JP2000180258A - Discrimination device for vessel type of sailing body and method thereof - Google Patents

Abstract

PROBLEM TO BE SOLVED: To automatically discriminate the vessel type of a sailing body from a hull acoustic signal. SOLUTION: The device is constituted by providing an acoustic signal receiver 1 for receiving a hull acoustic signal, a FFT processing means 2 for carrying out a frequency analysis, a memory 3 for storing frequency analysis data, a power spectrum calculating means 4 for calculating the power with regard to the respective frequencies from the frequency analysis data in the prescribed frequency band, a symbolic coding means 5 for obtaining the coding signal percentage value for each prescribed frequency band from both the entire energy of the power spectrum in a prescribed frequency band and the energy of power spectrum for each prescribed frequency band divided the prescribed frequency band by N and for producing the code row based on the coding signal percentage value, and a neural network processing means 6 for performing the neural network processing and for outputting the vessel type of the sailing body.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a code sequence of a coded signal including characteristics of a noise signal peculiar to a vehicle from a hull acoustic signal transmitted in water (including the sea) generated by the vehicle. And the neural network operation processing is performed using the coded signal of this code string as input data of a multilayered neural network by backpropagation learning to automatically identify the type of ship. The present invention relates to a body type identification device and a method thereof.

[0002]

2. Description of the Related Art The noise sources of a hull sailing on or underwater consist of mechanical noise, propeller noise and hydrodynamic noise. For example, FIG. 12 shows a frequency-power spectrum analysis diagram of an example of a hull acoustic signal generated by the hull, where a circle indicates a propeller noise caused by a propeller, and a circle indicates mechanical noise caused by an engine. Stuff,
The circles marked with ○ indicate mechanical noise caused by the refrigerator or cooler.

The type of propeller mounted on the hull, ie, the type of screw, is determined by the type of ship, and the shape of the screw is about 20 types.

A conventional method of receiving a hull acoustic signal generated by a hull and discriminating the type of the hull from the acoustic signal is a mechanical method that removes radiation noise of the hull, that is, hydrodynamic noise. By performing FFT (Fast Fourier Transform) processing on the radiated noise of the vehicle based on the noise and propeller noise and creating a loafgram, an experienced person who has been trained specially for a long time can see the sound pattern. The ship's ship type.

[0005]

However, even if a person who has been trained specially for a long period of time as in the prior art is identified, the identification of the ship type may be incorrect depending on the physical condition and environment at that time or the waveform of the loafgram. May occur.

In addition, it is desired to automatically identify the type of a moving vehicle approaching a device installed underwater and to output a signal from the device only to a predetermined type of ship. The case exists.

The present invention has been made in view of the above points. Applicant has reported that the radiating noise of the vehicle is composed of mechanical noise and propeller noise, and the mechanical noise caused by uneven rotation of engines, reduction gears, generators, various pumps, etc. Changes in sound pressure and frequency, and propeller noise is due to cavitation and vibration characteristics of the propeller.Cavitation noise is a continuous frequency component spectrum.The sound pressure decreases as the cruising depth decreases and the speed increases. While the level increases, noise due to the vibration characteristics determined by the shape and material of the propeller has a specific frequency component spectrum, and the frequency distribution does not change even if the propeller rotation speed increases or decreases. In addition, the frequency spectrum distribution is almost constant according to the type of ship, and the noise Appears as decrease of the noise Minamotooto pressure, causing a decrease of the power spectrum of several local maxima in the frequency spectrum distribution, however in the frequency spectrum distribution, the frequency band of the various noise was confirmed by experiments that does not change.

The present invention pays attention to these facts and, when identifying a ship type of a hulling vehicle by multi-layered neural network arithmetic processing by error back propagation learning, uses a hull acoustic signal to analyze a noise signal specific to the ship type. A code carrier of a coded signal including the features of the above, and the coded signal of the code sequence is used as input data to perform a neural network operation to automatically identify the type of the cruising ship. It is an object to provide a species identification device and a method thereof.

