RU2476825C2 - Method of controlling moving object and apparatus for realising said method - Google Patents

Abstract

FIELD: radio engineering, communication.

SUBSTANCE: cartographic reference and current images of a territory are obtained, variation parameters of the current cartographic image are compared with the reference cartographic image and noise is filtered from the images. Before comparing variation parameters of the images, preliminary coordinates of the object are determined, the section of the territory is recognised and an integral cartographic reference image is formed. Integral information-bearing parameters of the compared images are obtained, while selecting multidimensional parameters among them and increasing the latter by a value defined by the required accuracy of estimating coordinates of the object. Change in shape, topology and colour of the section of the territory is determined.

EFFECT: broader functional capabilities owing to higher accuracy, faster determination of coordinates of movement of the object and noise-immunity.

20 cl, 21 dwg

Description

The group of inventions relates to a method for controlling a moving object and a device for its implementation. The scope covers automatic and automated control of the movement of various objects, such as transport robots, ships, aircraft, etc.

Various methods and devices for determining the coordinates of an object are known and widely used, for example, inertial, optical-television or radar navigation devices [1, 2].

An essential feature of these methods and devices is that the analysis of the information obtained is carried out according to certain methods, algorithms or theoretical models, and then a conclusion is made about the coordinates of the object. So, for example, the conclusion about the location of the object is made when analyzing the radar image of the area [3].

The disadvantages of the methods and devices are the low accuracy of determining the coordinates of the object and the inability to quickly adapt when changing the navigation task. In addition, these methods and devices have low functionality, since they perceive navigation information from a limited number of sensors (one, two and three).

As a prototype, we consider a method for determining the coordinates of an object in which the reference and current images are read (perceived), resulting in a pair of compared images [4]. A device that implements this method includes a true course cartographic index (CICC), a basic cartographic index of a given course (CGC), filters for filtering interference CQC and CGC, a correlation-extreme information processing unit (CECI) and filter control units. With the help of the KUIK, the current image is perceived, and with the help of the CKPS, the reference image is reproduced, which are compared in the BKEOI and the coordinates of the object's location are determined.

The disadvantages of the method and device are the low accuracy of determining the coordinates of the object due to the low noise immunity and noise immunity (limited by the ability to filter interference TI, EI and VKF), the inability to quickly adapt when changing the navigation task, low functionality due to the use of one image sensor (cartographic pointer true course), low speed due to the presence of electromechanical filtering units of the compared images and filter control.

The objective is to create a way to control a moving object and a device with improved functionality by increasing accuracy, speed of determining the coordinates of the movement of the object, noise immunity and noise immunity, and as a result, increasing the efficiency of the navigation task.

The problem is achieved in that, as in the known method in the claimed method of controlling a moving object, get cartographic reference and current images of the territory, compare the change parameters of the current cartographic image with a reference cartographic image, filter the image from interference

The difference between the proposed method and the known one is that before comparing the parameters for changing the images, preliminary coordinates of the object are determined, a site area is recognized, an integrated cartographic reference image is formed and integrated informative parameters of the compared images are obtained, highlighting the multidimensional ones and increasing the latter by an amount determined by the required estimation accuracy the coordinates of the object, determine the change in the geometry, topology and color of the area, while up to olnitelno suppress interference neural network filtering, and after comparing the image change parameter estimates the position of the object and produce corrected route of the object.

At the same time, an integral cartographic reference image is formed by mounting local areas of the territory at reference points using neural network technologies.

In addition, receive integrated informative parameters of the compared images when forming radar, television color or infrared images of the underlying surface of the Earth or images of a map of the starry sky.

Generate radar or television color images on topographic maps of the Earth.

Determine the change in the geometry, topology and color of the area for the compared pair of images while eliminating the inconsistencies of the images according to the angle of their relative rotation, scale, roll, pitch and color.

In addition, the geometrical parameters of local and integral areas of the territory are determined in the form of perimeters, areas, radii, length, width, the number of inflection points of the contour, the geometric center of image elements, constantly observed areas, contour sweep, re-calculation in the polar coordinate system, image inversion, image formation microplan.

In this case, the transition to the type of image is carried out on the basis of calculating the adaptive coefficient by comparing the current and limit values of the integral informative geometric, topological and color parameters of the images, as well as the amplitude, spectral, structural and differential informative parameters of the cross-correlation function.

In addition, they adjust the route of movement of the object by reconfiguring the operation of the device.

In addition, they recognize a site using neural network technologies.

It is advisable that multidimensional integrated informative parameters of the compared images are increased up to 1000 times.

In this case, the territory may be land, underwater, air territory.

The task is also achieved by the fact that, like the known claimed device for controlling a moving object, it contains a serially connected synchronizer, a block of correlation-extreme information processing, a main cartographic indicator of a given course, a block of additional filters of the main cartographic specified course, with an output connected to the second input of a correlation block -extreme information processing, the input through the filter control unit is connected to the output of the correlation-extreme block th information processing, the latter is connected to the second input of the additional filter unit COG pointer mapping, and filter control unit is connected to the second output of the additional filter unit COG mapping and pointer control object.

What is new is that it additionally contains, firstly, a series-connected image sensor connection unit, an adaptive neural network interference suppression filter and an intelligent image processing unit, with a separate output connected to the input of the image sensor connection unit, which are inserted between the output of the radar cartographic indicator of the true course and a separate input of the correlation-extreme information processing unit, a separate output is connected to an intelligent processing unit of secondly, an infrared image sensor, an astro sensor unit, a television image sensor, a satellite navigation system unit, a topographic map image building unit are connected by separate bidirectional connections to the image sensor connection unit, thirdly, an inertial navigation system and structurally connected tunable control unit for individual inputs connected to the outputs of the correlation-extreme information processing unit, satellite navigation unit system, a block for constructing images from a topographic map and a synchronizer, and with separate outputs connected to a control object and a block for constructing images from a topographic map, with a separate input of a structurally tunable control unit connected to the output of a control unit, fourthly, a block of neural network image recognition by separate inputs connected to separate outputs of the intelligent image processing unit and synchronizer, and a separate output connected to separate inputs of the block relational-extreme information processing.

In addition, the image sensor connection unit is made in the form of a multi-channel multiplexer.

In addition, the astro sensors block contains sensors that receive a map of the starry sky and radio galaxies.

In addition, the satellite navigation system unit contains GLONASS, GPS and / or GALILEO.

In addition, the inertial navigation system includes blocks of gyroscopes, accelerometers, control, monitoring and information processing.

In addition, the block for constructing images on a topographic map consists of topographic and topological maps of the earth’s surface, an image reader from maps, and an automaton on tunable computing environments.

In addition, the television image sensor is made in the form of a television camera operating in the infrared optical range, and contains an infrared spotlight.

In addition, the block of additional filters of the cartographic indicator of the true and given courses and the filter control block are made on tunable computing environments.

In addition, a structurally tunable control unit is implemented on tunable computing environments.

The proposed method and device is illustrated by the drawings shown in figures 1-21.

Figure 1 presents the structural diagram of the device.

Figure 2 presents the structural diagram of an intelligent image processing unit.

Figure 3 presents the structural diagram of the block correlation-extreme information processing.

Figure 4 presents the structural diagram of an adaptive neural network filter suppression interference.

Figure 5 presents a diagram of a formal neuron.

Figure 6 presents the neural network.

7 shows the patterns of image features used in the method (when using neural networks).

On Fig presents the structure of a multidimensional storage device of the main cartographic indicator of a given course.

Figure 9 presents the field of informative parameters (IP) of the main radar cartographic index of a given course.

Figure 10 presents the IP field of the main infrared cartographic indicator of a given course.

Figure 11 presents the IP field of the main television cartographic index of a given course.

On Fig presents the field IP of the main astronomical cartographic index of a given course.

On Fig presents the IP field of the main satellite map of the specified course.

On Fig presents field IP of the main topographic cartographic index of a given course.

On Fig presents a structural diagram of the main cartographic indicator of a given course.

