JP2020034382A - Signal processor, signal processing method, and program of signal processing method - Google Patents

Abstract

To obtain a signal processor, etc., with which it is possible to perform more accurate identity determination.SOLUTION: Provided is a signal processor 100 for performing an identity determination process of determining whether or not the same signal is included in a plurality of signals x. The signal processor 100 comprises: a feature extraction unit 111 for extracting multiple types of feature quantity from the plurality of signals x; a dimensionless unit 112 for rendering the multiple types of feature quantity dimensionless; a feature correlation unit 113 for calculating, with respect to the correlation of multiple types of dimensionless feature quantity, a feature correlation value that indicates linearity between the multiple types of feature quantity; a correlation determination unit 114 for determining, on the basis of the feature correlation value, whether or not the same signal is included; a feature angle correlation unit 115 for calculating a feature angle correlation value that represents the inclination of distribution in the multiple types of feature quantity; and a feature angle correlation determination unit 116 for determining, on the basis of the feature angle correlation value, whether or not the same signal is included.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a signal processing device, a signal processing method, and a program for a signal processing method. In particular, the present invention relates to identity determination for determining whether a plurality of signals include the same signal.

  For example, there is a signal processing device that processes a signal such as sound received by a receiver to specify a target, estimate a direction of the target, and the like. Here, in the signal processing device, identity determination processing is performed to determine whether the same signal is included in the signals received by the plurality of receivers, for example, from the same sound source (for example, And Patent Document 1).

JP 2003-139856 A

Here, the identity determination process by calculation using the cross-correlation function will be described. Assuming that the signals output from the receiver a and the receiver b are x _a [n] and x _b [n], respectively, the cross-correlation functions r _{xa, xb} [ τ] is represented by equation (1).

When the same signal is included in the signal x _a [n] and the signal x _b [n], the cross-correlation function r _{xa, xb} [τ] is the arrival time difference between the signals at the receiver a and the receiver b. Has a peak in the sample corresponding to Then, when a threshold value γ _r that can be regarded as the same signal is set by the cross-correlation, the identity can be determined by the equation (2). Here, max is a function that outputs the maximum value among the input samples. Flag _r is a flag indicating the result of the identity determination.

As shown in Expression (2), when it is determined that the same signal is included in the signal x _a [n] and the signal x _b [n], flag _r = 1. If it is determined that the same signal is not included in x _b [n], flag _r = 0. As described above, by using the cross-correlation function, it is possible to perform the identity determination processing for determining whether or not the signals output from the receiver a and the receiver b include the same signal.

  However, in the above-described identity determination processing using the cross-correlation function, a peak value is obtained due to the integration effect as shown in Expression (1). Therefore, it is considered that when the duration of the common signal is long, the identity determination can be performed as expected. On the other hand, when a common signal having a relatively short duration is a target, the determination is performed with a small integrated value, so that the accuracy of the identity determination is significantly reduced.

  Therefore, it has been desired to realize a signal processing device, a signal processing method, and a program for a signal processing method that can perform more accurate identity determination.

  A signal processing device according to the present invention is a signal processing device that performs an identity determination process of determining whether or not the same signal is included in a plurality of signals. The feature extraction unit to be extracted, the dimensionless unit for dimensionlessly converting a plurality of types of feature amounts, and the correlation between the plurality of types of nondimensionalized features, the linearity between the plurality of types of feature amounts. A feature correlation unit that calculates a feature correlation value to be indicated, a correlation determination unit that determines whether the same signal is included based on the feature correlation value, and a feature angle correlation value that represents a gradient of a distribution in a plurality of types of feature amounts And a characteristic angle correlation determining unit that determines whether the same signal is included based on the characteristic angle correlation value.

  Also, the signal processing method according to the present invention includes a feature extraction step of extracting a plurality of types of feature amounts from a plurality of signals, a dimensionless step of non-dimensionalizing the plurality of types of feature amounts, respectively. A feature correlation value calculating step of calculating a feature correlation value indicating linearity between the plurality of types of feature amounts, and determining whether the same signal is included based on the feature correlation values. A feature angle correlation value calculating step of calculating a feature angle correlation value representing a gradient of a distribution in a plurality of types of feature amounts when the same signal is temporarily determined to be included; A characteristic angle correlation determining step of determining whether or not the same signal is included based on the correlation value.

  Also, the program of the signal processing method according to the present invention includes a feature extraction step of extracting a plurality of types of feature amounts from a plurality of signals, a dimensionless step of dimensionlessly converting the plurality of types of feature amounts, respectively. A feature correlation value calculating step of calculating a feature correlation value indicating linearity between the plurality of types of feature amounts, and the same signal is included based on the feature correlation values. A correlation determination step of temporarily determining whether or not, when temporarily determining that the same signal is included, a feature angle correlation value calculation step of calculating a feature angle correlation value representing the gradient of the distribution in a plurality of types of feature amounts; And a feature angle correlation determining step of determining whether or not the same signal is included based on the feature angle correlation value.