[0009]

SUMMARY OF THE INVENTION In order to solve the above-mentioned object, a vehicle type identification apparatus according to the present invention receives a hull acoustic signal generated by a vehicle and detects the acoustic signal of the vehicle from the acoustic signal. In a marine vessel type identification apparatus for identifying a type of ship, an acoustic signal receiver for receiving a hull acoustic signal, FFT processing means for frequency-analyzing the acoustic signal received by the acoustic signal receiver,
A memory for storing the frequency analysis data processed by the T processing means, a power spectrum calculation means for reading frequency analysis data of a predetermined frequency band from the memory and calculating a power for each frequency; From the total energy of the power spectrum and the energy of the power spectrum for each predetermined frequency width obtained by dividing the predetermined frequency band into N, a coded signal percentage value for each predetermined frequency width is obtained, and a code sequence based on the coded signal percentage value is obtained. And a neural network processing means for performing neural network arithmetic processing and outputting the type of the hull as a coded signal of a code string generated by the coding means as input data, and Automatic identification of ship type based on acoustic signal received by signal receiver It is characterized by a door.

The encoding means converts the percentage value of the coded signal into a value of a predetermined significant digit and generates a code string of the numerical value of the significant digit, or the encoding means outputs the coded signal percentage value to the encoding means. A case is provided in which a comparison unit for comparing with a predetermined threshold value is provided, and the coded signal percentage value is binarized to generate a code string.

The method for identifying a ship type of a vehicle according to the present invention comprises the steps of: receiving an acoustic signal of a hull generated by the vehicle and identifying a ship type from the acoustic signal; Based on the received hull acoustic signal of the hull, the frequency analysis is performed by the FFT processing means, and the power for each frequency in a predetermined frequency band among the frequencies processed by the FFT processing means is calculated, The predetermined frequency band is divided into N predetermined frequency widths, and the energy of the power spectrum of the predetermined frequency band is obtained, and the total energy of the power spectrum of the predetermined frequency band is obtained, and the total energy of the power spectrum of the predetermined frequency band is obtained. The coded signal percentage value indicating the ratio of the energy of the power spectrum of the predetermined frequency width to the predetermined frequency width is obtained for each of the predetermined frequency widths, A code sequence based on the signal percentage value is generated, and a coded signal of the generated code sequence is used as input data to perform neural network operation processing for identifying a ship type of the vehicle by a neural network processing means, and receive an acoustic signal. It is characterized in that the type of the ship is automatically identified from the acoustic signal received by the vessel.

On the basis of the received hull sound signal of the hull,
In a predetermined frequency band, a code sequence of a coded signal including characteristics of radiation noise peculiar to the vehicle is generated. By using the coded signal of this code string as input data and performing a neural network operation using a neural network processing means that has been learned in advance, a ship type of a vehicle based on the specific radiation noise is automatically identified. You.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows the configuration of an embodiment of a vehicle type identification apparatus according to the present invention.

In FIG. 1, a hull acoustic signal of a marine vessel traveling above or below the water surface is received by an acoustic signal receiver 1 such as a hydrophone. The acoustic signal received by the acoustic signal receiver 1 is digitized by the A / D converter, and then input to the FFT processing means 2 for frequency analysis. As the FFT processing means 2, a conventional well-known FFT processing means is used, and the resulting frequency analysis data is stored in the memory 3. Although the FFT processing means 2 has been described as being digital, the acoustic signal received by the acoustic signal receiver 1 is processed by the analog FFT processing means 2, digitized, and stored in the memory 3. May be.

The power spectrum calculation means 4 reads out frequency analysis data of a predetermined frequency band from the memory 3 in order from a lower frequency side to a higher frequency, and obtains a power spectrum P (i) for each frequency f (i). calculate.

The encoding means 5 comprises a predetermined frequency width energy calculating section 7, a memory 8, an integrating section 9, and a dividing section 10, and the predetermined frequency band read from the memory 3 and calculated by the power spectrum calculating means 4 % Of the power spectrum of the power spectrum of the predetermined frequency band and the total energy of the power spectrum of the predetermined frequency band obtained by dividing the predetermined frequency band into N. And a code sequence is generated based on the coded signal percentage value.

That is, based on the power spectrum P (i) input from the power spectrum calculating means 4, the predetermined frequency width energy calculating section 7 divides the power spectrum by an energy 所 定 P (C
Find i).

That is, the predetermined frequency width energy calculator 7
Is the energy ΔP from the power spectrum of the lowest frequency sequentially output by the power spectrum calculation means 4 to the power spectrum of the frequency having the predetermined frequency width W.
(C1) is obtained by calculation.

