Home – Home – Jornals – Avtometriya 2022 number 1

The hyperspectral method for analyzing the Earth's surface is very effective in solving problems of classification of both objects located on it and the state of these objects (for example, agricultural crops). However, a full-scale hyperspectral analysis is a very expensive job, and the search for ways to reduce the cost of this procedure is quite understandable. The most logical way is to reduce the number of spectral components - classification features - by choosing (or forming from them) the most informative ones. In this paper, to implement it by using neural network technologies is proposed. By an example of processing a 200-channel hyperspectral image, it is shown that reducing the dimension of the feature space using these technologies makes it possible to achieve high-accuracy classification with the accuracy exceeding that obtained by other known methods.

The article presents a complex algorithm for combining hyperspectral and highly detailed panchromatic images. The algorithm includes preliminary procedures for the selection of contours and subsequent alignment of hyperspectral and panchromatic images at the corresponding points of their contours. As a result of precise alignment, the accuracy of spectral separation of objects in hyperspectral images is also increased. The resulting hyperspectral image has a high spectral and spatial resolution. The results of numerical studies confirm the high efficiency of the developed algorithm.

An important task of a modern radar is the study of approaches aimed at creating broadband radar systems (radars) with an adaptive phased antenna array (APAA) and the development of principles for their construction and algorithms for processing broadband signals in radars with an APAA. The main advantage of such approaches is an increase in the resolution of the system, which makes it possible to classify targets better. The article focuses on the issue of adaptive spatial filtering of active interference in a radar using digital multichannel signal processing in the frequency domain.

A method is proposed for fast bandwidth selection of kernel functions in nonparametric probability density estimation of a two-dimensional random variable with independent components. The distribution laws under study belong to the family of single-mode and symmetric probability densities. The possibility of their optimization is justified on the basis of the analysis of the asymptotic expressions of the mean square deviations of the components of a two-dimensional random variable. Each component is characterized by an optimal bandwidth of the kernel functions, which depends on the nonlinear functional of the probability density. Its functional dependence on the antikurtosis coefficient of a one-dimensional random variable is established. The effectiveness of the proposed technique is confirmed by the results of analytical studies.

A projection shadow method for measuring the diameter of cylindrical objects under conditions of a limited dynamic range of a photodetector is proposed. The method involves the use of telecentric optics and an incoherent light source. The method is based on determining the position of the edge of the object in the image based on the position of the maximum of the derivative of the intensity dependence. Theoretical and experimental estimates of the error of the proposed method are carried out. The error of measuring the diameter of cylindrical objects at the level of 0.3 * 10⁻⁵ has been achieved. The result obtained confirms the efficiency and prospects of the proposed method.

Two methods for measuring the optical response to ammonia of a sensor based on film opal-like silica materials are proposed, and the corresponding devices have been developed. The explanation of the selective response of the sensor to ammonia associated with its high solubility in water has been experimentally confirmed. Experimental results have been obtained on operation of such a sensor with ammonia concentrations from several ppm to 20vol.% with fast response.

Based on the model of equivalent diagrams, the Fraunhofer diffraction patterns (spectra) of extended (in depth) asymmetric slit-type holes with absolutely absorbing inner faces and different input (D) and output (D₁) apertures are calculated in the far field zone. The behavior of the spectrum of an extended object is studied analytically in the case of differences in apertures 2÷Dç =÷D₁-Dç noticeably smaller than the size of the Fresnel zone d_d = Öld (l is the light wavelength, and d is the hole depth). It is shown that, in the range of angles ÷qç<< q_кр= Öl/d, the observed diffraction pattern of an extended object is equivalent to the diffraction of light by a flat slit (d=0) with an effective width D_eff=D+D-q_d /(Ö2p). On the basis of a constructive approximation of the Fresnel integral function, the features of light diffraction by volumetric holes, whose apertures differ significantly from each other: 2÷Dç>> d_d, are studied analytically. Calculations have shown that, in the cases of expanding (D₁> D) and narrowing (D₁< D) apertures, the behavior of the minima of the observed diffraction patterns in the angle ranges ÷qç<÷q_Dç= ÷Dç/d differs only slightly from the equidistant behavior for a flat slit (d =0) with widths D and D₁, respectively. The results obtained can be used in the development of optoelectronic systems for dimensional inspection of plates with holes.

