RU2730048C1 - Adaptive dichotomous classification method of marine objects - Google Patents

Abstract

FIELD: hydro acoustics.SUBSTANCE: invention relates to hydroacoustics, can be used in solving problems of processing sea objects signal with passive and active hydroacoustic means and is intended for separation of detected objects into two alternative classes. Method is based on successive reception of hydroacoustic signals of objects, determination of classification characteristics of objects and their comparison with threshold for decision on class of objects. When implementing the method, during its operation, a statistical description of alternative classes is formed in terms of a given classification characteristic, which enables to use an adaptive threshold for classification, which is dynamically formed on the previous cycle of operation of the device.EFFECT: method enables classification of objects in real natural conditions when hydro-acoustic conditions and interference-signaling environment in the marine environment are not known or known in part.1 cl, 1 dwg

Description

The invention relates to the field of hydroacoustics, can be used in solving problems of signal processing of marine objects by passive and active hydroacoustic means and is intended to divide the detected objects into two alternative classes.

Known methods [1-3] for the recognition of marine objects detected by their primary or secondary hydroacoustic field into two alternative classes. The disadvantage of these methods lies in the use of a threshold that does not take into account all the features of the hydrological-acoustic and noise-signaling environment, which leads to insufficient efficiency of the classification in natural conditions.

The closest analogue of the proposed invention is a method for classifying marine objects in a typical sound direction-finding station [4].

In the prototype method, the following operations are performed:

set the type of the classification characteristic as the ratio of the received signal power to the angular velocity of the object,

form a theoretical threshold for classifying objects into two classes with respect to a given classification characteristic in given hydrological and acoustic conditions,

receive the signal of the object with a hydroacoustic antenna,

determine the value of the classification characteristics of the object,

make a decision about the class of the object based on the comparison of its classification characteristics with the threshold.

The method allows using one classification characteristic, defined as the ratio of the received signal power to the angular velocity of an object, to divide objects into classes of low-noise and high-noise in a two-dimensional feature space "noise" - "speed".

The disadvantage of this method lies in the complexity of the formation of the threshold in the transition from the space "noise" - "speed", in which the threshold C is fixed on the basis of the statistical description of classes of objects [5], to the space "received signal power" - "angular velocity of the object". In the prototype method, it is proposed to use the SA threshold for this transition, which is the product of the C threshold by the coefficient A, which characterizes the signal attenuation during propagation in the medium, and determined in the assumed hydrological-acoustic conditions according to the theoretical method [6]. Despite the fact that, theoretically, the indicated technique allows one to take into account the vertical distribution of the speed of sound along the depth and the reflecting and absorbing properties of the bottom and surface, in real field conditions, these parameters may differ from the average statistical ones or may not be fully known. In addition, the methodology does not allow taking into account the noise-signaling situation in the area, that is, the presence and structure of noise interfering with navigation or interference of biological origin. All these factors lead to the fact that the SA threshold, calculated theoretically, does not allow effectively separating objects in real field conditions.

The objective of the invention is to improve the efficiency of classification of objects detected by hydroacoustic means in natural hydrological and acoustic conditions.

To solve the set problem in the method of adaptive dichotomous classification of sea objects, in which the type of classification characteristic is set, a theoretical threshold for classifying objects into two classes X and Y is formed with respect to a given classification characteristic under given hydrological-acoustic conditions and it is taken as the current threshold, a signal is received the first object with a hydroacoustic antenna, determine the value of the classification characteristic of the first object, make a decision on the class of the first object based on comparing its classification characteristic with the current threshold,

introduced new features, namely

create two empty arrays of fixed length measurements, one for each class X and Y,

write in the array corresponding to the adopted decision on the class of the first object, the value of its classification characteristic,

receive the signal of the second and subsequent objects,

determine the values of the classification characteristics of the accepted objects, make a decision on the class of the accepted objects based on their comparison

classification characteristics with the current threshold,

complement the arrays corresponding to the adopted decision on the class with the values of the classification characteristics of the accepted objects, using sliding windows for this,

где μ_X и μ_Y - математические ожидания, a w_X и w_Y - величины, обратные дисперсиям, для совокупности классификационных характеристик в массиве класса X и массиве класса Y, соответственно,when the arrays are full, form a new adaptive classification threshold

where μ _X and μ _Y are mathematical expectations, aw _X and w _Y are the inverse of variances for a set of classification characteristics in an array of class X and an array of class Y, respectively,

take the received adaptive threshold as the new current threshold and update it after each next filling of the arrays.

The technical result of the invention is to provide the possibility of classifying objects in real natural conditions, when the hydrological-acoustic conditions and noise-signal conditions in the marine environment are not known or are known in part.

Let us show the possibility of achieving the specified technical result by the proposed method.

Hydroacoustic means for which both the prototype method and the proposed method are used are never used to detect a single object. In the water area in which they are used, there are always several objects that are detected and classified by hydroacoustic means sequentially. This allows us to propose to use the statistics accumulated during the classification of the first n objects for the classification of n + 1 objects, and so on.