The energy ΔP (C
1) is stored in the memory 8 and input to the accumulator 9. The predetermined frequency width W input by the integrating unit 14
Is integrated at this time, and since this is the first time, the energy ΔP of this predetermined frequency width W is reduced to zero.
(C1) is integrated, and the integration unit 14 integrates ΣP (C1).

This repetition is executed for the above-mentioned predetermined frequency band, and the last predetermined frequency width W
When the energy ΔP (CN) is obtained by the predetermined frequency width energy calculating unit 7 and stored in the memory 8 and integrated by the integrating unit 9, the memory 8 stores the N obtained by the predetermined frequency width energy calculating unit 7. Predetermined frequency width W
Are stored in a predetermined order, respectively, and the total energy ΔP for the predetermined frequency band is obtained by the integrating unit 9.

The dividing unit 10 stores the data of the total energy ΔP of the integrating unit 9 and the predetermined frequency width W read from the memory 8 in order from the lower frequency side, for example.
From the data of the energy ΔP (Ci) of the coded signal, a percentage (energy ratio) value ΔP (C
i) Perform an operation to obtain / ΣP. At this time, in obtaining the coded signal percentage value, the division unit 10 outputs a numerical value of a predetermined significant digit. For example, when the significant digit is 1, the percentage value of the coded signal is, for example, “0”, 0.01 from 0.00 to less than 0.01%.
Output as "1" from% to less than 0.02%, "2" from 0.02% to less than 0.03%, "9" from 0.09% to less than 0.10% I do. In this way, a code sequence based on the coded signal percentage value is generated.

That is, the division unit 15 calculates the data of the total energy ΔP from the integrating unit 9 and the energy ΔP (C1) of the predetermined frequency width W having the lowest frequency from the memory 8.
, The encoded signal percentage value is obtained, and the numerical value of the significant digit described above is output. The value of this significant digit is input as input data to the # 1 input terminal of the neural network processing means 6 learned by the error back propagation learning method described later.

Similarly, the divider 10 reads out the data of the energy ΣP (C2) having the next lower predetermined frequency width W from the memory 8 and obtains the coded signal percentage value. Output a digit number. The value of this significant digit is obtained by
The data is input to the input terminal as input data.

The encoding means 5 performs the remaining N
-2 times over the predetermined frequency band, it is possible to generate a code sequence based on the coded signal percentage value. An example of a code string of the coded signal percentage value having one significant digit is shown in FIG. 3 (I) (FIG. 3).
(The predetermined frequency width-encoded signal percentage diagram of (II) does not always correspond one-to-one.)

Then, the coded signal percentage value of the number of significant digits of the code string generated by the coding means 5 is subjected to neural network arithmetic processing to perform the # 1 through # 1 processing of the neural network processing means 6 for identifying the type of the vehicle. ♯N are input to the input terminals as input data.

As the neural network processing means 6, a conventionally known one is used, and the number of {1 to ΔN inputs corresponding to the number of the coded signal percentage values of the code string input from the coding means 5 is used. It has a terminal and has a neural network of a multilayer structure based on error back propagation learning that has been learned in advance. Then, a neural network operation is performed based on the input data of the numerical value of the significant digit of the coded signal percentage value input from the coding means 5, and an identification signal for designating a ship type is output.

FIG. 2 shows an embodiment of the neural network processing means used in the present invention.

In FIG. 1, the neural network processing means 6 includes an input unit 21 having N input terminals # 1 to #N, seven output units 22 for outputting the ship types 1 to 7, and a hidden unit. It is composed of a three-layered neural network of 16 intermediate units 23 as units. However, there is only a link from the input unit 21 to the output unit 22, and there is no reverse link.
The weight of each link is determined by learning in advance.
Since the learning method is performed by a conventional method, the description is omitted.

A specific example of encoding of neural network input data and identification of a ship type using the input data as input in the vehicle type identification apparatus and method of the present invention having the above-described configuration will be described below. I do.

FIG. 4 is a diagram for explaining an embodiment of the ship type identification.

In the figure, (I) is a hull sound signal of the hull detected by the sound signal receiver 1. The FFT processing means 2 performs FFT processing on the acoustic signal of the hull acoustic signal at a certain time point T1, and determines a predetermined frequency band from F ₀ Hz to F ₁ (F ₁ > F ₀ ) from the frequency analysis data.
The power spectrum shown in FIG.