On Fig presents a structural diagram of a structurally tunable control unit.

On Fig presents an embodiment of a block tunable computing environment.

On Fig presents a structural diagram of a unit for connecting an image sensor.

On Fig presents a new type of cross-correlation function (VKF) for various sizes of image elements and various types of images.

On Fig presents graphs of the dependencies of the accuracy characteristics of the device from the steepness of the VKF or from the geometric characteristics of the images.

On Fig presents a view of the VKF with various types of images.

Figure 1 gives the following notation:

1 - control object (OS);

2 - additional infrared image sensor (IDN);

3 - an additional block of astro sensors (BA);

4 - radar cartographic indicator of the true course (RKUIK);

5 - additional television cartographic indicator of the true course (OTDI);

6 - additional satellite cartographic indicator of the true course (BSNS);

7 - inertial navigation system (ANN);

8 - additional shaper of the cartographic indicator of the true course on the topographic map (block for constructing images on the topographic map) (BPITK);

9 - unit for connecting an image sensor (BPDI);

10 - intelligent image processing unit (OBOI);

11 - structurally tunable control unit (SPBU);

12 - synchronizer;

13 - filter control unit (FUF);

14 - block additional filters cartographic pointer true course (BDFKUIK);

15 is a block of additional filters of the main cartographic index of a given course (BDFOKUZK);

16 - the main cartographic index of a given course (OKUZK);

17 - block correlation-extreme information processing (BKEOI);

18 - adaptive neural network interference suppression filter (ANFPP);

19 is a block neural network image recognition (BNRI).

Determining the coordinates of an object using the proposed methods and devices is carried out when comparing images of various dimensions and chromaticities based on calculation estimates:

where R is a measure of the proximity of the current image (cartographic indicator of the true course) and the reference image (cartographic indicator of the given course);

F ₁ (λ), F ₂ (λ, ν) - functions that describe the cartographic indicators of the true and given courses, respectively;

λ, ν are respectively the constant and variable parts of the parameters.

First, we consider algorithms that can function with arbitrary images (with an arbitrary arrangement of TI and EI elements).

It is known [5] that if the angle of rotation between TI and EI is zero α = 0), then the coordinates of the main maximum of the CCF are determined by the following expressions:

,

- координаты центров тяжести совпадающих элементов ТИ, ЭИ соответственно; a, b - относительные смещения коррелируемых изображений.Where

,

- coordinates of the centers of gravity of the coincident elements TI, EI, respectively; a, b are the relative displacements of the correlated images.

If during angular alignment of images, the coordinate systems associated with the TI and the registration plane of the VKF are fixed, and the EI rotates around the optical axis of the correlator, then the coordinates of the main maximum of the VKF are determined approximately

To accurately determine the coordinates of the main maximum, the expression (3) should be transformed as follows:

Expressions (4) mean the re-calculation of coordinates associated with EI into coordinates associated with EI.

Obviously, when using algorithm (3), an error occurs in the calculation of linear coordinates

So if

so that δ ' _ξ = δ' _η = 0, you must first determine the approximate values of ξ ₁ , η ₁ , and then calculate the exact coordinates using formulas (4).

Using (3) and the methods of geometric optics, we can also write

It follows from (7) that if

then with the angular alignment of the images, the linear coordinates do not depend on the rotation angle of the TI and EI relative to the initial position and are characterized only. expressions (2).

Thus, when combining the compared images without changing the EI rotation, it is necessary either to refine the obtained approximate coordinates, or to simplify the process of calculating the coordinates, combine the center of rotation of the EI and the center of gravity of the image with the optical axis of the correlator.

Let us now consider the application of this algorithm in the case of using a differentiated method for determining coordinates, in which the signs of coordinates are calculated as follows:

where Z ₁ ÷ Z ₄ - signals characterizing the illumination of the corresponding quadrant of the correlation plane.

In expression (9), the sign Δ ₁ corresponds to the coordinate η, and the sign Δ ₂ corresponds to the coordinate ξ.

Expression (9) is valid if the center of rotation of the EI and the center of gravity of the image are aligned with the optical axis of the correlator, and the turn of the EI does not exceed the angle π / 2. When the EI is rotated by an angle greater than π / 2, the signs will already be:

As can be seen from the comparison of expressions (9), (10) when the EI is rotated by an angle exceeding π / 2, errors arise in the determination of linear coordinates

Expression (11) should be understood as follows: the magnitude of the error along the coordinate ξ corresponds to the offset Δη along the coordinate η, and the magnitude of the error along the coordinate η corresponds to the offset Δξ along the coordinate ξ.

To eliminate errors δξ _α ; δη _α . it is necessary, using a priori information about the reversal of the EI with respect to the correlation plane, to connect the signals Z ₁ ÷ Z ₄ taken from the output of the photodetector to the corresponding inputs of the linear coordinate estimation block.

When recognizing the color of the elements of the territory, the color of the zone “colored” in any tone or spectrum is determined based on the following rule:

where C _to , C _z , C _s , C _p , C _W , C _g , C _b - respectively, the levels of red, green, blue,

purple, yellow, blue, white,

J _R , J _G , J _B - respectively VKF, characterizing red, green and blue colors. Thus, by the combination of VKF J _R , J _G , J _B , any color of the zone can be determined

the investigated surface.

The terms of the VKF are described by the following expressions:

- функция описывающая цветное ТИ (в системе координат X₁, O₁, Y₁);Where

- a function describing the color TI (in the coordinate system X ₁ , O ₁ , Y ₁ );

- функции, описывающие ЭИ (в системах координат

,

,

), пропускающие соответственно красную R, зеленную G и синюю В компоненту цвета;

- functions that describe EI (in coordinate systems

,

,

), letting through respectively the red R, green G and blue B component of the color;

K is the coefficient of proportionality.

The geometric characteristics of the zones of the investigated surface are determined by the following well-known expressions.

where X _C , Y _C - coordinates of the center of gravity of the structural element of the image area;

ΔM _KX , ΔM _KY - elementary moments in the direction of the corresponding axes;

N _to - the number of points (pixel) in the circuit;

i is the circuit number;

S _SE - the area of the structural element of the image area, limited by the contour;

K _f - form factor;

P is the perimeter of the circuit;

P _i - inflection of the contour;

Nп - the number of inflections of the contour;

j is the number of the inflection of the i-th circuit.

The area S of the _{solar cell} and the perimeter P of the structural element are determined bypassing the zone contour. After obtaining the necessary information, the current and reference information (images) are compared in accordance with expression (1):

,

,

,

,

,

- матрицы информативных признаков (параметров) текущих изображений, формируемых в данный момент времени радиолокационным, инфракрасным, телевизионным, астрономическим, спутниковым и топографическим картографическим указателями истинного курса, соответственно;Where

,

,

,

,

,

- a matrix of informative features (parameters) of current images generated at a given time by radar, infrared, television, astronomical, satellite and topographic cartographic indicators of the true course, respectively;

- матрицы информативных признаков (параметров) эталонных изображений (запоминаемых заранее), формируемых радиолокационным, инфракрасным, телевизионным, астрономическим, спутниковым и топографическим картографическим указателями заданного курса, соответственно.

- a matrix of informative features (parameters) of reference images (stored in advance) generated by radar, infrared, television, astronomical, satellite and topographic cartographic indicators of a given course, respectively.

These matrices can be formed in the form of images (photographs) and mathematical models of territories.

The device operates in two modes: preparation of a standard and measurement (determination of coordinates). In the standard preparation mode, a reference image (EI) is formed that characterizes the required position of the OS 1 and provides the specified solution to the navigation problem. For this, all image sensors IDI 2, BA 3, RKUIK 4, OTDI 5, BSNS 6 and BPITK 8 are connected using the BPDI 9 by the signal from the synchronizer 12, the images are read, the images of the OBOI 10 are processed and recorded in the RAM 16, which is the recording unit (storage) EI. At this mode of operation of the device "preparation of the standard" ends.

