JP2001281328A - Device and method for identifying target - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a target identifying device with high target identifying precision to a low S/N ratio echo capable of considerably lowering an erroneous target ratio even when a detection ratio of the low S/N ratio echo is increased. SOLUTION: This target identifying device is provided with at least a wave transmission/reception part 1 transmitting a sonic wave with a modulated frequency and receiving its reflection wave from a target, a preprocessing part 2 analyzing a frequency of a signal received in the transmission/reception part 1, intensifying characteristics of an echo constituent as the reflection wave from the target and forming neutral network inputting data with weakened non-echo characteristics, and a neutral network processing part 3 having a predetermined neutral network and determining whether the neutral network inputting data formed in the preprocessing part 2 is an echo or a non-echo.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus for receiving an acoustic signal (a reflected wave of a modulated sound wave) from a target and identifying the target based on information about the target contained in the received signal. About the method.

[0002]

2. Description of the Related Art Conventionally, as this type of target identification method, there has been known a method in which information on a target included in a received signal is displayed in a useful form, and an observer visually identifies the target to identify the target. Also, in order to make it easier for the observer to visually identify the target, for example, various signal processing is performed on the received signal, and the position of the target obtained from the information on the direction and distance of the target included in the received signal is calculated. There is also proposed a method of displaying a plurality of gradations on a display device according to the signal intensity. In addition, there is a method in which the observer directly listens to an acoustic signal from the target via an output device such as a speaker to identify the target.

[0003] In any of the above target identification methods, the ability of the observer greatly affects the identification accuracy. So recently,
For the purpose of improving identification accuracy, a target identification method in which target identification processing is automated has been proposed as a method that does not rely on the ability of the observer. As an example, Japanese Patent Application Laid-Open No. 5-297114 discloses an echo and a reverberation (here, meaning noise).
A sonar signal processing device using a neural network that has learned the characteristics of (1) is disclosed. FIG. 10 is a block diagram showing a main configuration of the sonar signal processing device.

[0004] Referring to FIG. 10, the sonar signal processing device includes a transmitting / receiving section 101, a preprocessing section 102, and a neural network processing section 103. The transmission / reception unit 101 transmits a frequency-modulated sound wave (FM wave) and receives an echo signal from a target. The pre-processing unit 102 includes the transmitting / receiving unit 10
1. A received signal of a predetermined time length is extracted from the signal received at 1 while repeatedly shifting the signal by a required time, and the extracted received signals of the predetermined time length are sequentially subjected to fast Fourier transform (hereinafter, referred to as FFT) to obtain spectral data. Ask for. Further, the preprocessing unit 102 extracts, from the obtained spectrum data, spectral data having a time length corresponding to the pulse length of the FM wave which is a transmission wave while repeatedly shifting the required time, and extracts the three-dimensional “frequency-amplitude-time”. Neural network input data is created by creating data and performing normalization processing on the data so that the maximum value of the spectrum value becomes 1. The neural network processing unit 103 uses the neural network in which the characteristics of the echo and the reverberation are learned in advance, and converts the input data (three-dimensional data) of the neural network created by the preprocessing unit 102 into an echo (a reflected wave from a target). ) Or reverberation (reflection from something other than the target).

[0005] The neural network used here is the most typical three-layer network of a layered model, and comprises an input layer, a middle layer, and an output layer. The input layer has a unit having a two-dimensional structure corresponding to the distribution frequency of the amplitude spectrum of the neural network input data (three-dimensional data) created by the preprocessing unit 102, and each unit has a spectrum value of each distribution frequency. Is entered. The intermediate layer has units of one-dimensional structure, each unit being coupled to each unit of the input layer. In the intermediate layer, for each unit, a product-sum operation of the spectrum value and the weight input to each unit of the input layer is performed, and further, threshold processing is performed, and the result is output as an intermediate output. You. The output layer has an echo output unit and a reverberation output unit, each coupled to each unit of the intermediate layer. In this output layer, a product-sum operation of the intermediate output output from the intermediate layer and the weight is performed for each output unit, and further, threshold processing is performed. Based on the echo output output from the echo output unit and the reverberation output output from the reverberation output unit, it is determined whether the neural network input data (three-dimensional data) is an echo or a reverberation.