The physical essence of the proposed method lies in the formation of a threshold for the classification of objects for a given type of classification characteristics based on a dynamically updated statistical description of classes of objects directly in the process of operation. For example, for the classification characteristic of the prototype method [4], defined as the ratio of the received signal power to the angular velocity of an object, there is no theoretical description of object classes, since such a description will be different for different hydrological-acoustic and noise-signal conditions. For the methods [2, 3], used in passive and active sonar, and using the angle of arrival of the signal in the vertical plane as a classification characteristic, a theoretical description of object classes is possible, but it does not take into account the full-scale noise-signal conditions. And the new procedures of the proposed method allow you to create a statistical description of the classes of objects, and, therefore, to form the actual threshold, in the dynamics of its work, using for this the classification characteristics of the first, second, and subsequent objects.

When using the principle of maximum likelihood to form a threshold separating alternative classes, it is enough to obtain two statistical parameters for each of the classes: the mathematical expectation and variance [7] of the distribution of the classification characteristic. To do this, create two arrays of measurements, one for each class X and Y, in which the values of the measured classification characteristics are dynamically stored during the production process. The array of statistics of class X stores the values of the classification characteristics of those objects for which the decision about class X was made in the previous step, and the array of statistics of class Y stores the values of the classification characteristics of other objects for which the decision about class Y was made. When filling the arrays, sliding windows are used : the oldest information is erased, the entire array is shifted down, and the current information is recorded as the most current. This allows you to constantly update the statistics of classes, adapting it to the current hydrological-acoustic and noise-signaling environment and generate an updated threshold.

Based on the information of the filled arrays, the main statistical parameters for each of the classes are dynamically calculated - the mathematical expectation and the variance of the distribution of the values of the classification characteristic. Then the classification threshold is formed based on the maximum likelihood principle [7].

During the operation of the method, the statistics of the classification characteristics for alternative classes are constantly updated, that is, the description of the classes of objects for the current hydrological-acoustic and noise-signal conditions is dynamically refined, on the basis of which the actual threshold for classification is formed.

Thus, the efficiency of the classification in the claimed method is increased relative to the prototype and other dichotomous classification methods using a fixed threshold, analytically calculated on the basis of a theoretical model of sound propagation.

The essence of the invention is illustrated in FIG. 1, which shows a block diagram of a device that implements the proposed method.

The device (Fig. 1) contains series-connected blocks: hydroacoustic antenna 1, block 2 of spatial and time-frequency signal processing (PSHP), block 3 for deciding on the class of the object, indicator 4 (IND), block 5 of read-only memory (ROM) , block 6 for forming an adaptive threshold (FAP), and the output of the latter in the chain of block 6 is connected by feedback to the second input of block 3. The device also contains a block 7 for forming a theoretical threshold (FTP), the output of which is connected to the third input of block 3.

Hydroacoustic antenna 1 is known from [8], while antenna modules can be made according to [9]. The procedures of block 2 of spatial and frequency-time processing are known from [10] and can be performed in modules of programmable signal processors and in modules of the universal part of a digital computing complex [11]. Block 3 for making a decision on the class of the object, block 6 for forming an adaptive threshold and block 7 for forming a theoretical threshold can be performed in the modules of the universal part of the digital computing complex [11]. The indicator 4 and the read-only memory unit 5 are implemented, for example, in a control panel and an optical disk drive according to [11].

The method is implemented as follows.

When the device is turned on in block 7, a theoretical classification threshold is formed according to the methodology for a given classification characteristic. For example, according to the method [4] for the classification characteristic, the ratio of the received signal power to the angular velocity of the object, or according to the method [2] for the classification characteristic, the angle of arrival of the signal in the vertical plane, or according to any other method for a given classification characteristic.

In the dynamics of the device operation, the noise signal of the first, second and subsequent objects, received by antenna 1, enters block 2, where typical noise direction finding procedures are performed [10]: spatial processing and time-frequency processing of the signal. At the output of block 2, the value of the given classification characteristic Z of each object is formed, for example, according to [4], the ratio of the received signal power to the angular velocity of the object. The classification characteristic enters block 3. At the same time, the theoretical classification threshold comes to block 3 from block 7, which at this stage of the device operation is taken as the current threshold. In block 3, a threshold classification scheme is implemented [10], as a result of which the object is referred to one of two alternative classes X or Y. The classification characteristic and the decision made about the class of the object are sent to block 4, where they are indicated to the operator, and to block 5 of permanent memory devices.

In block 5, the current value of the classification characteristic is recorded into one of two arrays, depending on the decision made about the object class: into the array of object statistics X (X _i = Z) or into the array of object statistics Y (Y _i = Z). Recording is carried out sequentially, as the classification characteristics are found and developed, for the first, second and subsequent objects. At the same time, when filling the arrays, sliding windows are used: the oldest information is erased, the entire array is shifted down, and the current information is recorded as the most relevant. Arrays of statistics of classification characteristics for two classes of objects are fed to block 6 for forming an adaptive classification threshold.