In the "measurement (determination of coordinates)" mode, the device operates as follows. The signal from the synchronizer 12 includes BKEOI 17, SPBU 11, INS 7, IBOI 10 and BPDI 9, which connects the image sensors IDI 2, BA 3, RCUIK 4, OTDI 5, BSNS 6 and BPITK 8. An additional infrared image sensor IDI 2 senses thermal image of the Earth’s surface, and an additional television cartographic indicator of the true course of OTDI 5 - color images of the Earth’s surface. An additional block of astro sensors BA 3 forms an optical-television image of a map of the Starry sky or a map of the radio field of the Starry sky. RKUIK 4 reads the radar image (RLI) of the Earth's surface. BSNS 6 receives signals from navigation satellites and determines the preliminary (rough) coordinates of the control object OU 1, the proportional signals of which are sent to SPBU 11. In addition, signals from BSNS 6 are sent through BPDI 9 to the intellectual block of image processing of OBOI 10. Block BPITK 8 synthesizes TI, using the preliminary navigation coordinates of the control object OS 1, obtained from ANN 7 and BSNS 6. All signals describing the current images received by IDN 2, BA 3, RKUIK 4, OTDI 5, DSNS 6 and BPITK 8 are received through BPDI 9 and hell An ANPPP 18 active neural network filter for interference suppression (ANPPP 18 embodiment is shown in FIG. 4) into an intelligent image processing unit IBOI 10, which performs the following operations: converting the TI into contour or point images, electronically extracting the information part (s) of the TI, converting TI according to a given law, filtering TI from mismatched parts (interference). Thus, IBOI 10 submits to BNRI 19, BDFKUIK 14, BDFOKUZK 15 or OKUZK 16 a complete (not converted) image (the embodiment of IBOI 10 is shown in FIG. 2). After that, the neural network image recognition unit BNRI 19 is turned on, which compares (recognizes) the images of the territory (the embodiment is shown in FIGS. 5, 6). After this, BUF 13 includes BDFKUIK 14 and BDFOKUZK 15, which optically eliminate additional mismatched parts of the images on TPSU 16 and the cartographic indicator of the true course coming from IBOI 10. Further, the BKEOI 17 determines the coordinates (ξ, η) of the object as follows:

,

- фильтры, представляющие собой маску с прозрачной областью, равной размеру кадров КУИК и ОКУЗК;Where

,

- Filters, which are a mask with a transparent area equal to the size of the KUIK and OKUZK frames;

- фильтр, выполненный в виде маски, в которой изображения элементов прозрачны, по форме и расположению идентичны ЭИ, а границы элементов изображения на маске расширены по сравнению с ЭИ на величину, равную суммарной ошибке работы предлагаемого устройства;

- a filter made in the form of a mask in which the image of the elements are transparent, identical in shape and arrangement to the EI, and the boundaries of the image elements on the mask are expanded compared to the EI by an amount equal to the total error of the proposed device;

Z ₃ (ξ ', η') - filter, represents a diaphragm.

,

обеспечивают фильтрацию помех основного картографического указателя заданного курса 16, а фильтр Z₃(ξ',η') - фильтрацию побочных максимумов взаимно-корреляционной функции. Сигналы, пропорциональные величинам ξ', η', поступают от БКЭОИ 17 на структурно-перестраиваемый блок управления 11, который использует информацию с БСНС 6, ИНС 7 и БПИТК 8, управляет движением объекта управления 1. При изменении оперативной задачи на СПБУ 11 поступают соответствующие сигналы управления от задатчика управления. Информация с CПБУ 11 поступает также на БПИТК 8.Moreover, the filters

,

provide interference filtering of the main cartographic indicator of a given course 16, and filter Z ₃ (ξ ', η') - filtering side maxima of the cross-correlation function. Signals proportional to ξ ', η' are received from BKEOI 17 to a structurally tunable control unit 11, which uses information from BSNS 6, ANN 7 and BPITK 8, controls the movement of control object 1. When the operational task changes to SPBU 11, the corresponding control signals from the control unit. Information from SPBU 11 also arrives at BPITK 8.

This completes one cycle of the device. The next cycle starts again at the signal from the synchronizer 12.

Figure 2 presents the structural diagram of an intelligent image processing unit 10. IBOI 10 consists of the following blocks:

20 - block selection contours (BVK);

21 - block forming points (BFT);

22 - invert block (BI);

23 - block re-calculation in other coordinate systems (BPDSK);

24 is a block selection of constantly observed parts of the image (MNIS);

25 is a block filtering mismatched parts of the image (BFNCHI);

26 - microplan forming unit (BFM);

27 - block forming a scan circuit (BFRK);

28 - block calculating the coordinates of the center of gravity of the image (BVKTSTI);

OBOI 10 works as follows.

During the operation of IBC 10, image segmentation and coding algorithms are implemented. Segmentation is the process of dividing the image of the working scene into its component parts: objects, their fragments or characteristic features. Under coding is understood the compression of video information for storage in the memory of an OBOI [3]. At this stage, the following operating modes are possible:

1st mode - without image conversion - using ordinary (full) images;

2-nd mode - conversion of complete images to contour ones (forming contours of image elements or forming a scan of image contours);

3rd mode - conversion of full images to point-based ones (for example, by forming image points at the inflection points of the contour, by calculating the coordinates of the center of gravity of a part or the whole image);

4th mode - selection of the informative part (s) of the image (invert, selection of constantly-observable parts - selection of image elements);

5th mode - image conversion according to a given law (recalculation into other coordinate systems);

6th mode - filtering images from mismatched parts (extrapolating image bias, using the microplan mode of the image, using adaptive neural network filtering).

The choice of the operating mode of the OBI is based on some characteristics of the image and the VKF. In this case, the geometric characteristics of the elements of the complete images are used (S _GT is the area of the object, l is the length (or width) of the object, S _mu is the image characterizing the area of the object and the noise and frame (screen) of the image sensor (S _E is the area, l _E diameter (width and height)). To change the operating mode of the OBI (i.e. the image sensor), the following relationships are used:

Relations (17), easily implemented in software or hardware, should be used as criteria for switching some operating modes of the intelligent image processing unit 10. Moreover, the transition to another mode must be carried out if one (or several) of the following conditions is met:

wherein K * _s, K * _f, K * _s, ΔS * - critical (limit) values the corresponding values _{K s, K e, K '} s, ΔS, above which will cause a malfunction SEAL 17 (value K _s, K _e , K * _s , K * _e obtained experimentally).

It is clear that K _s , K _e should be used in the absence of interference on the transformer, and K ' _s , ΔS - in the presence of interference on the transformer. Situations are also possible when the operating modes of IBOI 10 need to be controlled using two criteria: one - in the absence of interference, the other - if any. Thus, according to the values of K _s , K _e , K ' _s , ΔS, a decision is made on the choice of any operating mode of OBOI 10.

We give a scheme for choosing the main modes (the decisive rule of the RP) of the operation of OBOI 10:

1st mode is used for K _s ≤K * _s , K _e ≤K * _e , K ' _s ≤K' * _s , ΔS≤ΔS *;

2nd mode is applied when K _s > K * _s , K _e > K * _e , K ' _s ≤K' * _s , ΔS≤ΔS *, while the interference should not distort the contour of the object;

3rd mode is used for K _s > K * _s , K _e > K * _e , K ' _s >K' * _s , ΔS> ΔS *;

The 4th mode is used when the use of the previous modes does not have a positive effect, in particular, if the elements of the TI fluctuate in brightness and coordinates, and the informative part of the TI is not perceived by the TI sensor;

5th mode is used if it is necessary to provide an estimate of the angular coordinate with high accuracy, without using the rotation of the EI;

6th mode is used if K _s > K * _s , ΔS> ΔS *. The critical values of K * _s , K * _e , K * _s , ΔS are selected on the basis of preliminary studies, taking into account the desired accuracy characteristics of the proposed device.

