RU2687994C1 - Method of classification of marine objects based on weight coefficients of their classification features - Google Patents

Abstract

FIELD: physics.SUBSTANCE: present invention relates to radio electronics and hydroacoustics, namely to devices for classification of noise and signals of sea targets. Device enables to classify noise and target signals with the specified accuracy for each time interval from the moment of detection. Invention is based on using an artificial neural network with calculated weight coefficients of input synapses. Distinctive feature of the method is the possibility of determining the accuracy of automatic classification at each time interval from the moment of target detection.EFFECT: technical result is possibility of automatic (automated) classification of noise and signals of sea targets in terms, which comply with standards of hydroacoustic operators of submarines, as well as high accuracy of classification due to use of additional classification features.1 cl, 14 dwg, 10 tbl

Description

The present invention relates to the field of hydroacoustics and can be used to build systems for the automatic (automated) classification of objects found in various paths of a hydroacoustic complex.

The task is solved by creating a matrix of classification signs that are detected in the main paths of the sonar complex (such as noise finding, sonar, mine detection, detection of sonar signals, communication and recognition, classification (spectral analysis of the envelope and carrier signal), the noise control system, as well as in the auditory canal) with the most informative.

Achievable technical result is to determine the weighting coefficients of classification signs on the basis of the frequency of their appearance for different classes of goals. The high efficiency of this device makes it possible to determine the following classes of goals with minimum time costs with high accuracy:

- salvo torpedo weapons (ZTO);

- mine (M);

- combat swimmer (BP);

- rocket weapon launch (SRO);

- explosion (B);

- submarine (submarine);

- PL simulator (IPL);

- device hydroacoustic suppression (HAP);

- noise of a biological nature (SHBH);

- fishing trawler (RT);

- transport (TP);

- Battle surface ship (BNK);

- anti-submarine helicopter (BT);

- anti-submarine aircraft (SM);

- group goal (GR);

- iceberg (A);

- polynya, razvodye, the edge of the ice field (L);

- fixed structures at sea, drilling platforms (ST);

- underwater navigation obstacle (NP);

- own disturbance (P).

Prototype. Currently, to solve this problem, the most widely used method is the frequency-time signal processing of the received antenna path of the direction finding noise [1, 3, 4]. Of the preceding methods, the most interesting is the method of obtaining information about an object that is noisy at sea (prototype; patent No. 2156984 dated 09/27/2000). The device operation scheme is shown in FIG. one,

This method contains the following operations:

- receiving sonar noise signal using the generated characteristics of the antenna;

- time-frequency signal processing, including the formation of frequency ranges, detection, temporal averaging;

- indication on the screen of the trace of signals of a noisy object in each frequency range.

The method provides the ability to simultaneously detect, classify and determine the distance to a noisy object.

To solve the problem of classification, the method of obtaining information about a noisy object, which includes receiving a sonar noise signal, frequency-time signal processing with the formation of frequency ranges, detection and averaging of signals in frequency ranges, indication of the signal path in each frequency range, has the following operations:

- before display, the signal of each frequency range is coded with its own color, while the brightness of each color component is determined by the signal level in the corresponding frequency range, and the connection of color components into a color representation is produced, for example, in the same way as on the color kinescope screen;

- produce an indication of the color image of the signal of a noisy object in time, with the resulting color at each point on the screen when the display is determined by the ratio of signal levels in the frequency ranges (target indicator trace);

- determining the class of a noisy object and estimating the distance to it by comparing the color of the indicator trace of the object signal with the color scales of noisy objects of different classes by distance.

The disadvantages of this method is to use only the direction finding system for classifying, ignoring information from sonar routes, mine detection, detection of sonar signals, communication and identification, classification (spectral analysis of the envelope and carrier signal), the noise monitoring system, as well as in the auditory channel. In addition, only 2 classification signs are used in the noise-finding line - the frequency range and the signal-to-interference ratio, ignoring such classification features as target length, bearing change rate, stability of automatic target tracking, number of directional angles where the target is observed.