With respect to this power spectrum, a predetermined frequency band W = (F ₁ −F ₀ ) / N Hz width is obtained by dividing a predetermined frequency band F ₀ Hz to F ₁ Hz into N (N columns).
The coded signal percentage value ΣP (Ci) / ΣP described above is obtained, and the code string of the significant digits of the 50 coded signal percentage values ΣP (Ci) / ΣP as shown in FIG. Partial display) is generated.

The numerical values of the N significant digits of this code string are shown in FIG.
Are input as input data to the corresponding # 1 to #N input units 21 of the neural network processing means 6 shown in FIG. The neural network operation is performed by the neural network having a three-layer structure based on the backpropagation learning that has been trained in advance, so that the neural network processing means 6 outputs “1” representing the identification signal for designating the ship type. That is, from the output unit 22 (IV)
A pattern as shown is output. As a result, the ship type 1 is obtained from the hull acoustic signal of the marine vessel detected by the acoustic signal receiver 1.
Is identified and determined to be “Vessel A”.

FIGS. 5 and 6 show another embodiment of the ship type identification for different ship types, and are the same as FIG.

5 and 6, (I) is a hull acoustic signal of the hull detected by the acoustic signal receiver 1, (II) is a frequency power spectrum at a certain point in time,
FIG. 3 (III) shows a code string of numerical values of significant digits of a coded signal percentage value generated by the coding means 5 and input to the neural network processing means 6, and FIG. 4 (IV) shows an output pattern of the neural network processing means 6. In FIG. 5, the neural network processing means 6 identifies the ship type 2 and determines that the ship type is “Vessel B”. In FIG. 6, the neural network processing means 6 identifies the ship type 7 and calls "ship G".
Is determined. Since the procedure for identifying the ship type 2 and the ship type 7 is the same as that in the case of FIG. 4, the description is omitted.

FIG. 7 shows another embodiment of the encoding means.

In the figure, the same components as those of the encoding means 5 shown in FIG. 1 are denoted by the same reference numerals, and the encoding means 5 of FIG. 7 is provided between the division unit 10 and the neural network processing means 6. Is characterized in that a comparison unit 11 is provided.

That is, a predetermined threshold value is input to the comparison unit 11, and the comparison unit 11 encodes the threshold value and the encoding for each of the predetermined frequency widths input from the division unit 10. The signal percentage value ΔP (Ci) / ΔP is compared. The coded signal percentage value at this time {P (Ci) /}
P is a division value of the coded signal percentage value itself, different from the output of the division unit 10 in FIG.

The comparing section 11 compares the coded signal percentage value ΔP (Ci) / ΔP for each of the predetermined frequency widths with a threshold value. This coded signal percentage value {P (Ci) /
When ΣP is larger than a predetermined threshold value, for example, “1”, when the coded signal percentage value ΣP (Ci) / ΣP is smaller than a predetermined threshold value, coding is performed as “0”. . That is, the coded signal percentage value ΣP (Ci) / ΣP for each of the predetermined frequency widths is converted into a binary signal according to the magnitude. This binarized signal is input as input data to the ♯i input unit 21 of the neural network processing means 6 learned by the error back propagation learning method described above.

The operation after the division unit 10 will be briefly described.
The data of P and the data of the energy ΔP (C1) of the predetermined frequency width W having the lowest frequency from the memory 8 are read, and a coded signal percentage value ΔP (C1) / ΔP is obtained.
The coded signal percentage value ΣP (C1) / ΣP is compared with the comparison unit 1
Send to 1. The comparing unit 11 compares the threshold value with the coded signal percentage value ΣP (C1) / ΣP, and determines “1” or “0” according to the magnitude of the coded signal percentage value ΣP (C1) / ΣP. Is converted into a binary signal. This binarized signal is input to the # 1 input unit 21 of the neural network processing means 6 as input data.

Similarly, the dividing unit 10 reads out the data of the energy ΔP (C2) having the next lower predetermined frequency width W from the memory 8 and calculates the encoded signal percentage value ΔP (C2) / ΔP. The comparison unit 11 compares the coded signal percentage value ΣP (C2) / ΣP with the above threshold value and converts the coded signal into a binary signal. This binarized signal is input to the # 2 unit 21 of the neural network processing means 6 as input data.

The coding means 5 can generate a binary code string based on the coded signal percentage value for the predetermined frequency band by repeating the same processing. The binarized signal of the code string generated by the encoding means 5 is input to the # 1 to #N input units 21 of the neural network processing means 6 for performing the neural network operation processing and identifying the type of the vehicle. Is entered as FIG. 8 shows an example of this binary code string.
(I) (predetermined frequency width of FIG. 8 (II) −
(It does not necessarily correspond one-to-one to the coded signal percentage diagram).