Information from IBOI 10 is fed to BPDI 9, which determines the adaptive coefficient K _a using expression (17), (18) and the decision rule of the RP. Based on this coefficient K _a , the structure of the compared images and the type of image sensor used are selected.

Figure 3 presents the structural diagram of the block correlation-extreme processing of information 17.

BKEOI 17 consists of the following blocks:

29 - shaper correlation function (CF);

30 - KF analyzer;

31 - classical correlation calculator (KB);

32 - HF with filtration;

33 - HF with the combination of TI and EI in angle (KVSTIIU);

34 - HF without combining TI and EI in angle (KVBSTIEIU);

35 - non-standard (dynamic) KB (BDKV);

36 - KB with a given arrangement of terms VKF (KVZRSVKF);

37 - KB with color decomposition (CMEC);

38 - classical correlation analyzer (KA);

39 - spacecraft with coordinate correction (AS);

40 - differentiated KA (DKA);

41 - differential spacecraft (RCA);

42 - spectral KA;

43 - structural spacecraft;

44 - quadratic KA (KKA);

45 - stereoscopic spacecraft;

46 - linear spacecraft (LCA);

47 - spacecraft with extrapolation (CAE);

48 - adaptive spacecraft;

49 - amplitude spacecraft.

BKEOI 17 works as follows. Shaper correlation functions (CF) implements the following methods of forming CF. The simplest classical correlation analyzer (KA) 38 calculates the VKF or convolution integral. The spacecraft in the classical version requires significant computing power to perform the operations of multiplication and integration for all shifts and turns of the TI and the reference image (EI). SC with filtering 32 provides filtering (spatial or otherwise) from interference TI, EI and VKF. A spacecraft with a combination of TI and EI at an angle of 33 in comparison with a spacecraft without an angular alignment of images 34 allows TI to be rotated to any angle and requires a circular rotation of the EI in angle relative to the TI. The standardless KB 35 synthesizes VKF using sequential rewriting of TI and EI and is based on the use of a dynamic correlator [5]. The spacecraft, which forms the terms of the VKF according to the given law 36, is characterized by the design of the correlator and a certain arrangement of the elements TI and EI. KB with color decomposition 37 provides the formation of VKF corresponding to the terms of the color of the images. The use of a specific shaper KF is determined by the task performed by the BKEOI and therefore the proposed device. The KF 30 analyzer operates as follows. Amplitude spacecraft 49 retrieves the necessary information based on an estimate of the amplitude of the main maximum of the VKF. This is the simplest spacecraft. A higher accuracy characteristics compared to the amplitude KA 49 provides a modified amplitude correlation calculation in which the necessary information is extracted based on the analysis of parameters (width, slope) VKF. Spacecraft with coordinate correction 39 allows continuous rotation of EI with angular alignment of images. In this case, the obtained rough estimate of the coordinates is refined. In the differential CA 40, the angular rotation of the EI and the correction of coordinates by commutation of the corresponding quadrants of the photodetector of the KF 30 analyzer are also carried out. The difference KA 41, based on the element-by-element calculation of the differences in the intensities of the TI and the EI, has the advantage over the classical KA 38, requiring the operation of multiplication. The spectral KA 42, which determines the angular coordinate, is based on the Fourier transform and calculates the difference between the spectra of TI and discretely deployed EI. Structural (syntactic or linguistic) CA 43 is based on three procedures: allocation of TI contours, selection of descriptors (lines of a certain shape, segments) and a description of their parameters, parsing using grammar (classification). When implementing the CA 43 algorithm, features are compared, not full images. This significantly reduces the required memory capacity of the proposed device and reduces the amount of computational operations.

Quadratic spacecraft 44 is implemented by squaring the correlation function, thereby increasing the difference between the main and secondary (secondary) maxima. This greatly increases the noise immunity of the proposed device.

Stereoscopic and linear modified KA 45, 46 are intended for comparing three-dimensional images (i.e. analysis of three-dimensional scenes). In this case, the stereoscopic spacecraft 45 is based on a correlation estimation of parallax, while the linear modified one is based on a correlation comparison of a combination of flat (two-dimensional) images of the territory in the working area.

A spacecraft with extrapolation 47 filters TI and EI by predicting TI bias. However, such a spacecraft 47 requires a priori information about the displacement of the object in the working area.

Adaptive spacecraft 48 is synthesized based on the spacecraft described above and adjusts the parameters of the proposed device (for example, identifying informative features of TI, adjusting the sensitivity of the KF analyzer, filtering TI, EI and VKF, etc.) in the presence of interference in the compared images and changing the situation in the working area. The choice of a specific algorithm is determined by the task performed by the device.

Figure 4 presents the structural diagram of an adaptive neural network interference suppression filter (ANFP) 18.

ANFPP solves the problem of extracting a useful signal distorted by broadband interference using a neural network, which acts here as a filter that suppresses broadband interference (noise). Consider the case where the useful signal is distorted by broadband interference, while there is no external reference signal containing the useful signal itself. In general, an adaptive interference suppression device cannot be used to reduce or eliminate this type of interference. However, upon receipt of the reference signal directly from the input signal after a predetermined delay, as shown in FIG. 4, broadband interference can be suppressed.

ANFPP consists of the following elements:

50, 51 adders;

52 - delay line;

53 - filter.

In Fig. 4, m [k] = m (t _k ), t _k = k * dt denotes a useful harmonic signal. By v [k] = v (t _k ) the broadband interference is indicated. The input signal is denoted by x [k] = x (t _k ), (x [k] = m [k] + v [k]), and the reference signal is denoted by x [k-dh] = x (tk-dh) . By y [k] = y (t _k ), the signal at the filter output is designated, and y [k] = M [k], where M [k] is the estimate of the signal m [k] at time t _k . The difference between x [k] and y [k] is denoted by e [k] = e (t _k ), and e [k] = x [k] -y [k]. Here, dt denotes the interval of measurement discreteness.

The neural network image recognition unit is based on the use of various types of neural networks [6, 7]. A neural network (NS) consists of elements called formal neurons that mimic the functioning of neurons in the cerebral cortex. Each neuron converts a set of signals entering its input into an output signal. An example of a formal neuron is shown in FIG. The structure of the neural network is shown in Fig.6.

A connection between individual neurons can exist, and it is communications between neurons, encoded by weight coefficients, that play a key role in the functioning of NS. The connections between the neurons in Fig.6 are indicated by the Latin letter W. The index in the upper part indicates the affiliation of the weight coefficient to the layer. One of the advantages of the NS is the possibility of parallel functioning of all elements, which significantly increases the efficiency of solving the problem as a whole. This feature of NS has been successfully used in pattern recognition systems. A neural network has inputs X and outputs Y, representing a system that generates an output state depending on the input. The presence of weighting coefficients that can be determined algorithmically allows us to give the NS the most important property - the ability to learn. Today, there are a huge number of weight adjustment algorithms that successfully cope with the task of training a neural network.

.Consider an algorithm for recognizing images of a territory using a neural network. Suppose that a photograph (reference image) presents an image of the image and the recognition system should identify it. For proper identification, the system must be trained in this. Since a digital photograph is a matrix of intensity values, for example, for a black-and-white photograph, the gray scale has a length from 0 (black) to 255 (white). Thus, the value of each pixel of a digital photograph can be fed to the input of a neural network in the form of an information matrix of vectors

.

Suppose the system is trained to recognize 10 images of a territory. Therefore, the output layer of the neural network must contain 10 neurons, each of which is “trained to recognize” the corresponding territory. If at the output of a neuron the signal is maximum in relation to others, then this neuron is the “winner” of recognition. Since it is associated with the image of a certain territory from this ten, the territory to which this neuron corresponds will be identified.

Note that when applying all the described methods for image recognition of three-dimensional objects, difficulties arise associated with spatial rotations and changes in lighting conditions. Images for different angles of rotation of the object can vary significantly, and some of the information in the image is lost. The difficulties associated with intraclass variations are significant. For territories, such variations are changes in the color, lighting, and geometric characteristics of the territory.

