it simplifies the parallelization process by mapping the different blocks over different processing units. The spatial-spectral classification approach aims at refining the classification results previously obtained by using a K-Nearest Neighbors (KNN) filtering process, in which both the pixel spectral value and the spatial coordinates are considered. To do so, KNN needs two inputs: a one-band representation of the hyperspectral image and the classification results provided by a pixel-wise classifier. Thus, this algorithm is divided into three different stages: a Principal Component Analysis (PCA) algorithm for computing the one-band representation of the image, a Support Vector Machine (SVM) classifier, and the KNN-based filtering algorithm itself. The parallelization of these algorithms shows promising results in terms of computational time, as the mapping of them over different cores presents an important speedup. Consequently, experimental results demonstrate that real-time processing of hyperspectral images is achievable.

There is a trend to develop swarm groups of ultra-small spacecraft (USS) in the area of Earth remote sensing (ERS) that led to a necessity to miniaturize computers and USS software without damage of their functional qualities. According to specialists’ forecasts more than a half of launched USS will pertain to ERS swarm groups by 2020. It opens a new market where consumers of ERS data will be ordinary users and small businesses. USS groups can update data several times per day that discovers completely new possibilities for cartography, monitoring of various processes and development of cyber-physical systems. Task of processing and analysis of obtained ERS data on USS board is actual taking into consideration significant expenditures of energy for data transfer and low productivity of computers. Thereby, there is an issue of effective and reliable storage of the general information flow obtained from onboard systems of information collection, including ERS data, into a specialized data base (DB). The paper has considered peculiarities of database management system (DBMS) operation with the multilevel memory structure. Components of the DB physical structure have been determined. Known methods of access have been analyzed from the point of view of their usage for solution of tasks to design an onboard DBMS. Their labor intensity has been estimated. Analysis of above mentioned methods of access have shown that the method of the binary aligned tree for access by the primary, unique or external key better corresponds to task terms. The paper has suggested rejecting indexing by the external key and executing search of a record under check of the referential integrity by sequential scanning of the table in order to reduce volumes of the used memories. For storage of data in DB the format has been developed that describes a DB physical structure which contains required parameters for information loading: DB headline, table descriptors, field descriptors, reference descriptors, list of data blocks for tables. Data record contains information on values of all fields of the record in a binary code corresponding to the type of such fields. There are fields containing information on location of a record in all trees of indices according to which the record is indexed in the record structure in order to obtain values of records sequentially by the key. Such structure allows reducing a memory size occupied by DB because it is not necessary to store values of keys separately. Experiments of the developed USS realization have executed in language С++ in the programming system Microsoft Visual Studio Community with possibility of cross-platform transfer of the executable code. Besides, possibility of USS realization as a mobile application in operating systems Android and iOS has also been considered. Data changed within operation are stored in the random-access memory and, if necessary, can be copied into the nonvolatile FLASH-memory. Both multiuser access to separated data and parallel transactions where a user views changes only of those data which this user made, can be organized on the basis of such architecture.

Supervised classification allows handling a wide range of remote sensing hyperspectral applications. The spatial organization of the pixels over the image has proven to be beneficial for the interpretation of the image content, thus increasing the classification accuracy. Different methods are widely used to this end: segmentation, morphological profiles, attribute profiles, etc. Denoising in the spatial domain of the image has also been shown as a technique that enhances the structures in the image. In this paper the authors propose in a multi-component noise reduction method in order to increase the classification accuracy. The method com- bines a feature extraction based on a 1D-discrete wavelet transform (DWT) applied in the spectral dimension followed by an Extended Morphological Profile (EMP) and a classifier (SVM or ELM). The multi-component noise reduction is applied to the EMP just before the classification. The proposed noise reduction is applied to each component of the EMP. The denoising recursively applies a separable 2-D DWT after which the number of wavelet coefficients is reduced by using a threshold. Finally, inverse 2D-DWT filters are applied to reconstruct the noise free original component. The impact of the number of times that the wavelet filter is applied and of the threshold value are studied, as they determine the amount of noise removed. The accuracy improvements obtained by this classification scheme de- pend on characteristics of the images such as their spatial resolution and on characteristics of the reference data used for the supervised classifica- tion. The computational cost of the classifiers as well as the cost of the whole classification scheme is high but it is much reduced achieving real time behavior for some applications through their computation on Nvidia GPU platforms.