At this time, a conventionally known neural network processing means 6 is used, and the number of {1 to #N input signals corresponding to the number of binary signals of the code string input from the coding means 5 is used. It has a terminal, that is, a ♯1 to ♯N input unit 21 and has a multi-layered neural network based on error backpropagation learning previously learned. Then, a neural network operation is performed based on the input data of the binary signal pattern input from the encoding means 5, and an identification signal for designating the type of ship is output.

FIG. 9 is a diagram for explaining an embodiment of the ship type identification using the encoding means of FIG.

In the figure, (I) is a hull sound signal of the hull detected by the sound signal receiver 1. The FFT processing means 2 performs FFT processing on the acoustic signal of the hull acoustic signal at a certain point in time T1, and obtains a predetermined frequency band F ₀ Hz to F ₁ Hz from the frequency analysis data in FIG.
The power spectrum shown in I) is obtained.

With respect to this power spectrum, a predetermined frequency band W = (F ₁ −F ₀ ) / N Hz width is obtained by dividing a predetermined frequency band F ₀ Hz to F ₁ Hz into N (N columns).
The coded signal percentage value ΣP (Ci) / ΣP described above is obtained, and a code string (N) of N coded signal percentage values ΣP (Ci) / ２P binary signal as shown in FIG. Partial display) is generated.

The N binary signals of this code string are input as input data to the corresponding # 1 to #N input units 21 of the neural network processing means 6 shown in FIG. The neural network operation is performed by a neural network having a three-layer structure based on error backpropagation learning that has been learned in advance.
Will output. That is, the output unit 22 outputs a pattern as shown in (IV). As a result, the ship type 1 is identified from the hull acoustic signal of the hull body detected by the acoustic signal receiver 1, and is determined to be “Vessel A”.

FIGS. 10 and 11 show another embodiment of the ship type identification for different ship types, and are the same as FIG.

10 and 11, (I) shows the hull acoustic signal of the hull body detected by the acoustic signal receiver 1, and (II) shows the frequency power spectrum at a certain point in time. (III) shows a binary code string of a coded signal percentage value generated by the coding means 5 and input to the neural network processing means 6, and (IV) shows an output pattern of the neural network processing means 6, respectively. In FIG. 10, the neural network processing means 6 identifies the ship type 2 and determines that the ship type is "Ship B". In Figure 11, ship type 7
Is identified by the neural network processing means 6, and "G
Ship ". Since the procedure for identifying the ship type 2 and the ship type 7 is the same as that in the case of FIG. 9, the description is omitted.

In the above description, the ship type 1 to ship type 7 are all specified. However, the neural network processing means 6 is set in advance such that the ship type 7 is set to "no applicable ship type". By learning, it is possible to discriminate between a target marine vehicle and a non-target marine vehicle for discriminating a ship type, and it is possible to improve a discrimination rate. This processing depends on how the neural network processing means 6 learns.

The series of ship type identification processing is repeated every time S required for the ship type identification processing. Then, based on the type of ship identified and output at each time S, for example, a type of ship that has been output the most number of times within a predetermined time, or a type of output type of ship having a probability equal to or more than a predetermined number is selected. By providing an identification improvement processing means such as selecting the same ship type when the same ship type is identified continuously for a predetermined number of times, the identification rate of ship type identification can be improved.

As described above, since the type of the hull can be automatically identified from the acoustic signal received by the acoustic signal receiver 1 instead of the human being, the hull type identification apparatus or method according to the present invention is provided. If the previously installed equipment is installed in the water in advance, the equipment will automatically identify the approaching ship type and output signals from the equipment only for the predetermined target ship type. Can be. In addition, if a device for detecting the right and left of the traveling body approaching the device is provided, a signal can be generated when the traveling body of the target ship type passes right and left of the device.

Further, each configuration of the encoding means 5 shown in FIG. 1 is not limited to the configuration shown in FIG. 1, but shows one embodiment, and the encoding means 5 having this configuration is shown. It goes without saying that any processing method or circuit configuration may be used as long as it has a function equivalent to that of. Also, the encoding of the encoding means 5 exemplifies the one which has obtained a good result by an experiment, and is not limited to the above-described one, but may employ an appropriate code conversion.

According to the present invention, since the type of ship can be automatically identified, it is possible to monitor whether a regular passenger ship or the like is operating on a regular basis in a harbor. Demonstrate.

[0055]

As described above, according to the present invention, based on a received hull acoustic signal of a hull, a code string including characteristics of radiation noise peculiar to the hull is generated.
Since each signal of this code string is used as input data and the neural network processing is performed using neural network processing means that has been learned in advance, the ship type of the vehicle is automatically identified.

In identifying the type of the vehicle, the encoding signal percentage value of claim 2 is set to a predetermined number of significant digits, and the encoding means is adapted to generate a code string of the number of significant digits. In such a case, the same type of ship can be identified to a more detailed level. On the other hand, a comparison unit for comparing the coded signal percentage value of claim 3 with a predetermined threshold value is provided, and the coded signal percentage value is calculated. In the case of the coding means for generating a code string by binarization, there is an advantage that a rough ship type can be identified.

Further, the identification rate of the type of ship can be improved by providing an identification improvement processing means which is repeated every predetermined time S and is based on the number of identified types of ships to be identified and outputted.

Since the ship type of the vehicle can be automatically identified from the acoustic signal received by the acoustic signal receiver, the vehicle type identification apparatus or the apparatus provided with the method according to the present invention is installed in water in advance. If so, the device can automatically identify the type of the approaching hull, and can output a signal from the device only to the hull of the predetermined target type. .

[Brief description of the drawings]

FIG. 1 shows the configuration of an embodiment of a vehicle type identification apparatus according to the present invention.

FIG. 2 shows an embodiment of a neural network processing means used in the present invention.

FIG. 3 is an explanatory view of an example of a code string of a vehicle having a coded signal percentage (energy ratio).

FIG. 4 is an explanatory view of one embodiment of a ship type identification.

FIG. 5 is an explanatory view of another embodiment of the ship type identification for different ship types.

FIG. 6 is an explanatory view of another embodiment of the ship type identification for different ship types.

FIG. 7 shows another embodiment of the encoding means.

FIG. 8 is an explanatory diagram of an example of a binarized code sequence of a vehicle having a coded signal percentage (energy ratio).

FIG. 9 is an explanatory view of one embodiment of a ship type identification using the encoding means of FIG. 7;

FIG. 10 is an explanatory view of another embodiment of the ship type identification for different ship types.

FIG. 11 is an explanatory view of another embodiment of the ship type identification for different ship types.

FIG. 12 is a frequency-power spectrum analysis diagram of an example of a hull acoustic signal generated by the hull.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Acoustic signal receiver 2 FFT processing means 3 Memory 4 Power spectrum calculation means 5 Encoding means 6 Neural network processing means 7 Predetermined frequency width energy calculation means 8 Memory 9 Accumulation unit 10 Division unit 11 Comparison unit

────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] October 21, 1999 (1999.10.
21)

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] Claims

[Correction method] Change

[Correction contents]

[Claims]

[Procedure amendment 2]

[Document name to be amended] Statement

[Correction target item name] 0009

[Correction method] Change

[Correction contents]

[0009]

SUMMARY OF THE INVENTION In order to solve the above-mentioned object, a vehicle type identification apparatus according to the present invention has a unique structure for the vehicle itself.
The hull acoustic signal to be generated in the hull type identification device for identifying the type of the hull from the acoustic signal,
An acoustic signal receiver for receiving a hull acoustic signal, an FFT processing means for frequency-analyzing the acoustic signal received by the acoustic signal receiver, a memory for storing frequency analysis data processed by the FFT processing means, and Power spectrum calculating means for reading out frequency analysis data of a predetermined frequency band and calculating power for each frequency; for every predetermined frequency width obtained by dividing the total energy of the power spectrum of the predetermined frequency band and the predetermined frequency band by N Encoding means for determining a coded signal percentage value for each predetermined frequency width from the energy of the power spectrum, and generating a code string based on the coded signal percentage value, and a code string generated by the coding means. Using the coded signal as input data, perform neural network arithmetic processing and output the type of ship A chromatography neural network processing means, independently generate the Wataru Hashikarada acoustic signal wave receiver has received
It is characterized in that so as to automatically identify the acoustic signal the ship species Kou Hashikarada that.