When recognizing images, a person unconsciously attracts a huge supply of contextual knowledge, which he accumulates throughout his life. The neural network method implemented in the Neurociber navigation computer program of the trained neural network, potentially allows you to simulate the processes occurring during pattern recognition in the human brain, which in a first approximation can be represented as a neural network of great complexity.

Consider the results of experimental studies of the Neurociber program. After receiving the current image frame, it is subsequently loaded into the buffer of the search module of the anomalous area on the territory image (or search for another image).

The module for searching for an abnormal area in the image of the territory is designed to detect areas containing these objects or with suspicion of these changes in the image, for subsequent transmission of the contents of these areas to the module for recognizing changes based on neural networks. The abnormal area search module on the territory image implements an algorithm for detecting objects using cascades of classifiers working with similar features.

The algorithm uses a cascade of classifiers stored in memory, created as a result of “training” the cascade on an array of “positive” images (containing change data) and “negative” images (images of the same size that do not contain changes in areas of the territory).

After loading the cascade, it can be applied to the plot in the image. The result of the operation is the value "1" if the site contains a similar to an abnormal site in the image of the territory, and the value "0" otherwise. When analyzing the full image, the cascade is scaled several times and “moves” throughout the image to detect abnormal areas in the image of the territory located in different parts of the image.

The word “cascade of classifiers” in the name of the algorithm means that the object under this name contains several levels of classifiers, which are applied sequentially, level by level, to the studied image section, until at one of the levels the image section is recognized as “negative”, or all levels will be passed and the image section will be considered “positive”. Haar-like features are inputs of individual classifiers and are calculated as follows (see Fig. 7).

The feature used in a particular classifier is determined by its position inside the considered image area, scale and pattern (Fig. 7).

For example, in the case of the third linear feature (Fig. 7, template 2c), the response is calculated as the difference between the sum of the pixels of the image under the rectangle that includes the entire feature (two white stripes and a black strip in the middle) and the sum of the pixels of the image under the black strip multiplied by 3 to compensate for the difference in the size of the areas.

Based on the totality of these features, the search module for an anomalous area in the image of the territory detects all areas containing abnormal areas in the image of the area, cuts them out and passes it to the recognition module for recognizing the anomalous area in the image of the area based on neural networks.

The module for recognizing an abnormal area in the image of a territory based on neural networks, when loading, reads a database of a neural network trained to recognize this change in computer memory.

When the search module for the anomalous area in the territory image transmits an image area that contains the anomalous area, the neural network recognition module supplies the pixels of this image presented in the range [-1; +1], to the input of a trained neural network. As a result, at its output, the value +1 appears on the neuron corresponding to one of the type of image change (or type of image) stored in the database. The remaining neurons in this case should have a value of -1. Otherwise, due to the influence of interference during recognition (for example, low or uneven illumination, a large angle of rotation of the vehicle, etc.), the values differ from 1. If the difference exceeds the threshold set during compilation, the object is considered unrecognized and the recognition module of the anomalous section is the image of the territory based on neural networks returns the string "unknown section". In case of successful recognition, the recognition module of the anomalous area on the image of the territory based on neural networks searches the database associated with the recognized change in the line (description of this change) and returns it as the result of recognition. The Neurociber program identifies an identified abnormal area in the image of the territory.

Since the problem solved by the neural network is classification, then one output is enough to solve this problem of the neural network. The output value of the neural network is in the range [-1; 1], which accordingly means the absence or presence of the investigated object in the classified image.

The input layer of 28 × 32 neurons serves to feed the input image into the neural network. Next to the input layer is a convolutional layer C1, which consists of 5 planes and coagulates the input image using a 5 × 5 synaptic mask.

The size of the convolution plane is determined in accordance with the following expression:

.

.

h _c = h _u -K + 1,

where w _u , h _u are the width and height of the convolution plane, respectively;

w _c , h _c - width and height of the plane of the next layer;

K is the width (height) of the scan window.

The plane size of the convolutional layer C1 is 24 × 28 neurons.

Receptive areas of neurons intersect. Neurons extract the same features of the input image regardless of their exact location. The subsampling layer S1 following layer C1 consists of 5 characteristic maps and provides local averaging and subsampling. After the subsampling operation, the exact location and specific features of each feature become less important, which gives a rather large degree of invariance of this network. Each plane of layer S1 is associated with only one plane of layer C1. The size of each plane of the S1 layer is 12 × 14 neurons, which is half that of the plane of the previous layer. The convolutional layer C2 consists of 20 planes 8 × 10 neurons in size, layers S1 and C2 are cross-linked. Different feature maps are needed to extract different characteristics because they receive different sets of inputs. Thus, the network is added the ability to combine different kinds of features to make up new, less dependent on the distortion of the input image. Layer S2 consists of 20 planes, each on one plane of layer C2, the size of each plane is 4x5 neurons. Layers N1 and N2 contain simple sigmoid neurons. The role of these layers is to ensure classification after the extraction of features and reduction of the dimension of the input are performed. There are 20 neurons in the N1 layer (one for each plane of the S2 layer), each neuron is completely connected to each neuron of only one plane of the S2 layer, it passes the result through the activation function. The only neuron of the N2 layer is completely connected with all the neurons of the N1 layer. The role of this neuron is to calculate the final classification result. The output of this neuron is used to classify the input image into a given territory and an arbitrary (random) territory.

All teaching methods are based on a single principle - error minimization. In multilayer networks, the optimal values of neurons of all layers except the last layer, as a rule, are not known. Therefore, such networks can no longer be trained, guided only by the magnitude of the errors at the outputs of the neural network. The main idea of the used method of back propagation of the error is that the error signal propagates from the outputs of the network to its inputs, in the direction opposite to the direction of propagation of signals in normal operation.

To test the program, 150 preset and 50 arbitrary images of the territory were selected. The testing methodology was as follows. When the first image was fed into the neural network, weights were randomly generated. For the remaining images, the weights were adjusted sequentially based on previously calculated values. For arbitrary images, the initial value of the weight coefficients was taken from the calculation of the given images. Then the weights were subsequently adjusted by image processing.

The test results — the responses of the program — are given in Tables 1 and 2. Thus, to solve the problem of image classification, artificial neural networks can be effectively used, namely convolutional neural networks, which are resistant to changes in scale, displacements, rotations, change of angle and other distortions , which is very important when processing images.

Tests have shown that the Neurociber program is able to memorize in real time using the built-in tools to add new anomalous areas to the database image of the area and then recognize the anomalous areas on the image of the territory under fixed lighting conditions.

Table 1 The results of testing a neural network on given images of the territory No. Network response No. Network response No. Network response No. Network response No. Network response one 0,992 31 0,999 61 0.989 91 0,999 121 0,999 2 0,999 32 0,999 62 0,999 92 0.979 122 1,000 3 1,000 33 0.980 63 0,999 93 0.989 123 , 1,000 four 0,999 34 1,000 64 0,999 94 0,995 124 0,999 5 0,999 35 0,999 65 0,999 95 1,000 125 0,998 6 0,999 36 1,000 66 1,000 96 0,999 126 0,996 7 0,999 37 0,993 67 1,000 97 0.974 127 0,999 8 0.986 38 1,000 68 0,999 98 0,999 128 1,000 9 0,999 39 0,995 69 0,999 99 1,000 129 1,000 10 1,000 40 1,000 70 1,000 one hundred 1,000 130 0,999 eleven 0,999 41 0,999 71 0.972 101 1,000 131 1,000 12 1,000 42 0,999 72 0,999 102 1,000 132 1,000 13 0,999 43 1,000 73 0.988 103 0,999 133 0.955 fourteen 0.988 44 0,999 74 1,000 104 1,000 134 0,999 fifteen 0,999 45 1,000 75 1,000 105 0.959 135 0,999 16 0,994 46 0,999 76 0,991 106 0,999 136 1,000 17 1,000 47 0.982 77 1,000 107 1,000 137 1,000 eighteen 0,999 48 0,998 78 1,000 108 0,996 138 1,000 19 0,999 49 1,000 79 0,999 109 0,993 139 1,000 twenty 1,000 fifty 1,000 80 0,999 110 1,000 140 0,999 21 1,000 51 0,999 81 1,000 111 1,000 141 0,999 22 0.955 52 0,998 82 1,000 112 1,000 142 0,999 23 1,000 53 1,000 83 1,000 113 1,000 143 0,999 24 0.960 54 0.987 84 1,000 114 0,999 144 0.969 25 1,000 55 0,999 85 0.962 115 0,999 145 0,999 26 1,000 56 0,999 86 0.981 116 0,995 146 1,000 27 0,994 57 1,000 87 1,000 117 1,000 147 1,000 28 0,999 58 0,996 88 1,000 118 0,998 148 1,000 29th 1,000 59 0,999 89 0,999 119 1,000 149 1,000 thirty 1,000 60 1,000 90 0,999 120 0,999 150 1,000