  According to the present invention, when both the feature correlation value calculated by the feature correlation unit and the feature angle correlation value calculated by the feature angle correlation unit based on a plurality of types of features in the signal satisfy the condition, the same signal is output. Are included by the feature correlation unit and the feature angle correlation unit, so that more accurate identity determination can be performed.

FIG. 1 is a diagram illustrating a configuration of a signal processing system centered on a signal processing device 100 according to Embodiment 1 of the present invention. FIG. 3 is a diagram showing a relationship between a plurality of receivers 10 and a sound source according to Embodiment 1 of the present invention. FIG. 3 is a diagram for explaining feature data stored in a feature database 121 according to Embodiment 1 of the present invention. FIG. 3 is a diagram showing a configuration of a feature angle correlation unit 115 according to Embodiment 1 of the present invention. FIG. 3 is a diagram for explaining distribution of characteristics of a signal x according to Embodiment 1 of the present invention. FIG. 9 is a diagram illustrating a configuration of a signal processing system mainly including a signal processing device 100 according to Embodiment 2 of the present invention. FIG. 9 is a diagram for describing effects according to the second embodiment of the present invention.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. Here, in the following drawings, components denoted by the same reference numerals are the same or equivalent, and are common to the entire text of the embodiments described below. In addition, the forms of the components shown in the entire text of the specification are merely examples, and are not limited to these descriptions. In particular, the combination of components is not limited to the combination in each embodiment, and the components described in other embodiments can be appropriately applied to another embodiment. In addition, when it is not necessary to particularly distinguish or specify a plurality of devices of the same type that are distinguished by subscripts, the subscripts and the like may be omitted.

Embodiment 1 FIG.
FIG. 1 is a diagram showing a configuration of a signal processing system mainly including a signal processing device 100 according to Embodiment 1 of the present invention. FIG. 2 is a diagram illustrating a relationship between the plurality of receivers 10 and the sound source according to Embodiment 1 of the present invention. In FIG. 2, two sound sources respectively emit signals. The receiver 10 converts a received signal into an electric signal. Here, it is assumed that the receiver 10 receives a signal related to sound. In the first embodiment, two receivers 10a and 10b are provided. In the signal processing device 100 according to the first embodiment, the sound signals received by the receivers 10a and 10b each include the same signal with the same specifications, such as a signal emitted from the same sound source. A description will be given as an apparatus for determining whether or not the operation is performed.

  The signal processing device 100 according to the first embodiment includes an arithmetic processing device 110 and a storage device 120. The arithmetic processing device 110 according to the first embodiment is a device that performs identity determination processing based on signals transmitted from the plurality of receivers 10. The arithmetic processing device 110 includes a feature extraction unit 111, a dimensionless unit 112, a feature correlation unit 113, a correlation determination unit 114, a feature angle correlation unit 115, and a feature angle correlation determination unit 116. The storage device 120 is a device that stores data related to the processing of the arithmetic processing device 110. The storage device 120 has a feature database 121.

The feature extraction unit 111 (111a, 111b) of the arithmetic processing unit 110 performs a feature extraction step of processing a plurality of signals from the receiver 10 and extracting a plurality of feature amounts _fk extracted. Then, the feature extracting unit 111 outputs a feature vector f which is a column vector having the feature amount f _k as an element. Here, the characteristic amount f _k is a numerical value of each of a plurality of types of characteristics of the signal.

The dimensionless unit 112 (112a, 112b) performs a dimensionless step of dimensionlessizing each feature amount f _k in the feature vector f based on the feature data stored in the feature database 121. The feature data will be described later. Then, a dimensionless feature vector f _ND having the dimensionless feature quantity f _NDk as an element is output. Here, in FIG. 1, the feature extraction units 111a and 111b and the dimensionless units 112a and 112b are provided according to the number of the receivers 10, but the invention is not limited to this. is not. One feature extraction unit 111 and dimensionless unit 112 may process a plurality of signals.

The feature correlation unit 113 performs a feature correlation value calculation step of calculating and outputting a feature correlation value r _vector based on the plurality of dimensionless feature vectors f _ND sent from each dimensionless unit 112. The feature correlation value r _vector is an index for determining whether or not a plurality of signals include the same signal. The correlation determination unit 114 performs a temporary determination process based on the feature correlation value r _vector from the feature correlation unit 113 and performs a correlation determination step of outputting a temporary determination result flag _vector .

The feature angle correlation unit 115 determines the feature angle correlation based on the gradient of the distribution of each feature amount f _NDk of the dimensionless feature vector f _ND relating to a plurality of signals provisionally determined to include the same signal in the determination result flag _vector . A feature angle correlation value calculation step of calculating and outputting the value r _angle is performed. The characteristic angle correlation value r _angle is an index for determining whether a plurality of signals include the same signal. The feature angle correlation determination unit 116 performs a determination process based on the feature angle correlation value r _angle from the feature angle correlation unit 115, and performs a feature angle correlation determination step of outputting a determination result flag _feature .

  Here, the arithmetic processing device 110 is generally configured by a device that performs control arithmetic processing such as a computer centered on a CPU (Central Processing Unit), for example. Then, the arithmetic processing device 110 implements a program of the procedure of the signal processing method performed by each unit in advance, and implements the processing of each unit. Here, for example, the storage device 120 has program data. However, the present invention is not limited to this, and each unit may be separately configured by a dedicated device (hardware).