[Procedure amendment 3]

[Document name to be amended] Statement

[Correction target item name] 0011

[Correction method] Change

[Correction contents]

The method for identifying a ship type of a ship according to the present invention receives a sound signal of a ship body generated independently by the ship itself and identifies the ship type of the ship by the sound signal. In the method, based on the received hull acoustic signal of the hull,
The frequency analysis is performed by the FFT processing means, the power for each frequency of a predetermined predetermined frequency band among the frequencies processed by the FFT processing means is calculated, and the predetermined frequency band is divided into N predetermined frequency widths. Calculating the energy of the power spectrum of the predetermined frequency band, obtaining the total energy of the power spectrum of the predetermined frequency band, and indicating the ratio of the energy of the power spectrum of the predetermined frequency width to the total energy of the power spectrum of the predetermined frequency band. A coded signal percentage value is determined for each of the predetermined frequency widths, a code sequence based on the coded signal percentage value is generated, and the coded signal of the generated code sequence is used as input data by the neural network processing means by the neural network processing means. Neural network operation to identify the type of ship Own occurrence of the coastal Hashikarada but received
It is characterized in that the type of ship is automatically identified from the acoustic signal to be made.

[Procedure amendment 4]

[Document name to be amended] Statement

[Correction target item name] 0019

[Correction method] Change

[Correction contents]

The energy ΔP (C
1) is stored in the memory 8 and input to the accumulator 9. The integrator 9 integrates the input energy ΔP (Ci) of the predetermined frequency width W. Since the energy ΔP (Ci) is the first in this case, the energy ΔP (P) of the predetermined frequency width W is set to 0.
(C1) is integrated, and the integration unit 9 integrates ΣP (C1).

[Procedure amendment 5]

[Document name to be amended] Statement

[Correction target item name] 0032

[Correction method] Change

[Correction contents]

With respect to this power spectrum, a predetermined frequency band W = (F ₁ −F ₀ ) / N Hz width is obtained by dividing a predetermined frequency band F ₀ Hz to F ₁ Hz into N (N columns).
The coded signal percentage value ΣP (Ci) / ΣP described above is obtained, and the code string of the significant digits of the N coded signal percentage values ΣP (Ci) / ΣP as shown in FIG. Partial display) is generated.

[Procedure amendment]

[Submission Date] January 25, 2000 (2000.1.2
5)

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] Claims

[Correction method] Change

[Correction contents]

[Claims]

[Procedure amendment 2]

[Document name to be amended] Statement

[Correction target item name] 0009

[Correction method] Change

[Correction contents]

[0009]

SUMMARY OF THE INVENTION In order to solve the above-mentioned object, a vehicle type identification apparatus according to the present invention receives a hull acoustic signal independently generated by a marine vessel, and uses the acoustic signal to perform navigation. In a vehicle type identification device for identifying a type of a vehicle,
An acoustic signal receiver for receiving a hull acoustic signal, an FFT processing means for frequency-analyzing the acoustic signal received by the acoustic signal receiver, a memory for storing frequency analysis data processed by the FFT processing means, and Power spectrum calculating means for reading out frequency analysis data of a predetermined frequency band and calculating power for each frequency; for every predetermined frequency width obtained by dividing the total energy of the power spectrum of the predetermined frequency band and the predetermined frequency band by N From the energy of the power spectrum, an energy percentage value for each predetermined frequency width is obtained, an encoding unit that generates a code sequence based on the energy percentage value, and an encoding pattern of the code sequence generated by the encoding unit. Performs neural network arithmetic processing and outputs the type of ship as input data And a neural network processing means, is characterized in that so as to automatically identify the ship type of domestic Hashikarada from the acoustic signal is Wataru Hashikarada itself acoustic signal wave receiver has received is independently generated.

[Procedure amendment 3]

[Document name to be amended] Statement

[Correction target item name] 0010

[Correction method] Change

[Correction contents]

The encoding means converts the energy percentage value into a predetermined number of significant digits and generates a code string of the significant number, or the encoding means outputs the energy percentage value to the encoding means. A case is provided in which a comparison unit for comparing with a predetermined threshold value is provided, and the energy percentage value is binarized to generate a code string.