table 2 Test results of a neural network in arbitrary images of the territory No. Network response No. Network response No. Network response No. Network response one -0.999999 fourteen -0.999984 27 -0.999999 40 -0.999999 2 -0.999999 fifteen -0.999999 28 -0.999999 41 -0.999999 3 -0.999999 16 -0.999999 29th -0.999999 42 -0.999999 four -0.999999 17 -0.999999 thirty -0.999999 43 -0.999999 5 -0.999999 eighteen -0.999999 31 -0.999999 44 -0.999999 6 -0.999999 19 -0.999999 32 -0.999999 45 -0.999981 7 -0.999999 twenty -0.999999 33 -0.999999 46 -0.999999 8 -0.999999 21 -0.999999 34 -0.999999 47 -0.999999 9 -0.999999 22 -0.999999 35 -0.999999 48 -0.999999 10 -0.999999 23 -0.999983 36 -0.999999 49 -0.999999 eleven -0.999999 24 -0.999999 37 -0.999999 fifty -0.999999 12 -0.999983 25 -0.999999 38 -0.999999 13 -0.999999 26 -0.999999 39 -0.999999

On Fig presents the structure of a multidimensional storage device of the main cartographic index of a given course, which includes the following elements:

55 - main radar cartographic index of a given course (ORKUZK);

56 - the main infrared cartographic indicator of a given course (OIKUZK);

57 - the main television cartographic index of a given course (HOSE);

58 - the main astronomical index of a given course (OAKUZK);

59 - the main satellite map index of a given course (OSKUZK);

60 - the main topographic cartographic index of a given course (HOLES);

61-66 - reference marks (beginning of the first line of the image frame); respectively, ORKUZK (55), OIKUZK (56), HIGH SCHOOL (57), OAHK (58), OSKUZK (59), HOLIDAY (60);

67-72 - reference marks (end of the first line of the image frame); respectively, ORKUZK (55), OIKUZK (56), HIGH SCHOOL (57), OAHK (58), OSKUZK (59), HOLIDAY (60);

73-78 - reference marks (the beginning of the last line of the image frame), respectively, ORKUZK (55), OIKUZK (56), HOLIDAY (57), OAKZK (58), OSKUZK (59). HOLIDAY (60);

79-84 - reference marks (end of the last line of the image frame), respectively, ORKUZK (55), OIKUZK (56), OTKUZK (57), OAKZK (58), OSKUZK (59), OTKUZK (60).

A multidimensional main cartographic indicator of a given course is a multilayer digital model or photograph of an image.

Figure 9 presents the field of informative parameters (IP) of the main radar cartographic index of a given course (RE). The following notation is introduced here:

- координаты центра «тяжести» элемента изображения в системе координат XOY;

- coordinates of the center of gravity of the image element in the XOY coordinate system;

- коэффициент формы;

- form factor;

- число перегибов контура элемента изображения;

- the number of bends in the contour of the image element;

- площадь элемента изображения;

- the area of the image element;

- периметр элемента изображения;

- the perimeter of the image element;

a - image elements;

i, j - pixel coordinates in the XOY coordinate system.

Figure 10 shows the IP field of the main infrared cartographic index of a given course (FROM). Here, all the symbols correspond to the symbols presented in Fig.9.

Figure 11 presents the IP field of the main television cartographic index of a given course. The following notation is introduced here:

,

,

,

,

,

,

- соответственно, уровни красного, зеленого, синего, пурпурного, желтого, голубого и белого цветов;

,

,

,

,

,

,

- respectively, the levels of red, green, blue, magenta, yellow, cyan and white;

,

,

- соответственно, длина, ширина и ориентация элемента изображения (угол между продольной осью элемента и осью системы координат изображения);

,

,

- accordingly, the length, width and orientation of the image element (the angle between the longitudinal axis of the element and the axis of the image coordinate system);

,

,

,

,

- значения обозначений соответствуют обозначениям, представленным на фиг.9.

,

,

,

,

- the values of the symbols correspond to the symbols presented in Fig.9.

12 shows the IP field of the main astronomical cartographic index of a given course (A3). The following notation is introduced here:

,

,

- соответственно, координаты центра «тяжести» элементов (совокупности изображений звезд) ОАКУЗК;

,

,

- accordingly, the coordinates of the center of gravity of the elements (the totality of the images of the stars);

- угол между продольной осью совокупности ориентиров (звезд) и какой-либо осью системы координат XOY;

- the angle between the longitudinal axis of the set of landmarks (stars) and any axis of the XOY coordinate system;

- координаты элементов изображения, преобразованные в полярную систему координат;

- coordinates of image elements converted to a polar coordinate system;

- количество звезд;

- number of stars;

,

- соответственно, минимальный и максимальный радиус-векторы локальных (или интегральных) участков изображений;

,

- respectively, the minimum and maximum radius vectors of local (or integral) image sections;

and - image elements.

On Fig presents the field IP of the main satellite map of the specified course (SZ). The following notation is introduced here:

r ^ij , ν ^ij - respectively, radius (orbits), angular velocity and satellite acceleration;

,

- координаты центра «тяжести» локального участка (интегральной площади) изображения;

,

- coordinates of the center of gravity of the local area (integral area) of the image;

α ^ij is the angle between the longitudinal axis of the local section or the integral image area;

and - image elements.

On Fig given IP field of the main topographic cartographic index of a given course (TPZ). The following notation is introduced here:

- высоты соответствующих уровней z трехмерного изображения территорий;

- heights of the corresponding levels z of a three-dimensional image of territories;

the remaining values of the symbols correspond to the symbols shown in Fig.9, 10, 11, 12.

The main meanings of the designations shown in Figs. 9-14 are determined by expressions (2) - (14).

Multidimensional storage device of the main cartographic index of a given course works as follows. In the recording (preparation) mode of the reference image, the reference image is recorded in the cartographic pointers of the given course 55-60. At the same time, reference marks (optical or electronic) are placed at points 61-84, with the help of which the correct "stitching" of image frames during installation of the integrated main cartographic index of a given course is ensured. Stitching occurs according to the rule: the end of the first line of the i-th image frame - the beginning of the first line of the (i + n) -th image frame; the end of the last line of the 1st image frame is the beginning of the last line of the (i + n) -th image frame.

On Fig presents a structural diagram of the main cartographic index of a given course (OKUZK) 16, which includes the following products:

85 - multidimensional storage device (MZU85);

86 - control unit write-read (BUZS86);

87 - block offset, reversal and change the image frame (BSRSKI87).

OKUZK 16 works as follows. The signals from the correlation-extreme information processing unit 17 record the corresponding image from the intelligent image processing unit 10 through the image recording unit 86 into a multidimensional storage device 85; this image then from the multidimensional storage device enters the block of additional filters of the main cartographic index of the specified course 14.

On Fig presents a structural diagram of a structurally tunable control unit (SPBU) 11.