As a target identification process using a neural network, besides the above, there is an automatic target detection method described in Japanese Patent Application Laid-Open No. 8-152472. This target automatic detection method includes a reverberation receiving unit, a preprocessing unit, and a neural network determining unit. An echo signal from the target is received by the echo receiver. The pre-processing unit cuts out the received signal in a predetermined section and converts it into a detection pattern characterized by the spectrum of the echo signal (Fourier transform). The neural network determination unit uses a neural network that has learned in advance the relationship between the detection pattern and the desired output to determine from what object the detection pattern input from the preprocessing unit is reflected. A determination is made.

[0007]

However, any of the publications described in the above publications uses an FFT in the preprocessing unit.
Although spectrum data is obtained by using a frequency analysis method such as a frequency method, no condition is given to the frequency characteristics, and therefore, there are the following problems.

In general, the FFT method does not have a high instantaneous frequency resolution, so that for an echo having a low S / N ratio, data for inputting a neural network in which the difference between the echo and the reverberation is not clear is created. Further, in the conventional apparatus, the preprocessing unit has no means for enhancing the characteristics of the echo components, and even if it has one, it merely weakens the characteristics of the non-echo components, and has a low S / Its effect on N-ratio echoes was not satisfactory. Therefore, the conventional one has a low target identification accuracy with respect to the echo having a low S / N ratio. Where S
The / N ratio is the ratio of echo level to reverberation level. The instantaneous frequency resolution means the frequency resolution at the time of frequency analysis of the received signal for each short time (instantaneous).

It is important to increase the detection rate of the echo having a low S / N ratio in order to extend the distance at which the target can be identified. However, in the prior art, the erroneous target rate increases when trying to increase the detection rate of an echo with a low S / N ratio, and the erroneous target rate decreases when trying to decrease the erroneous target rate. There is also a problem that the detection rate of the echo decreases. Here, the false target rate is a probability that the reverberation is identified as an echo.

[0010] An object of the present invention is to solve the above-mentioned problems and to reduce the low S
It is an object of the present invention to provide a target identification apparatus and a target identification method which can suppress the erroneous target rate to some extent even if the detection rate of the E / N ratio echo is increased, and which has a high target identification accuracy for the low S / N ratio echo. .

[0011]

In order to achieve the above object, a target identifying apparatus according to the present invention comprises: a transmitting / receiving means for transmitting a frequency-modulated sound wave and receiving a reflected wave from a target;
Frequency analysis of the signal received by the transmitting and receiving means,
Preprocessing means for creating neural network input data in which the characteristics of the echo component that is the reflected wave from the target are strengthened and the features of the non-echo components are weakened, and the neural network input data as input. A predetermined neural network, and at least a neural network processing means for determining from the output of the neural network whether the input data for the neural network is an echo or a non-echo.

A method for identifying a target according to the present invention comprises the steps of: transmitting a frequency-modulated sound wave and receiving the reflected wave from the target; analyzing the received signal by frequency to obtain an echo of the reflected wave from the target. A step of creating neural network input data in which the feature of the component is strengthened and the feature of the non-echo component is weakened; and using a predetermined neural network, the neural network input data is an echo or non-echo. Identifying at least one.

The following is an operation of the target identification apparatus and the target identification method of the present invention.

[0014] By creating neural network input data that makes the difference between echo and reverberation more pronounced,
Neural network input data in which the difference between the echo and the reverberation is clear can be created even for an echo having a low S / N ratio.

In the present invention as described above, the neural network input data in which the characteristics of the echo components are strengthened and the characteristics of the non-echo components are weakened are created. In the network input data, the difference between the echo and the reverberation is more prominent. Therefore, even for echoes with a low S / N ratio,
Neural network input data in which the difference between echo and reverberation is clear can be created. Creation of such neural network input data can be realized by using a method having a high instantaneous frequency resolution for frequency analysis of a received signal. In the present invention, Wigner distribution processing or AR model distribution estimation processing is used for frequency analysis, and any of these processings has higher instantaneous frequency decomposition than the conventionally used FFT method.