[Procedure amendment 4]

[Document name to be amended] Statement

[Correction target item name] 0011

[Correction method] Change

[Correction contents]

The method for identifying a ship type of a ship according to the present invention receives a sound signal of a ship body generated independently by the ship itself and identifies the ship type of the ship by the sound signal. In the method, based on the received hull acoustic signal of the hull,
The frequency analysis is performed by the FFT processing means, the power for each frequency of a predetermined predetermined frequency band among the frequencies processed by the FFT processing means is calculated, and the predetermined frequency band is divided into N predetermined frequency widths. Calculating the energy of the power spectrum of the predetermined frequency band, obtaining the total energy of the power spectrum of the predetermined frequency band, and indicating the ratio of the energy of the power spectrum of the predetermined frequency width to the total energy of the power spectrum of the predetermined frequency band. An energy percentage value is obtained for each of the predetermined frequency widths, a code string based on the energy percentage value is generated, and a coding pattern of the generated code string is used as input data by the neural network processing means by the neural network processing means. Performs neural network operation processing to identify the ship type of Is characterized in that No. receivers is adapted to automatically identify the ship type of domestic Hashikarada from the acoustic signal which uniquely generated Wataru Hashikarada itself received.

[Procedure amendment 5]

[Document name to be amended] Statement

[Correction target item name] 0016

[Correction method] Change

[Correction contents]

The encoding means 5 comprises a predetermined frequency width energy calculating section 7, a memory 8, an integrating section 9, and a dividing section 10, and the predetermined frequency band read from the memory 3 and calculated by the power spectrum calculating means 4 From the energy of the power spectrum of each predetermined frequency band obtained by dividing the predetermined frequency band into N and the total energy of the power spectrum of the predetermined frequency band , an energy percentage value representing the ratio for each of the predetermined frequency bands.
(Hereinafter referred to as “encoded signal percentage value”), and a code sequence is generated based on the encoded signal percentage value.

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Yoshiyuki Nomoto 7-12-12 Tomioka Nishi, Kanazawa-ku, Yokohama-shi, Kanagawa Prefecture (72) Inventor Kunihiko Yoshimi 2-29-1, Heian-cho, Tsurumi-ku, Yokohama-shi, Kanagawa 2G064 AA14 AB16 AB21 BA02 BC37 CC29 CC42 CC43 CC47 DD29

Claims (4)

[Claims] A marine vessel type identification device for receiving a hull acoustic signal generated by a marine vessel and identifying a hull type of the marine vessel from the acoustic signal. F, which analyzes the frequency of the acoustic signal received by the acoustic signal receiver
FT processing means, a memory for storing frequency analysis data processed by the FFT processing means, power spectrum calculation means for reading frequency analysis data of a predetermined frequency band predetermined from the memory, and calculating power for each frequency; From the total energy of the power spectrum of the predetermined frequency band and the energy of the power spectrum for each predetermined frequency width obtained by dividing the predetermined frequency band into N, a coded signal percentage value for each predetermined frequency width is obtained, and the coded signal percentage value is obtained. Encoding means for generating a code string based on the above, and a neural network processing means for performing a neural network operation process and outputting a type of a vehicle as a neural network by using an encoded signal of the code string generated by the encoding means as input data From the acoustic signal received by the acoustic signal receiver. A hull type identification device characterized by automatic identification. 2. The apparatus according to claim 1, wherein said encoding means sets the percentage value of the encoded signal to a value of a predetermined significant digit, and generates a code string of the value of the significant digit. Vessel type identification device. 3. The coding means includes a comparing unit for comparing the coded signal percentage value with a predetermined threshold value, and generates a code string by binarizing the coded signal percentage value. 2. The vehicle type identification device according to claim 1, wherein: 4. A hull body type identification method for receiving a hull acoustic signal generated by a hull body and identifying the type of the hull body from the acoustic signal. Based on the above, a frequency analysis is performed by the FFT processing means, and a power for each frequency of a predetermined frequency band among the frequencies processed by the FFT processing means is calculated, and the predetermined frequency band is divided into N predetermined frequencies. Divided into widths, the energy of the power spectrum of the predetermined frequency band is obtained, and the total energy of the power spectrum of the predetermined frequency band is obtained. The energy of the power spectrum of the predetermined frequency width with respect to the total energy of the power spectrum of the predetermined frequency band A coded signal percentage value indicating the ratio of the coded signal is determined for each of the predetermined frequency widths, and a code sequence based on the coded signal percentage value is calculated. Generates coded signals of the generated code string as input data, performs neural network operation processing for identifying the type of hull by neural network processing means, and sails from the acoustic signal received by the acoustic signal receiver. A method for identifying the type of a moving ship, wherein the type of the ship is automatically identified.