SPBU consists of the following blocks:

88 - block functional converter of the input signals (BFPvh);

89 - block tunable computing environment (BPVS);

90 - tuning unit (BN);

91 - block functional converter output signals (BFPvy). A key component of the control unit under consideration is BPSA 89. By BPSA 89 we mean a set (Z ^k , V, Q), where Z ^k is the set of k-dimensional vectors with integer coordinates that specify the nodes of a k-dimensional integer lattice; V = {v ₁ , v ₂ , ..., v _{m + p} } is a tunable automaton placed in each node of the lattice with n + p input channels, m + p output channels (m main, p side channels).

Here

,

,

,

t=log₂R;where σ _ij ∈ {0,1} (j = 1, ..., t),

,

t = log ₂ R;

- i-я функция алгебры логики (i=1, 2,…, R), реализуемая на j-м выходе автомата (j=1, 2,…, m+p);

- the i-th function of the algebra of logic (i = 1, 2, ..., R), implemented at the j-th output of the automaton (j = 1, 2, ..., m + p);

x ₁ , ..., x _{(n + p)} - information inputs;

z ₁ , ..., z _t - additional inputs used to configure the machine to the desired structure;

Q = {υ ₁ , υ ₂ , ..., υ _p }, where υ _i ∈ Z ^k , i = 1, ..., p is a neighborhood pattern: each vector υ ∈ V for an automaton with coordinates υ defines an automaton with coordinates υ + υ ₁ , with the i-th side output channel of which is connected the i-th side input of the machine in the node υ, i = 1, ..., p.

By a tunable automaton we mean an automaton for which a set of automaton mappings implemented by it is given and an algorithm for tuning to implement each of these automaton mappings is defined. A tunable automaton can be represented as a set of automata with the same outputs, and the setting determines the automaton whose outputs are considered the outputs of the entire tunable automaton at this setting.

BFPvh 88 converts the input information into binary form; BN 90 on the basis of the input information generates codes for setting up the computing environment for the algorithm of work necessary at the current time; BFPvy 91 converts a binary signal from the computing environment into a signal convenient for perceiving and processing the received information in other blocks.

BPVS 89 is an m × n matrix formed by a team of calculators based on tunable automata. In Fig.17 shows an example of BPS 89 for (m = 6 is the number of rows, n = 7 is the number of columns) and for a tunable automaton described by a system of formulas:

Here f ₁ , f ₂ are the outputs of the automaton;

y ₁ , y ₂ - information inputs;

x is the input;

z ₁ , z ₂ , z ₃ , - settings inputs.

On Fig for each tunable automaton in the lower left corner its code for setting the current automatic display is shown, so for the automaton with coordinates (1, 1) the code 001 is shown, which corresponds to z ₁ = 0, z ₂ = 0, z ₃ = one.

The fixed four inputs of the tunable computing environment are supplied with values (h _k , k = 1, 2, ..., 4). For the settings presented in Fig.17, and at the outputs f ₁ and f ₂ computing environment functions are implemented:

and when you change the setup code of one of the machines with coordinates (2, 7) in the matrix of the computing environment on the code z ₁ = 0, z ₂ = 0, z ₃ = 1 (Fig.17, b) at the outputs f ₁ and f ₂ computing environment implemented functions:

On Fig given a structural diagram of a unit for connecting an image sensor (BPI) 9, which operates as follows. The operation of block 9 begins with a signal from the synchronizer 12. The signals are transmitted to the adder of the structural parameters of the images from the cartographic pointers through a multiplexer 96 from the intelligent image processing unit 10. These signals characterize the type of image structure (ordinary, contour, point, etc.) and are calculated by formulas (12) - (14), (24) are fed to the first input of the comparison circuit 93, the second input of which receives signals from the adder of parameters (steepness) of the cross-correlation function 95. Information to the adder 95 come from IBOI 10. The third input of the comparator circuit 93 receives the reference signals _ie K, K _fe reference parameters from a memory 94. The comparison circuit 93 operates according to the rule:

where K _a is the adaptive integral coefficient;

K _i , K _i - coefficients characterizing the current and reference informative geometric parameters of the images;

K _F , K _Fe - coefficients characterizing the current and reference steepness of the VKF, respectively;

K _C , K _Tse - coefficients characterizing the current and reference color images.

When condition (24) is fulfilled, the required sensor (or sensors) of the cartographic indicator of the true course is turned on via the multiplexer 96.

Other blocks (astro sensors block 3, RKUMK 4, OTDI 5, BSNS 6, BINS 7, BPITK 8, IDI 2, BPDI 9, SPBU 11, BDFKUIK 14, BDOKUZK 15, OKUZK 16) were made in accordance with the recommendations set forth in [1 -5, 8].

Compared with the known, the proposed method and device have improved functionality by increasing the accuracy, speed of determining the coordinates of the movement of the object, noise immunity and noise immunity.

We show this by example. High accuracy characteristics of the proposed device are ensured, firstly, by introducing additional image sensors (additional cartographic indicators of the true course), secondly, by introducing a neural network interference suppression filter, and thirdly, by efficiently converting images (highlighting informative features) fourthly, by introducing a neural network image recognition unit, fifthly, using a structurally tunable control unit, sixthly, due to the combination of tions compared images in color, scale, roll, pitch and angular misalignment.

The introduction of additional cartographic indicators of the true course will ensure the connection, if necessary, of the required sensor, which will form the most reliable (integral) image of the territory (radar, television, infrared, optical, radio engineering, gravity, topographic). Using an additional inertial cartographic index 7 provides the provision of preliminary navigation parameters; this makes it possible to withdraw the object to a given area area. Moreover, the adaptive neural network interference suppression filter will provide effective filtering of the useful signal from interference coming from the image connection unit 10. This will provide with high accuracy the location (determination of coordinates) of the object.

The neural network image recognition unit 19 will recognize the area with high accuracy (probability) even if there is a significant level of interference; this will eliminate a malfunction in the block of correlation processing of information 17, which then will carry out a more accurate location of the object, compared with the known method and device.

The accuracy of positioning is also improved by combining the compared images in color, scale, roll, pitch and angular mismatch. That is, the mismatch of the cartographic index of a given course relative to the true course is eliminated

where C, H, β, γ, α are the values of color, scale, roll, pitch and angular mismatch of the compared images, respectively.

The use of a structurally tunable control unit 11 will provide precise control of the movement of the object (for example, precise control of movement along a given route) by structural and parametric optimization (i.e., optimal change in its structure and selection of its parameters).

The expansion of functionality is carried out by, firstly, obtaining additional information from DIKUIK 2, DAKUIK 3, DTKUIK 5, DSKUIK 6, DIKUIK 7, DFKUIKTK 8 about the territory in which it is necessary to make an exact location, and secondly, the introduction of BNRI 19, providing recognition of a wide class of images (compared with the known method and device), thirdly, the introduction of SPBU 11, which implements control of the movement of the object when changing the operational navigation task.

We present the results of experimental studies of the proposed device.

On Fig, a presents a section of the VKF J (ξ) along the Oξ axis (in meters) when combining ordinary (not converted) images of large-scale elements (curve 1), elements of the rod type (curve 2) and small-scale elements (curve 3). On Fig, b given the section of the VKF for ordinary (curve 1), contour (curve 2) and point (curve 3) images. From the above graphs it follows that, firstly, the finer the structure of TI and EI (the smaller the image elements), the more steeply the VKF is formed; secondly, point images form VKF with greater steepness in comparison with contour or ordinary images.

On Fig, a shows graphs of the dependences of the random component of the error σ (in% of the allowable displacement and turn) to determine the angular (curve 1) and linear (curve 2) coordinates of the object from the steepness K _{F of the} main maximum of the VKF. Here, the dotted lines indicate the regions K _F for ordinary O, contour K, and point T images. As can be seen from the graphs, it is more advisable to use contour or point images for which the steepness K _{F is the} largest and the error σ is the smallest.