Hyperspectral imaging systems provide images in which single pixels have information from across the electromagnetic spectrum of the scene under analysis. These systems divide the spectrum into many contiguos channels, which may be even out of the visible part of the spectra. The main advantage of the hyperspectral imaging technology is that certain objects leave unique ‘fingerprints’ in the electromagnetic spectrum, known as spectral signatures, which allow to distinguish between different materials which may look like the same in a traditional RGB image. Accordingly, the most important hyperspectral imaging applications are realted with distinguishing or identifying materials in a particular scene. In hyperspectral imaging applications under real-time constraints, the huge amount of information provided by hyperspectral images have to be rapidly processed and analysed. For such purpose, parallel hardware devices, such as Field Programmable Gate Arrays (FPGAs) are typically used. However, developing hardware applications typically requires expertise in the specific targeted device, as well as in the tools and methodologies which can be used to perform the implementation of the desired algorithms in the specific device. In this scenario, the Open Computing Language (OpenCL) emerges as a very interesting solution in which a single high-level synthesis design language can be used to efficiently develop applications in multiple and different hardware devices. In this work, the Fast Algorithm for Linearly Unmixing Hyperspectral Images (FUN) has been implemented into a Bitware Stratix V Altera FPGA using OpenCL. The obtained resutls demonstrate the suitability of OpenCL as a viable design methodology for quickly creating efficient FPGAs designs for real-time hyperspectral imaging applications.

Hyperspectral image unmixing is an important part of hyperspectral data analysis. The mixed pixel decomposition consists of two steps, endmember (the unique signatures of pure ground components) extraction and abundance (the proportion of each endmember in each pixel) inversion. Particle Swarm Optimization (PSO), which was improved into a discrete particle swarm optimization (DPSO) algorithm, has been applied to accurately extract endmembers with high optimal performance. The D-PSO algorithm needs a large number of iterative loops which takes a lot of time. Considering the high computational amount, the DPSO endmember extraction algorithm is time-consuming for hyperspectral image analysis. Developing fast processing techniques for DPSO has received considerable attention, especially in analysis scenarios with real-time constraints. The DPSO endmember extraction algorithm was first implemented with MATLAB, then accelerated by using NVIDIA Tesla K20GPU with 2496 CUDA cores. In the experiment, several key calculation nodes of the algorithm was designed in parallel: the update of the best fitness, the update of the optimal positions (both each particles’ best position and global best position), and the update of the particles’ speed and positions. Meanwhile, different combination schemes of particles were tested in the experiments. The global memory, shared memory, L1 cache, L2 cache and L3 cache were carefully designed in the proposed parallel implementation for further improving the performance of the algorithm. And real hyperspectral data collected by the Airborne Visible-Infrared Imaging Spectrometer (AVIRIS), which was built by the National Aeronautics and Space Administration (NASA)’s Jet Propulsion Laboratory, used in the experiments. Results show that the computational performance of DPSO based on GPU has significantly improved in the analysis of hyperspectral images.

It has been already made, calibrated and tested a geometry normalized electromagnetic system (GNES) for metal defect examination. The GNES has an automatic data acquisition system which supporting the efficiency and accuracy of the measurement. The data will be displayed on the computer monitor as a graphic display then saved atomatically in the Microsoft Excel format. The transmitter will transmit the frequency pair (FP) signals i.e. 112.5 Hz and 337.5 Hz; 112.5 Hz and 1012.5 Hz; 112.5 Hz and 3037.5 Hz; 337.5 Hz and 1012.5 Hz; 337.5 Hz and 3037.5 Hz. Simultaneous transmissions of two electromagnetic waves without distortions by transmitter will induce an eddy current in the metal. This current in turn will produce secondary electromagnetic fields which are measured by the receiver together with the primary fields. Measurement of percent change of vertical component of the fields will give the percent response caused by the metal or the defect. The response examinations were performed by the models with various type of defect for the master curves. The materials of samples as a plate were using Aluminium, Brass, and Copper. The more of the defects is the more reduction of the eddy current response. The defect contrasts were tend to decrease when the more depth of the defect position. The magnitude and phase of the eddy currents will affect the loading on the coil thus its impedance. The defect must interrupt the surface eddy current flow to be detected. Defect lying parallel to the current path will not cause any significant interruption and may not be detected. The main factors which affect the eddy current response are metal conductivity, permebility, frequency and geometry.