[0016]

Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing a main configuration of a target identification device according to one embodiment of the present invention. This target identification device is an active sonar device,
It comprises a pre-processing unit 2 and a neural network processing unit 3. The transmission / reception unit 1 receives a linear frequency-modulated sound wave (hereinafter referred to as
LFM wave) and receives its reflected wave from the target. The pre-processing unit 2 creates neural network input data from the signal received by the transmitting / receiving unit 1. The neural network processing unit 3 performs preprocessing by using a neural network configured by combining a plurality of units, which has learned characteristics of an echo (a reflected wave from a target) and a reverberation (a reflection from a target other than the target). It discriminates whether the neural network input data input from the section 2 is an echo or a reverberation, and outputs the result.
The transmission / reception unit 1 and the neural network processing unit 3
It has a well-known configuration, and the one shown in the above-described conventional example can be used.

The feature of this embodiment resides in the processing in the pre-processing unit 2. The pre-processing unit 2 is configured to perform a frequency analysis process with a high instantaneous frequency resolution on the signal received by the transmission / reception unit 1 to obtain spectrum data, thereby obtaining an echo having a low S / N ratio. This makes it possible to create neural network input data that emphasizes the difference between the sound and the reverberation. The main configuration of the preprocessing unit 2 includes a frequency analysis processing unit 21, an extraction processing unit 22, an integration processing unit 23, and a normalization processing unit 24. Hereinafter, the operation of each unit of the preprocessing unit 2 will be described.

First, the operation of the frequency analysis processing section 21 will be described. FIG. 2 schematically shows the frequency analysis processing in the frequency analysis processing unit 21. The frequency analysis processing unit 21 sequentially extracts the reception signal b having the time length T from the reception signal a received by the transmission / reception unit 1 while repeatedly shifting the reception signal b by the time τ1, and performs the Wigner distribution for each of the reception signals b. The spectrum data c is obtained by performing a frequency conversion process by the method.

FIG. 3 shows a calculation image of the Wigner distribution method. In the Wigner distribution method, first, data b3 of the first half of received signal b is multiplied by data b2 of the second half to create data b3, and this data b3 is further subjected to FFT processing. For example, if the received signal b is data of X0 (N = 128 points), first, N / 2 point data X1 is created, and this is subjected to N point FFT processing.

When k = 0, the data X1 (k) is expressed as follows: X1 _real (k) = X0 _real (N / 2) × X0 _real (N /
2) + X0 _imag (N / 2) × X0 _imag (N / 2) X1 _imag (k) = X0 _imag (N / 2) × X0 _real (N /
2) −X0 _real (N / 2) × X0 _imag (N / 2), and when k = 1 to N / 2-1, X1 _real (k) = X0 _real (N / 2 + k) × X0
_real (N / 2-k) + X0 _imag (N / 2 + k) × X0
_imag (N / 2-k) X1 _imag (k) = X0 _imag (N / 2 + k) × X0
_real (N / 2-k) -X0 _real (N / 2 + k) × X0
_imag (N / 2-k).

The instantaneous frequency resolution of the Wigner distribution method is higher than that of the FFT method. FIG. 4A shows a spectrum obtained by the FFT method, and FIG. 4B shows a spectrum obtained by the Wigner distribution method, both of which are calculated from the same received signal data. FIG. 4 (a),
Comparing the spectra of (b), it can be seen that the Wigner distribution method is sharper than the FFT method.

Next, the extraction processing of the extraction processing section 22 will be specifically described with reference to FIG. In this extraction processing, first, spectrum data c ′ of a predetermined frequency portion of each spectrum data c obtained by the frequency analysis processing unit 21 is sequentially extracted. Here, the predetermined frequency portion is defined as the predicted maximum value of the Doppler frequency generated by the relative speed between the target and the sound source, and the integral value to be described later, in the sweep frequency width FW of the LFM wave centering on the center frequency of the sweep frequency of the LMF wave. The frequency portion (FW + 2) to which the margin frequency width λ added with the frequency width required by the processing unit 23 is added.
λ). Next, from the extracted spectrum data c ′, the time (PW + τ) obtained by adding the integration time τ3 in the integration processing unit 23 to be described later to the pulse length PW of the LFM wave.
3) A number of pieces of spectral data corresponding to minutes are extracted while repeatedly shifting by time τ2, and “frequency-amplitude-
3D data of "time" is created.

As shown in FIG. 6, the integration processing section 23 performs integration processing in the inclination direction of the sweep frequency line in the frequency width w1 centered on the sweep frequency line from the three-dimensional data extracted by the extraction processing section 22. Integration time τ along the sweep frequency line
Step 3 is repeated to create integrated data in which the features of the echo components are strengthened and the features of the non-echo components are weakened.