In block 6, the number of filled elements in arrays X and Y is analyzed. While the arrays are not completely filled, which will be in the first cycles of the device operation, the adaptive threshold for classification is not formed, and the theoretical classification threshold is still used. When the arrays are completely filled, then in block 6, first, statistical parameters for each of the classes are obtained, and then a threshold for classification by the least squares method is formed, which provides the maximum likelihood for further decision making [7]:

- математическое ожидание классификационной характеристики для класса X,

- mathematical expectation of the classification characteristic for class X,

- математическое ожидание классификационной характеристики для класса Y,

- mathematical expectation of the classification characteristic for class Y,

- дисперсия классификационной характеристики класса X,

- variance of the classification characteristic of class X,

- дисперсия классификационной характеристики класса Y,

- variance of the classification characteristic of class Y,

- адаптивный порог для классификации,

- adaptive threshold for classification,

where w _X = 1 / D _X , w _Y = 1 / D _Y are the inverse of variances.

The resulting adaptive threshold for classification goes to block 3, where it will be used to make a decision about the class at the next cycle of the device when classifying the next object. In this case, after the classification of the next object, its classification characteristic will again be written into one of the arrays of statistics of class X or class Y, depending on the decision made, and the classification threshold will again be updated based on new statistical parameters for each of the classes.

All of the above allows us to consider the problem of the invention solved. A method for adaptive dichotomous classification of marine objects is proposed, which provides the classification of objects in real natural conditions, when the hydrological-acoustic conditions and noise-signal conditions in the marine environment are not known or are known in part.

SOURCES OF INFORMATION

1. Velichkin S.M., Mironov D.D., Antipov V.A., Zelenkova I.D., Perelmuter Yu.S. RF patent No. 2156984 dated 09/27/2000. A method for obtaining information about an object noisy in the sea and a method for obtaining color scales for it. IPC G01S 3/84.

2. Volkova A.A., Konson A.D. RF patent No. 2703804 dated 10.22.2019, priority dated 10.22.2018. Method of classification of marine objects by passive hydroacoustic means. IPC G01S 3/80.

3. Libenson EB, Strelenko TB RF patent No. 2528114 dated 09/10/2014, priority dated 04/11/2013. Active sonar with object classification. IPC G01S 15/00.

4. Volkova A.A., Konson A.D. RF patent No. 2689968 from 05/29/2019, priority from 03/29/2018. Method of classification of offshore objects in a typical sound direction-finding station. IPC G01S 3/80.

5. Kobylyansky V.V. Development of algorithms for the classification of noise hydroacoustic signals based on the use of acoustic-structural characteristics of radiation objects and environmental models. Dissertation for the degree of Ph.D. L .: Central Research Institute "Morfizpribor". 1982.

6. Evtyutov A.P., Mitko V.B. Examples of engineering calculations in hydroacoustics. -L .: Shipbuilding. 1981.265 p.

7. Taylor J. Introduction to the theory of errors / Per. from English. - M .: Mir. 1985.272 p.

8. Litvinenko S.L. RF patent No. 2515133 dated 05/10/2014 Spherical hydroacoustic antenna. IPC G01S 15/00.

9. Smaryshev M.D., Chernyakhovsky A.E., Ivanov A.M., Shatokhin A.V., Seleznev I.A., Nikandrov V.A., Malyarov K.V., Barsukov Yu.V. RF patent No. 2539819 from 27.01.2015. Antenna module with digital output. IPC H04R 1/44.

10. Yu.Koryakin Yu.A., Smirnov S.A., Yakovlev G.V. Ship sonar technology: state of the art and current problems. SPb .: Science. 2004.

11. Betelin V.B., Kapustin G.I., Kokurin V.A., Koryakin Yu.A., Lise A.R., Nemytov A.I., Pershin A.S., Ryzhikov A.V., Chelpanov A.V., Shalin SA. RF patent No. 2207620 dated 03/10/2003. Digital computing complex for signal processing in hydroacoustic systems. IPC G06F 15/16, G01S 15/88.

Claims (1)

The method of adaptive dichotomous classification of sea objects, in which the type of classification characteristic is set, a theoretical threshold for classifying objects into two classes X and Y is formed with respect to a given classification characteristic under given hydrological-acoustic conditions and is taken as the current threshold, the signal of the first object is received by a hydroacoustic antenna, determine the value of the classification characteristic of the first object, make a decision about the class of the first object based on the comparison of its classification characteristic with the current threshold, characterized in that they create two empty arrays of fixed length measurements, one for each class X and Y, write to the array corresponding to the decision on the class of the first object, the value of its classification characteristics, receive the signal of the second and subsequent objects, determine the values of the classification characteristics of the received objects, make a decision on the class of the received objects on Based on the comparison of their classification characteristics with the current threshold, they supplement the arrays corresponding to the adopted decision on the class with the values of the classification characteristics of the accepted objects, using sliding windows for this, when the arrays are completely filled, form a new adaptive classification threshold

where μ _X and μ _Y are mathematical expectations, aw _X and w _Y are the reciprocal of variances for a set of classification characteristics in an array of class X and an array of class Y, respectively, take the resulting adaptive threshold as a new current threshold and update it after each next filling arrays.

Images