Device for classifying noisy objects

The invention relates to the field of hydro-acoustics and is intended for the recognition of objects by their noise.

In systems based on the analysis of noise emission for the detection and classification of objects, use features of the spectral composition of the signal, the so-called "portrait" [1]. Usually, a spectral “portrait” of an object is understood to mean a limited set of characteristic harmonics, the amplitude of which exceeds the continuous part of the spectrum formed by harmonics with close amplitudes. These characteristic harmonics are called discrete components, or discretes.

A device for classifying noise objects is known [2], based on measuring the average frequency of the noise emission band of a signal, and then comparing this frequency with the reference frequencies of objects of certain classes. Classification by this device is inefficient, since the average frequency of the spectrum band is determined primarily by the distance to the object [3], and only secondarily by the class of the object in terms of its noise emission.

A device for classifying noise objects is known [4], based on the analysis of the spectral composition of the signal noise emission, and the subsequent comparison of the detected discrete components with the reference discrete components of objects of certain classes. This device can basically work only if all noisy objects of one class have common discrete components, which is not observed for most modern noisy objects [5].

The closest analogue of the present invention is a device for the classification of noise objects [6], which is taken as a prototype.

The device prototype contains the following series-connected blocks:

multi-element acoustic receiving direction finding antenna;

analog-to-digital converter;

signal shaping unit;

discrete component detection unit;

block deciding on the class of the object

indicator.

The specified device by dividing the antenna into two halves and subsequent processing during the formation of the signal spectrum reduces the influence of the distributed noise interference received by the side field of the directivity pattern. However, the classification itself by this device is not effective enough for two reasons. First, the classification is based on an unstable procedure for detecting discrete components, in which there is always the likelihood of both missing detection of discretes and false detection of non-existent discretes [7]. Secondly, the device is not able to classify objects whose spectral composition of the signal is not known to the developer.

The objective of the invention is to increase the efficiency of the classification of noisy objects according to the spectral characteristics of their noise.

To solve this task, a noise-finding antenna analog-digital converter, a signal shaping unit that also contains a class of the object and an indicator sequentially connected to the multi-component acoustic receiving antenna, sequentially connected multi-element acoustic signals are added to the device for classifying objects, namely: forming a measure of similarity and a block of permanent memory, while the input of the block forming a measure of similarity is connected to the output of a signal shaping unit ala, and its output is connected to the input of the class decision making unit, the second output of which is connected to the input of the permanent memory unit, the output of the permanent memory unit is connected to the second input of the similarity measure forming unit, while the similarity forming unit is designed as a calculator, implements the calculation of a set of measures of similarity between the formed spectrum of the measured signal and the spectra of the reference classes from the base stored in the block of permanent memory, followed by the choice of the maximum measure of similarity, and the class decision block is made in the form of a threshold comparison device of the maximum measure of similarity with the threshold and output depending on the result of the comparison, or the class decision on the indicator, or the spectrum of the measured signal in the block of permanent memory to supplement the base of the spectra of the reference classes.

Additionally, to communicate with external devices, a read and write unit can be entered into the device, connected by two-way communication with the permanent memory unit, and realizing the movement of the base of spectra of the reference signals between the permanent memory unit and external storage media.

The technical result of the invention is the provision of the possibility of classifying objects, the spectral characteristics of which are not known a priori by the developer, and, in particular, of objects, in the spectrum of which there are no pronounced characteristic discrete components.

We show the possibility of achieving the specified technical result of the proposed device.

Introduction to the device unit of forming measures of similarity in the aggregate by the block of permanent memory allows to use for analysis not a limited set of discrete components, the reference list of which is fixed by the developer, as in the device-prototype, but a complete set of spectrum harmonics (the whole spectrum). The total set of harmonics includes both characteristic discretes and all other harmonics that form a continuous part of the spectrum. This makes it possible to take into account in classifying the full set of object properties present in its spectrum: both a set of characteristic discretes, as in [4, 6], and the shape of the solid part of the spectrum, and the average frequency of the spectrum band, as in [2]. In addition, the procedure for forming similarity measures [8], based on the analysis of the entire spectrum, does not require preliminary detection of characteristic discretes. Concerning, the procedure will work correctly even in the case when the object has characteristic voices that are weak or absent. A block of permanent memory, in addition to storage, allows the device to supplement the base of signal spectra, consisting of a complete set of harmonics of the spectrum, with new entries that were not known a priori by the developer, and thereby expand the list of classes.

To ensure the operation of the newly introduced blocks, the decision block is made in the form of a threshold device providing feedback between the similarity measure generation unit and the permanent memory block when the spectrum of the newly measured signal is different from all spectra of the reference classes from the base.

A device for reading and writing may additionally be introduced into the inventive device, which will allow transferring the base formed during the operation of the device to other device samples, ensuring the accumulation of a common body of knowledge about the spectral “portraits” of noisy objects. The accumulated body of knowledge can be used in the future, both for applied classification problems and for scientific problems of developing new classification methods.

Thus, the classification efficiency of the claimed device is increased relative to analogs that use only one property of the object spectrum for classification.

The invention is illustrated FIG. 1, which shows a block diagram of a device for classifying noisy objects.

The claimed device (Fig. 1) contains serially connected blocks: multi-element direction finding antenna 1, analog-to-digital converter 2, signal shaping unit 3, similarity measure generating unit 4, class decision making unit 5, indicator 6, and also unit 8 constant memory, the input of which is connected to the second output of block 5, and the output to the second input of block 4.

The device can be supplemented by a block 7 of reading and writing, connected by a two-way communication with block 8.

Multiple-element antenna 1 noise-finding is known from [9], while the antenna modules can be performed according to [10]. A / D conversion (block 2) is performed according to [11]. Block 3 of the signal spectrum formation includes the procedures of fast Fourier transform, accumulation over time, determination of the total and difference spectra of two semi-antennas for clearing the signal spectrum from interference according to [6]. Block 4 forming similarity measures and block 5 making a decision about a class can be performed in modules of programmable signal processors and in modules of the universal part of a digital computing complex [12]. As measures of similarity can be used any of the known measures of similarity, for example, the correlation coefficient or the Euclidean distance [8]. The threshold for decision making is selected based on the required probability of a false classification according to [7]. The indicator, the block of permanent memory and the block of reading and writing are implemented, for example, in a remote control device and an optical disk drive according to [12]. The base of spectra has a unified storage structure. Each record except the table of the signal spectrum with the columns “frequency” and “amplitude” contains related information: the name of the reference class, the date, time and conditions of the recording.

In the dynamics of the device operation, the noise signal of objects, received by antenna 1, enters block 2, in which the signal is transferred from analog to digital form. Next, the signal enters block 3, in which the signal is cleared from the distributed interference, transferred to the frequency domain and accumulatede in time. Next, the measured spectrum of the signal goes to block 4. At the same time, block 4 of block 8 receives the base of the spectra of the reference classes of signals, formed to the current point in time. In block 4, the procedure for forming measures of similarity between the measured signal spectrum and each signal spectrum of the reference class from the base is carried out. The maximum of the obtained similarity measures, the corresponding name of the reference class and the measured signal spectrum arrive in block 5, which implements the procedure for comparing the similarity measure with the threshold. If the value of the maximum measure of similarity exceeds the threshold, the name of the reference class comes from block 5 to indicator 1 for display. Otherwise, from block 5 to block 8 the measured spectrum of the signal enters. In block 8, the base of the spectra of the signals of the reference classes is supplemented by a new record containing the spectrum of the measured signal. At the next operation of the device in block 4 from block 8, the newly formed base of spectra of reference classes will be received. Then, in block 4, an additional set of measures of similarity will be formed, which will allow in block 5 to decide on the class and display it on the indicator 6.

Block 7 is included in the operation of the device as needed, and allows you to write the database to an external storage device for transfer to another sample device, or read the database formed on another sample device.

All of the above allows us to consider the problem of the invention solved. A classification device for noisy objects is proposed, which provides for the classification of objects whose spectral characteristics are not known a priori by the developer, and, in particular, for objects whose spectrum does not contain pronounced characteristic discrete components.

INFORMATION SOURCES

1. Burdick BC Analysis of hydroacoustic systems. L .: Shipbuilding. 1988

2. Timoshenkov V.G., Dyadchenko T.Z. The patent of the Russian Federation №2262121 from 04.24.2003. A way to classify rustling objects. IPC G01S 3/80

3. Horton J. U Basics of sonar. L .: Sudpromgiz. 1961

4. Deev V.V. and others. Analysis of information by the operator - hydroacoustics. L .: Shipbuilding. 1989

5. Myasnikov LL, Myasnikova E.N. Automatic recognition of sound images. L .: Energy. 1970

6. Afanasyev A.N. Znamenskaya T.K. The patent of the Russian Federation №2546851 from 11.12.2013. A method for classifying sonar acoustic signals of a marine object. IPC G01S 3/80

7. Tyurin AM Introduction to the theory of statistical methods in hydroacoustics. L. 1963

8. Tu J., Gonzalez R. Principles of pattern recognition / Per. from English M .: Mir. 1978

9. Litvinenko SL. RF patent №2515133 dated 05/10/2014 Spherical sonar antenna. IPC G01S 15/00

10. Smaryshev, MD, Chernyakhovsky, A.E., Ivanov, AM, Shatokhin, A.V., Seleznev, I.A., Nikandrov, V.A., Malyarov, K.V., Barsukov, Yu.V. The patent of the Russian Federation №2539819 from 01/27/2015. Antenna module with digital output. IPC H04R 1/44

11. IFisher E.S., Jervis B.U. Digital signal processing: a practical approach / Trans. from English M .: OOO “I.D. Williams. 2017

12. Betelin V.B., Kapustin G.I., Kokurin V.A., Koryakin Yu.A., Liss A.R., Nemytov A.I., Pershin A.S., Ryzhikov A.V., Chelpanov A.V., Shalin S.A. The patent of the Russian Federation №2207620 from 10.03.2003. Digital computing system for processing signals in sonar systems. IPC G06F 15/16, G01S 15/88