The normalization processing section 24 performs normalization processing according to the following equation so that the maximum value of the integral data becomes 1, and creates data for input to the neural network. This neural network input data is input to the neural network processing unit 3.

A ′ _ij = (A _ij −Min) / (Max−Min) Here, A indicates an integrated data level, and subscripts i and j indicate time and frequency, respectively. Min and Max indicate the minimum level and the maximum level of the integrated data, respectively.
A ′ indicates the level of the input data for the neural network.

FIG. 7A is an example of neural network input data obtained by the Wigner distribution method described above, and FIG. 7B is an example of neural network input data obtained by the conventional FFT method. Both are calculated from the same received signal data. As can be seen from these comparisons, the Wigner distribution method is
Compared with the FT method, the feature of the echo component is strong and the feature of the non-echo component is weak.

The neural network processing unit 3 receives the neural network input data created by the normalization processing unit 24, and uses the neural network that has learned the characteristics of echo and reverberation to generate the neural network input data. A distinction is made as to whether the data is echo or reverberation. For the neural network, for example, the above-described layered model can be applied.Also, for learning of the neural network, a well-known method, for example, the learning in the above-described JP-A-8-152472 can be applied. Can be. As an example, FIG. 8 shows an example in which the learning described in that publication is applied.

In FIG. 8, the neural network has a hierarchical structure consisting of three layers, an input layer, an intermediate layer, and an output layer, and all the units of each layer are mutually connected between the layers. In the learning of the neural network, first, the input data for the neural network is input to each unit of the input layer, and the output value of the output layer is calculated based on the input data. Subsequently, a difference between the output value and a teacher signal indicating the pattern input to the input layer is calculated as an error. Then, the weight between units and the threshold between units are updated based on this error. This series of processing is repeated until the error becomes a value equal to or less than a certain threshold value, and the optimum values of the weight and the threshold value are determined. This learning process is performed in advance.

The neural network processing unit 3 inputs the neural network input data created by the preprocessing unit 2 to the neural network whose weight and threshold are determined by the learning, calculates the output value, and calculates the calculation result. Is determined based on the result. This determination is made by determining which output signal is closer to the teacher signal used in the learning process.

(Other Embodiments) In the above embodiment, the Wigner distribution method is used for the frequency analysis processing, but an AR model estimation method can be used instead of the Wigner distribution method. Here, the AR model estimating method is a spectrum estimating method using a well-known AR model (also called an autoregressive model), which is one of the time series models.

FIG. 9 is a block diagram showing a main configuration of a target identification device according to another embodiment of the present invention. This target identification device has the same configuration as that shown in FIG. 1 described above, except that an AR model estimation processing unit 21 ′ is provided instead of the frequency analysis processing unit 21. 9, the same components as those shown in FIG. 1 are denoted by the same reference numerals, and the detailed description thereof will be omitted.

The AR model estimation processing section 21 ′ converts the reception signal (time series signal) received by the transmission / reception
Are sequentially extracted while repeatedly shifting the received signals by the time τ1, and each of the extracted received signals (time-series signals) is subjected to frequency conversion processing by an AR model estimation method to obtain spectrum data. Based on the spectrum data, a series of processes in the above-described extraction processing unit 22, integration processing unit 23, and normalization processing unit 24 are performed, and neural network input data is created.

In each of the embodiments described above, the Wigner distribution method and the AR model distribution estimation method have been described as examples of the method of analyzing the frequency of a received signal. However, the present invention is not limited to this. What kind of configuration can be applied if it is possible to create neural network input data that enhances the characteristics of the echo component that is the reflected wave from the target and weakens the characteristics of the non-echo component You may.

[0035]

As described above, according to the present invention, it is possible to create neural network input data in which the difference between an echo and a reverberation is made more prominent. Reduction can be achieved.

In addition, for an echo having a low S / N ratio,
Neural network input data in which the difference between the echo and the reverberation is clear can be created, so that the target identification accuracy can be improved with respect to the low S / N ratio echo, and the low S / N ratio echo can be generated. The detection rate can be further increased.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a main configuration of a target identification device according to an embodiment of the present invention.