Thus, in this method, the classification is generally made according to two classes - a highly noisy object and a low-noise object, which does not give a reliable picture to the operator of the hydroacoustic complex about the surface, underwater and air conditions.

Analog. Also known is the method of recognition (classification) of offshore objects, i.e. class definition of a noisy object based on the formation of attribute descriptions of signals using features of the spectral composition of noise emissions of objects of various classes with the operator presenting the results of spectral, correlation, cross-correlation and inter-spectral analyzes of the received signal in various frequency bands of hydroacoustic stations

This method (analog) [2, p. 141], which implements this classification method, is shown in FIG. 2,

The principle of operation of this device is as follows:

The recognition system (classification) is based on the use of an automatic system recognition system in the decision-making process about the object being analyzed using the operator.

For the formation of characteristic descriptions of hydroacoustic signals, such methods of measuring and analyzing random processes as spectral, correlation, statistical, cross-correlation, and inter-spectral analyzes are used. The choice of signs is made by analyzing all the information in the n-dimensional space of signs, from which a minimized description is obtained in another space.

A device that implements this method does not meet the basic technical requirements - reliability and recognition time of an object.

The proposed method. As a rule, a combination of different signs is used in all classification tasks, as each individually does not provide the required probability of correct target recognition. The target classification path uses information from both primary and secondary processing systems and its own devices to isolate the necessary features.

In the process of classification, it is necessary to choose those signs that are more informative. The number of signs used is determined by the probability of correct classification.

Classification signs can take on different values. Based on operating experience of hydroacoustic complexes, these values can be divided into four ranges for each classification attribute. For ease of use in digital processing, they are formalized by their digital values from 0 to 3. Possible options for the adoption of values are shown in Table 1.

For reasons of protecting state secrets, the ranges are designated by the letters A, B, C, and D for each attribute. These values are presented in [5].

Matrix classification signs for groups of objectives according to the degree of danger are presented in Table 2, Table 3 and Table 4, respectively. So in table 2 presents the matrix of classification signs for the purposes with signs of the use of weapons.

Table 3 presents the matrix of classification signs of warships and aircraft.

Table 4 presents the matrix of classification signs of civilian ships and noise that interferes with the operation of the hydroacoustic complex.

To determine the weighting coefficients of the classification signs of marine targets, it is necessary to determine the frequency of use of each value of classification signs. The rarer the characteristic value is, the higher the classification accuracy and the higher the weighting factor, respectively. Frequency of use of values of classification signs is given in table 5.

The weighting factors are calculated in the following sequence:

1. All factors are determined for which the weighting factor is to be determined.

2. Each factor is assigned a certain rank depending on the degree of significance of the factor.

3. The ranking values of all factors are summed up.

4. The rank of each factor is divided by the sum of all rank values.

Factors for which you need to determine the weighting factor are the classification signs of goals. As it was said earlier - the less often the value of a feature occurs, the higher the accuracy of classification and, correspondingly, the higher the rank of the significance factor. Let us take for the rank of significance the variance of the frequency of occurrence of a classification attribute. The higher the variance of the frequency of occurrence of a classification attribute, the higher the rank of significance, the more accurate the classification.

where D (x _i ) is the variance of the frequency of occurrence of the i-th classification characteristic, calculated according to the data of table 5 and presented in table 6;

X _i - the frequency of use of the variant of the value of the i-th classification characteristic, presented in table 5;

- математическое ожидание частоты использования варианта значения i-го классификационного признака;

- mathematical expectation of the frequency of use of the variant of the value of the i-th classification characteristic;

n is the number of variants of the classification attribute values.

Then the weight coefficient of the i-th classification feature will be equal to

Ranked weight data is presented in Table 6

The assignment of a weight coefficient to a discrete value of a classification attribute allows the classification ambiguity to be solved if the identified classification criteria are inconsistent. The data presented in the article ranking it is advisable to use both automatic and automated classification of noise and signals of marine targets.