Data of Earth remote sensing allow obtaining information on states of objects and territories being inaccessible for ground methods. Development of high volumes of heterogeneous spatial-temporal data, detection of significant information and knowledge, application of geoinformation methods for analysis and processing of geographical information on spatial-temporal objects, actions and processes allow increasing efficiency of application of Earth remote sensing data. There is an issue of availability of obtained data for finite users taking into account an access rate to information within usage of Earth remote sensing instruments. So, task of development of effective algorithms for search, indexing and cataloging of aerospace images is topical taking into consideration fast increasing volumes and heterogeneity of Earth remote sensing results. Cataloging of aerospace images is considered as creation of an archive for images and metadata attached to them with additional characteristics. ISO is used as a standard according to which metadata contain the following information: general description of data, location of data, date of obtaining, information on a supplier. List of metadata allows determining a specific object of geographical data and executing a direct request to it. Application of the geoinformation platform mutually with operative aerospace information allows significantly reducing time of access, processing and preparation of data that will increase quality of the provided information under solution of tasks of applied nature. Combined storage of aerospace images, digital terrain maps and relational databases generates a geo-relational structure based on geometrical cartographic models. There is a problem of conformity of heterogeneous geographical data by values of attributive data under usage of geo-relational structures. For realization of the cataloging process it is necessary to execute automatic linkage of new images to the cartographic base. For automation of this process we suggest to use correlated algorithms and spectral analysis. Three-stage algorithm has been suggested for combination of aerospace images being heterogeneous in time on the basis of automatic fitting of reference fragments, removal of information redundancy and sub-pixel improvement. We suggest to use a dyadic correlation calculated using the Vilenkin-Christenson transformation so that the correlation process has a fast algorithm and its realization is simple. Algorithms for data clustering allow searching of images by criteria of presence of specified elements. Modernized algorithm of the fuzzy clustering compares a RGB-value of each pixel with a value of the group center: the higher the value of belonging is, the higher possibility that the pixel belongs to this group is. Separation of various elements on the aerospace image allows expanding possibilities for search of images in the base of geographical data adding nonstandard parameters. Suggested algorithms are oriented to realization in systems for data processing on the basis of neural networks being able to training and self-training. Principles of such systems are realized in high-performance central processing units for digital signal processing of original architecture NeuroMatrix. Systems for distribution of Earth remote sensing data LandSAT and GeoNetCast have been used under execution of experimental researches.

Based on the scattering effect of optical vortex, the remote sensing for aerosol particles in marine atmosphere is investigated. The influences of orbital angular momentum (OAM) mode and beam width of Laguerre-Gaussian (LG) beam on the performance on the remote sensing for aerosol particles in marine atmosphere are analyzed numerically. The scattering properties of aerosol particles by LG beam are more sensitive to the change of relative humidity, compared to that of plane wave and Gaussian beam (GB). Remote sensing systems based on LG beam with OAM modes show superiority over traditional plane wave and Gaussian beam cases in detecting the changes in relative humidity of marine atmosphere aerosol. Results also indicate that, when the number of OAM modes p (l) increase, the Radar Cross-Section (RCS) of aerosol particles alternating appear extreme values in the forward and backward scattering, because of the special ring-shaped distribution of LG beam. The forward and backward scattering of aerosol particles decrease with the increase in beam waist. As beam waist is less than the radius of aerosol particle, there exists a minimum value in the forward direction.