FIG. 2 is a schematic diagram illustrating a frequency analysis process performed by a frequency analysis processing unit illustrated in FIG. 1;

FIG. 3 is a schematic diagram showing a calculation image of the Wigner distribution method.

4A is a diagram showing a spectrum obtained by the FFT method, and FIG. 4B is a diagram showing a spectrum obtained by the Wigner distribution method.

FIG. 5 is a schematic diagram for explaining an extraction process performed by an extraction processing unit shown in FIG. 1;

FIG. 6 is a schematic diagram for explaining an integration process performed by the integration processing unit shown in FIG. 1;

FIG. 7A is a diagram showing an example of neural network input data obtained by the Wigner distribution method;
(B) is a diagram showing an example of neural network input data obtained by the FFT method.

8 is a block diagram showing an example in which a learning function described in Japanese Patent Application Laid-Open No. 8-152472 is applied to the target identification device shown in FIG.

FIG. 9 is a block diagram illustrating a main configuration of a target identification device according to another embodiment of the present invention.

FIG. 10 is a block diagram showing a main configuration of a sonar signal processing device described in Japanese Patent Application Laid-Open No. 5-297114.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Transmitting and receiving part 2 Preprocessing part 3 Neural network processing part 21 Frequency analysis processing part 21 'AR model estimation processing part 22 Extraction processing part 23 Integral processing part 24 Normalization processing part

Claims (6)

[Claims] 1. A transmitting / receiving means for transmitting a frequency-modulated sound wave and receiving a reflected wave from a target, and a frequency analysis of a signal received by the transmitting / receiving means,
Preprocessing means for enhancing the characteristics of the echo component which is the reflected wave from the target, and creating the neural network input data in which the characteristics of the non-echo components are weakened; and inputting the neural network input data A target identification device comprising: a predetermined neural network; and neural network processing means for determining whether the input data of the neural network is an echo or a non-echo from an output of the neural network. 2. The target discriminating apparatus according to claim 1, wherein the frequency analysis is performed by a Wigner distribution method or an AR model distribution estimation method. 3. The pre-processing means extracts a reception signal having a predetermined time length from the signal received in time series by the transmission / reception means while repeatedly shifting the reception signal by a predetermined time, and obtains a spectrum of the extracted reception signal. Frequency analysis processing means for respectively obtaining data, From each of the spectrum data obtained by the frequency analysis processing means, of the frequency portion obtained by adding a required margin to the sweep frequency width of the transmission wave transmitted from the transmission and reception means Along with extracting the spectrum data in time series, from the extracted spectrum data, a group of spectrum data corresponding to a predetermined time is extracted while repeatedly shifting by a predetermined time, and the frequency, spectrum amplitude, Extraction to create three-dimensional data with time on each axis The three-dimensional data created by the extraction means is integrated by sequentially performing integral processing of a required time in a direction of inclination along the swept frequency line with a required frequency width centered on the swept frequency line. 2. The target identification apparatus according to claim 1, comprising: integration processing means for obtaining data; and normalization means for normalizing the integrated data and outputting the integrated data as neural network input data. Transmitting a frequency-modulated sound wave and receiving a reflected wave from the target; and analyzing the received signal by frequency to enhance characteristics of an echo component that is a reflected wave from the target. And generating neural network input data with reduced characteristics of non-echo components, and identifying whether the neural network input data is an echo or a non-echo using a predetermined neural network. And at least a target identification method. 5. The target identification method according to claim 4, wherein a Wigner distribution method or an AR model distribution estimation method is used as the frequency analysis. 6. The neural network input data generating step includes: extracting a received signal of a predetermined time length from the received signal while repeatedly shifting the received signal by a predetermined time; and obtaining respective spectral data of the extracted received signal. From each of the spectral data, spectral data of a frequency portion obtained by adding a required margin to the sweep frequency width of the transmission wave is extracted in a time series, and spectral data corresponding to a predetermined time is extracted from the extracted spectral data. Extracting the group while repeatedly shifting the group by a predetermined time, and creating three-dimensional data with frequency, spectrum amplitude, and time on each axis for each of the extracted spectral data groups; Required circumference around A step of sequentially performing integral processing of a required time in a direction of inclination along a swept frequency line at a wave number width to obtain integrated data; anda step of normalizing the integrated data and outputting this as data for neural network input. The target identification method according to claim 4, comprising: