Home – Home – Jornals – Avtometriya 2025 number 4

A method for processing a set of hyperspectral data in order to form a representative system of features is proposed, and object classification is carried out using 7 different systems of features in narrow spectral intervals (30 nm) of the visible and near IR spectral regions based on measured spectral brightness coefficients (SBC). It is shown that it is advisable to use systems of three features for classification of 12 types of vegetation and camouflage coverings. At the same time, traditional vegetation indices often used for plant research provide insufficiently high accuracy of classification of objects of selected types. Simultaneous use of two difference indices is more effective in comparison with them. However, the best classification accuracy is provided by systems of three features, which are integrated values of the SBC in specially selected spectral ranges. Note that the classification accuracy is equal to or close to 100% in almost all cases when classifying objects into two classes.

In this paper, three methods of spectral matching of remote sensing images are presented, namely, pixelwise linear, pixelwise nonlinear, and generalized nonlinear methods. Pixelwise nonlinear and generalized nonlinear methods are implemented as neural networks. The methods are compared using the Landsat-8 and Sentinel-2 images from the IEEE Data Fusion Contest 2021 dataset. According to the experimental results, the generalized nonlinear method of spectral matching of remote sensing images demonstrates the best matching quality.

Automated methods for counting the number of sunflower seedlings and constructing weed maps based on ultra-high spatial resolution RGB images obtained using an unmanned aerial vehicle are proposed. We propose automated methods for counting the number of sunflower seedlings and constructing weed vegetation maps from RGB images of ultra-high spatial resolution obtained by an unmanned aerial vehicle. The methods are based on the use of convolutional neural networks DeepLabv3+, ResNet-101, and YOLOv8. The results of experimental studies show that the accuracy of estimating the number of sunflower seedlings at early stages is 96% on the average.

The algorithm proposed by the author earlier for searching for small objects against a background of an inhomogeneous random texture has been modified. The modification allows one to work in parallel with multiple image segmentation. A weight is assigned to each segmentation, which is searched for by means of training based on the error backpropagation method. In the implemented process, five weights are determined by a set of a thousand of photographs of clouds measuring 1200×1200 pixels, on which model objects are applied - circles with a radius of two pixels. Initially, all weights are the same and are equal to 0.2. After learning, the maximum and minimum weights turn out to be 0.404 and 0.116. Graphs showing changes in weights indicate that the learning procedure converges. According to the results of the computational experiment, the modification turns out to be more effective than the original algorithm: the number of errors of the first and second kinds decreases by 1.23 and 1.8 times.

This paper suggests a method for reconstruction of optical satellite images using a neural network conditioned on co-registered synthetic aperture radar data. The solution is based on existing neural networks for image reconstruction - DiffCR and Palette, with a modified CBAM block for spatial-channel attention and a modified loss function. The experimental studies conducted show better quantitative and qualitative results in comparison with the baseline models.

A mathematical model of the characteristic of image displacement transformation into the measuring signal of the device for local vertical construction (DLVC) designed using a panoramic annular lens (PAL) and a focal plane array is considered. The deviation of the Earth's image center from the center of the annular image of space is calculated using the two-point centroid algorithm. A description of the laboratory setup for studying the DLVC transformation characteristic and experimental results are given.

The review article presents the results of an analysis of the most important feature of using special self-referential rotation units, which makes it possible to implement a calibration procedure for unique angular encoders included in high-precision angle measuring setups without using external flat angle standards.

Providing efficient operation and increasing the reliability of proton exchange membrane fuel cells (PEMFCs) require new tools for real time monitoring of their technical condition that have small-amplitude disturbances of the current operating mode. Currently, one of the most informative small disturbance methods for studying PEMFCs is impedance spectroscopy. The use of the wavelet analysis and broadband probing signals can significantly reduce the analysis time and, as a result, increase the efficiency of making diagnostic decisions. However, the problem of a significant loss of accuracy in assessing high-frequency components of the impedance has been identified. Within the framework of this work, it has been determined that the reason for this is a significant contribution of high-frequency spectral components with a low signal-to-noise ratio, reaching 4 dB. A technique for threshold signal processing based on the wavelet transform is proposed, which makes it possible to increase the accuracy of assessing the amplitude-frequency characteristics of the impedance up to 0.3% in the frequency range of 0.1-400 Hz without increasing the signal-to-noise ratio. The proposed tools of high-precision assessment of impedance characteristics based on the wavelet analysis of broadband signals will find applications both in research laboratories and in real operating power plants based on PEMFCs.

The front-end is a section of the synchrotron radiation (SR) beamline intended to interface the vacuum, radiation, and optical conditions of the SR source with those of the experimental stations. This work presents thermal analyses of key front-end components developed for undulator beamlines 1-1 and 1-2 of the Siberian Circular Photon Source (SKIF). The expected output power of the SR beam from a linear undulator is 7.1 kW. The finite element method implemented in the COMSOL® Multiphysics 6.2 software package, along with the method of the non-uniform power density distribution, is used for simulations. The paper reports the development of front-end components and 3D modeling results of temperature fields and mechanical stresses under continuous thermal loading in ultra-high vacuum conditions. Calculations are carried out for the following components: fixed mask, movable photon shutter, adjustable mask, radiation shutter, diamond vacuum windows, thermal filters, and SR beam monitors. The obtained results demonstrate that the designed cooling systems for the front-end elements efficiently dissipate the incoming thermal power of the synchrotron beam and ensure their reliable operation.

For a high-resolution spectral analysis of a fiber Bragg grating (FBG), the use of a new type of a tunable highly coherent laser, i.e., a fiber laser with vibration-induced sweeping, is demonstrated for the first time, in which wavelength tuning in the 2-nm range is induced by vibration of a part of the laser cavity. It is demonstrated that the use of the fiber laser makes it possible both to simplify the signal detection and to increase the accuracy of measuring the FBG spectra by an order of magnitude compared to schemes where pulsed self-sweeping lasers are used, due to averaging the signal over a larger number of points. The measurements show that the proposed technique can be used to resolve structures with a spectral width of about 64 MHz (~ 0.5 pm).

The paper describes the operation of a device for interrogating fiber sensors based on a Fabry-Perot interferometer (FPI) with a frame update rate of up to 70 kHz. The high performance is achieved due to the combination of a passive optical scheme based on a multichannel arrayed waveguide grating (AWG) and a phase algorithm for data processing. The possibility of high-rate interrogation of the sensor with accuracy of ~ 1% (~50 pm) of the measurement range is experimentally demonstrated.

A method compatible with the growth process has been proposed for processing reflectograms recorded in situ by a laser reflectometer during the growth of III-nitride layers by ammonia molecular beam epitaxy. This method includes data filtering based on the fast Fourier transform. The cause of an additional high-frequency signal appearing and interfering with the registration of informative interference oscillations of reflected light intensity has been established as the oscillation of the holder with the sample installed on it during growth rotation. Sample rotation is a necessary technological technique in molecular beam epitaxy growth, ensuring increased uniformity of epitaxial film parameters. A comparison of the thickness values of individual III-nitride layers, as determined by the processing of reflectograms using software, with values determined by scanning electron microscopy demonstrates good agreement between the two methods. Software has been developed that automatically filters out high-frequency signals and determines the growth rate of individual III-nitride layers grown by ammonia molecular beam epitaxy.