We have made an attempt to give an exact quantitative description and generalization for the specification of the spatial distribution of the x-ray polycapillary lens' spread function for different cases of the quality of its adjustment along angular and lateral coordinates. We have conducted series of experiments on defining the lens' spread function around the spatial areas whose offset from the focal point along the axial coordinate is large. As a result of analyzing results, we have proposed two approximation models for fitting the spread function with potentially good accuracy in the most appropriate cases of polycapillary lens' adjustment. We have demonstrated that the proposed models are universal and agree with the experimental distribution of the spread function with high accuracy by an example of a detailed analysis of three series of experiments on defining the spread function within large spatial areas after three different cases of polycapillary lens' adjustment.

The article deals with the layout of magnetometric gradiometric systems using three-component fluxgates and three-component accelerometric converters. These transducers are also widely used in attitude measurement systems. Basic mathematical models of three-component magnetometers included in the gradiometer structure are proposed, which are suitable for the ideal location of the three-component transducer in the device case. A diagram of the actual arrangement of fluxgates in the gradiometer bogy is shown, and the deviation angles of their sensitivity axes from the basis axes of the device are presented. Refined mathematical models of three-component fluxgate transducers are presented taking into account the angles of deviation of the their sensitivity axes. A technique for calibrating a three-component magnetometer with mechanical turns of the fluxgates on verification devices, installations (rotary tables) is proposed, which makes it possible to set and accurately monitor the required angles of spatial orientation of the device body. The errors of a three-component fluxgate converter are estimated by the method of computational experiment - computer simulation.

The solution of the problem of determining the coordinates of a radio beacon (including a mobile one) in real time using difference-rangefinder measurements taken by a grouping of small-sized unmanned aerial vehicles (UAV) is considered. To form this solution, the following equations are successively derived: equations describing the dynamics of changes in the navigation parameters of the UAV grouping and observers of these parameters based on GPS measurements, equations of motion of the mobile radio beacon and the observer of its navigation parameters, and, at the final stage, algorithms for filtering the UAV motion parameters by satellite measurements and the beacon motion parameters by measurements taken at the UAV. Numerical simulation of the filtering algorithms has shown the possibility of high-precision determination of the coordinates of a mobile radio beacon using the proposed approach.

The approach of applying machine learning methods for prediction of the component composition of the wide fraction of light hydrocarbons in pipeline transportation is investigated. The CatBoost library is used for building a machine learning model that allows the component composition of the mixture to be predicted with an error value of 2.263 by the MAPE metric.

Methods of interactive modeling and visualization of functionally defined objects with the help of a web browser using HTML5 and WebGL are proposed. An application for high-realism models has been developed, which is able to work with an interactive frame rate in real time. The user will be able to instantly change the appearance of the data by manipulating various display properties available through the user interface on the screen. Functional descriptions of objects that allow determining time-dependent geometric shapes, their appearance, and transformations using perturbation functions are proposed and implemented.

The paper presents a comparative analysis of various machine learning algorithms for solving the problem of predicting the heat transfer crisis during boiling in two-phase flows inside channels of various geometries. Twelve classical regression models implemented in the Scikit-learn, LightGBM, XGBoost, and CatBoost libraries, as well as neural network methods are considered. The models are compared with each other, as well as with traditional forecasting methods based on the use of skeletal tables, approximate semi-empirical ratios, and correlation formulas. Possibilities of hybrid models that combine the approach based on domain knowledge with machine learning algorithms are discussed. The results of experiments with a model that combines the CatBoost regressor with one of the traditional methods in a hybrid scheme are presented. The advantage of machine learning models over the traditional approaches is revealed. It is shown that the best performance for all metrics among machine learning models can be achieved by using ensembles of algorithms based on gradient boosting.