The criterion for the transition to contour or point images can be the coefficient K _F , which characterizes the steepness of the VKF (at the level of 0.7 of the maximum amplitude value), and the coefficient K _S = S _g / S _e (where S _g is the image area of the object on the map pointer screen true course, S _e is the screen area of the cartographic indicator of the true course) or the coefficient K _e = L _g / L _e (where L _{g is the} size (length) of the image of the elements on the screen of the cartographic pointer of the true course, L _e is the size (diameter) of the cartographic screen True Chicken Pointer a). On Fig, b shows the graphs of the dependences of the error σ of determining the coordinates on the values of K _S and K _e . Here, curves 1 and 1a characterize the angular coordinate, and curves 2, 2 a the linear coordinate (solid lines are the graphs of the dependences of σ on K _S , and dashed lines are the graphs of the dependences of σ on K _e ). As can be seen from the graphs obtained on the basis of experimental studies, for K _S <0.19 or K _e <0.35 (for this algorithm of operation of the VKF analyzer that implements the blind search method), it is necessary to convert images to contour or point (for example, when K _S = 0.21 device did not work).

Of particular interest is the use of images in the device when the TI is presented in the usual form, and the EI is in the contour form (this is possible if the TI indicator is implemented in a simplified version, i.e. without processing the TI, and to improve the accuracy characteristics, the EI is prepared in the contour form). Figure 21 shows the cross sections of the VKF J (η) confirming the possibility of using such images (here: 1-VKF of conventional TI and EI; 2-VKF of contour TI and EI; 3-VKF of conventional TI and contour EI). As you can see, curve 3 has a large slope in the region of the main maximum in comparison with curve 1, in addition, the slope of curve 3 is slightly different from the slope of curve 2.

Information sources

1. Anuchin O.N., Komarova I.E., Porfilyev L.F. Airborne navigation and orientation systems for artificial Earth satellites. - St. Petersburg: Publishing House of the SSC RF Central Research Institute "Elektropribor", 2004. S.80-120.

2. XVI St. Petersburg International Conference on Integrated Navigation Systems. Collection of materials. - St. Petersburg: Publishing House of the Central Research Institute "Elektribribor", 2009. - 302 p.

3. Syryamkin V.I. and other systems of technical vision. Directory. Tomsk: MGP "RASCO", 1992. 367 p.

4. Angelov M.P., Katyshev V.A., Reznik V.G., Syryamkin V.I. A device for determining the coordinates of a moving object. USSR copyright certificate No. 592229.

5. Beloglazov I.N., Tarasenko V.P. Correlation-extreme systems. M .: Sov.radio, 1974.- 392 p.

6. Golomb L.A., Lawrence D.T. and Sejnowski T. J, SexNet: A neural network identifies sex from human faces // Advances in Neural Information Processing Systems. Morgan Kaufmann Publishers. San Mateo. USA 1991. P.77-83.

7. Wassermen F. Neurocomputer technology: Theory and practice / Per. in Russian lang Yu.A. Zuev, V.A. Tocheno. 1992 - 184 p.

8. Information portal DailyTechInfo-http: //www.dailytechinfo.org/space/339-nasa-publikuet-samuyu-polnuyu-topologicheskuyu.html.

Claims (20)

1. A method of controlling a moving object, which consists in obtaining cartographic reference and current images of the territory, comparing the change parameters of the current cartographic image with a reference cartographic image, filtering the image from interference, characterized in that prior to comparing the image changing parameters, preliminary coordinates of the object are determined, they recognize a site of the territory, form an integrated cartographic reference image and receive integral informative the parameters of the compared images, distinguishing among them multidimensional and increasing the latter by an amount determined by the required accuracy of estimating the coordinates of the object, determine the change in the geometry, topology and color of the area, while additionally suppressing noise by neural network filtering, and after comparing the parameters of the image changes, the location of the object is evaluated and produced correction of the route of the object. 2. The method according to claim 1, characterized in that they form an integrated cartographic reference image by mounting local sections of the territory at the reference points using neural network technologies. 3. The method according to claim 1, characterized in that the integrated informative parameters of the compared images are obtained when forming radar, television color or infrared images of the underlying surface of the Earth or images of a map of the starry sky. 4. The method according to claim 3, characterized in that they form a radar or television color image on topographic maps of the Earth. 5. The method according to claim 1, characterized in that they determine the change in the geometry, topology and color of the area for the compared pair of images while eliminating the mismatches of the images according to their relative rotation angle, scale, roll, pitch and color. 6. The method according to claim 5, characterized in that the geometric parameters of the local and integral sections of the territory are determined in the form of perimeters, areas, radii, length, width, the number of inflection points of the contour, the geometric center of image elements, constantly observed sections, sweep contour, recalculation into the polar coordinate system, inverting the image, forming a microplan. 7. The method according to claim 6, characterized in that the transition to the type of image is carried out on the basis of calculating an adaptive coefficient by comparing the current and limit values of the integral informative geometric, topological and color parameters of the images, as well as the amplitude, spectral, structural and differential informative parameters, mutually correlation function. 8. The method according to claim 1, characterized in that they adjust the route of movement of the object by reconfiguring the operation of the device. 9. The method according to claim 1, characterized in that the area is recognized using neural network technologies. 10. The method according to claim 1, characterized in that the multidimensional integrated informative parameters of the compared images are increased up to 1000 times. 11. The method according to claim 1, characterized in that the territory may be land, underwater, air territory. 12. A moving object control device comprising a synchronizer, a correlation-extreme information processing unit, a basic cartographic index of a given course, an additional filter block of a basic cartographic given course, an output connected to a second input of a correlation-extreme information processing unit, an input through a filter control unit connected to the output of the block of correlation-extreme information processing, the latter by the second input is connected to the block to additional filters of the true-course cartographic index, and the second output filter control unit is connected to the additional true-course cartographic filters block, and a control object, characterized in that it additionally contains, firstly, an image sensor connection block, an adaptive neural network filter interference and an intelligent image processing unit, a separate output is connected to the input of the image sensor connection unit, which are entered between at the output of the radar cartographic indicator of the true course and a separate input of the correlation-extreme processing unit of information, a separate output is connected to the intelligent image processing unit, secondly, an additional infrared image sensor, an additional unit of astro sensors, an additional television cartographic indicator of the true course, an additional shaper of the cartographic pointer true course on topographic map connected by separate bidirectional by connections to the image sensor connection unit, thirdly, an additional inertial cartographic indicator of the true course and a structurally tunable control unit for individual inputs connected in series to the outputs of the correlation-extreme information processing unit, an additional satellite cartographic indicator of the true course, and an additional shaper of the true course cartographic indicator by topographic map and synchronizer, and connected to separate outputs by separate outputs control object and an additional shaper of the true course heading on the topographic map, with a separate input of a structurally tunable control unit connected to the output of the motion encoder, fourthly, the neural network image recognition unit with individual inputs is connected to separate outputs of the intelligent image processing unit and synchronizer, and to separate the output is connected to a separate input of the block of correlation-extreme information processing. 13. The device according to p. 12, characterized in that the unit for connecting the image sensor is made in the form of a multi-channel multiplexer. 14. The device according to p. 12, characterized in that the additional unit of astro sensors contains sensors that receive a map of the starry sky and radio galaxies. 15. The device according to p. 12, characterized in that the additional satellite cartographic indicator of the true course contains GLONASS, GPS and / or GALILEO. 16. The device according to p. 12, characterized in that the additional inertial cartographic indicator of the true course includes blocks of gyroscopes, accelerometers, control, monitoring and information processing. 17. The device according to p. 12, characterized in that the additional shaper of the cartographic indicator of the true course on the topographic map consists of topographic and topological maps of the earth’s surface, an image reader from maps and an automaton on tunable computing environments. 18. The device according to p. 12, characterized in that the additional television cartographic indicator of the true course is made in the form of a television camera that operates in the infrared optical range, and contains an infrared spotlight. 19. The device according to p. 12, characterized in that the block of additional filters of the cartographic index of the true course, the block of additional filters of the main cartographic index of a given course and the control unit of the filters are made on tunable computing environments. 20. The device according to p. 12, characterized in that the structurally tunable control unit is implemented on tunable computing environments.

Images