The invention works as follows:

The invention is shown schematically in FIG. 3. Information about the surrounding underwater, surface and air situation comes from the antennas of the hydroacoustic complex to the corresponding paths in which 31 classification signs are produced, indicated in Table 1, respectively:

- 6 signs in the path of noise-finding;

- 2 signs in the sonar path;

- 6 signs in the path of the mine;

- 3 signs in the path of detection of hydroacoustic signals;

- 2 signs in the communication and recognition path;

- 2 signs in the noise control system;

- 6 signs in the equipment of spectral analysis;

- 4 signs in the auditory canal.

31 signs presented by the data sets on correspondence to each of the possible options for the sign values, except for the option when there are no data on the sign (a total of 69 values) falls on the 69 inputs of the artificial neural network (as part of the classification path) shown in FIG. 4, FIG. 5, FIG. 6, FIG. 7, FIG. 8, FIG. 9, FIG. 10, FIG. 11 and FIG. 13. Values at the 20 outputs of the artificial neural network presented in FIG. 12, are numbers from 0 to 1, reflecting the assessment of the artificial neural network of correspondence between the sets of data on the signs supplied to the inputs and the 20 classes of targets presented in Table 2, Table 3 and Table 4. Then these values are given to (device for constructing a chart of values classification signs), where classes of goals are grouped according to Table 7, for display by color code. Then information about the target classes is fed to the indicator, where they are displayed in the form of a bar chart, shown in FIG. 14.

When preparing training samples for teaching an artificial neural network, the frequencies of using the values of classification signs presented in Table 5 were used. As an artificial neural network, a multilayer perceptron is used, with an input layer containing 69 artificial neurons, one hidden layer containing 31 artificial neurons, and output a layer containing 20 artificial neurons. The network structure is shown in FIG. 13. The parameters of the artificial neurons of the trained network are presented in Table 8.

The weights of the inputs of the hidden layer neurons are presented in Table 9

The weights of the inputs of the neurons of the output layer are presented in Table 10

Comparative analysis of the method of classification of marine objects on the basis of the weighting coefficients of their classification features with a prototype and analogue showed that the proposed approach has a more accurate classification than the prototype (31 evaluated classification attribute versus 2 classification features of the prototype), as well as less decision time than the analog.

List of used sources

1. Velichkin S.M., Mironov DD, Antipov V.A., Zelenkova I.D., Perelmuter Yu.S. The method of obtaining information about an object noisy at sea and a method of obtaining color scales for it. The patent for the invention №2156984 from 09/27/2000 (Prototype).

2. Deev V.V., Zabrodin Yu.B., Pakhomov A.P., Tenetko V.A., Titov M.S. Information analysis by hydroacoustic operator. - L .: Shipbuilding, 1989, p. 141 (Analog).

3. Burdick B.C. Analysis of sonar systems. L .: Shipbuilding, 1988, p. 377.

4. Evtyutov A.P., Kolesnikov A.S., Lyalikov A.P. Handbook of hydroacoustics. - L .: Shipbuilding, 1982, p. 14, 15.

5. Balakhonov P.N. Weighting coefficients of classification signs of marine targets. St. Petersburg: Interdepartmental Scientific and Technical Conference, reviewed collection of articles and reports No. 2 (20) - 2017, VMPI, p. 299-304.

Claims (1)

The method of classification of marine objects based on the weighting coefficients of their classification signs, containing a solution to the problem of automated (automatic) classification of marine objects, characterized by increased classification accuracy and reduced object recognition time, artificial neural network is also used for 69 inputs (by the number of classification signs) and 20 outputs (by the number of recognized classes of targets), when preparing training samples for an artificial neural network rimenyayutsya frequency of use of the classification feature values for all classes of targets.

Images