The storage device 120 stores data related to the processing of the arithmetic processing device 110, as described above. The feature database 121 of the storage device 120 is a database that stores feature data used in the processing performed by the dimensionless unit 112. The feature data is data relating to the value of the feature value _fk of each item of a signal regarding a non-stationary sound of a sample obtained in advance, with a plurality of features as items. Details will be described later. The storage device 120 includes a volatile storage device (not shown) such as a random access memory (RAM) that can temporarily store data and a nonvolatile auxiliary storage device (such as a hard disk and a flash memory that can store data for a long time). (Not shown).

The processing of each unit described above will be described in more detail. Here, the signal from the receivers 10a and _{x a.} The signal from the receiver 10b is _xb . The feature extraction unit 111 outputs a feature vector f representing a plurality of feature quantities f _k obtained from the signals x _a and x _b as a column vector. The feature amount f _k and the feature vector f can be represented by Expression (3). Here, k is a number assigned to the feature. K is the number of feature types to be used. G _k (x) is a function for calculating the feature quantity of the number k when the signal x output from the receiver 10 is input.

For example, as a function for calculating the characteristic amount f _k relating to frequency, a function which can be calculated using Fourier transform is a function for calculating a frequency at which energy peaks, a spread of energy regarding frequency, a center of gravity of energy regarding frequency, and the like. Can be adopted. In addition, as a function for calculating a time-related feature amount, a function for calculating a signal duration, a time at which energy peaks, a spread of energy over time, and the like can be employed. As a function for calculating a feature related to the distribution of energy values, a function for calculating a moment such as skewness and kurtosis can be adopted. As described above, various functions for extracting a feature amount related to a signal can be adopted, and a plurality of feature amounts f _k obtained based on the adopted function are used. Here, these feature values f _k are hardly affected by the SN ratio. The feature vector f obtained from the signal _{x a} and _{f a.} Further, the feature vector f obtained from the signal _{x b} and _{f b.} Then, a description will be given identity determination using the similarity of the features vectors f _a and the feature vector f _b.

FIG. 3 is a diagram illustrating feature data stored in feature database 121 according to Embodiment 1 of the present invention. In the feature database 121, as shown in FIG. 3, values of the feature amount _fk of the signal relating to the unsteady sound of the sample obtained in the past are arranged and stored as feature data. At this time, the type of the signal is clear. The dimensionless unit 112 uses the feature data stored in the feature database 121 to make the feature vector f dimensionless for each element. First, for all the feature vectors f stored in the feature database 121, the average value μ _k and the standard deviation σ _k of the feature amount f _k for each element are calculated as shown in Expressions (4) and (5). I do.

Further, the dimensionless unit 112 outputs a dimensionless feature vector f _ND represented by Expression (6) using the average value μ _k and the standard deviation σ _k of the feature amount f _k . In the first embodiment, the non-dimensional unit 112 outputs the dimensionless feature vector _{f NDa} and dimensionless feature vector _{f NDb} corresponding to the feature vector _{f a} and the feature vector _{f b.}

The feature correlation unit 113 outputs a feature correlation value r _vector representing a correlation between signals x of each feature based on the dimensionless feature vector f _NDa and the dimensionless feature vector f _NDb . The feature correlation value r _vector represented by Expression (7) is an index indicating linearity between a plurality of features.

The correlation determination unit 114 performs a determination process based on the feature correlation value r _vector from the feature correlation unit 113 as shown in Expression (8), and outputs a determination result flag _vector . Here, γ _vector is a threshold value at which it is considered that the same signal is included in the signal x of each receiver 10.

The feature angle correlation unit 115 receives the determination result flag _vector determined by the correlation determination unit 114, the dimensionless feature vector f _NDa, and the dimensionless feature vector f _NDb . When the determination result flag _vector = 0, the characteristic angle correlation unit 115 outputs only the determination result flag _vector because the same sound is not included in the signal.

On the other hand, when the determination result flag _vector = 1, the characteristic angle correlation unit 115 outputs the determination result flag _vector and the characteristic angle correlation value r _angle . Here, the feature angle correlation value r _angle is an index indicating the similarity between feature vectors.

As shown in FIG. 5A described later, the characteristic angle correlation value r _angle is an index related to the inclination when each characteristic amount is distributed. The correlation determination unit 114 determines the identity again, in order to determine, as different signals, signals that are regarded as the same even though they are signals of different sounds.

When obtaining the feature angle correlation value r _angle , it is necessary to calculate the inclination of the distribution of each feature f _k . As a means for calculating the gradient of the distribution, a method such as principal component analysis or the least squares method is used. Here, a method of calculating a gradient by performing principal component analysis using eigenvalue decomposition will be described.

  FIG. 4 is a diagram showing a configuration of the feature angle correlator 115 according to Embodiment 1 of the present invention. The feature angle correlation unit 115 further includes a correlation matrix calculation unit 115A, an eigenvalue decomposition unit 115B, and a feature angle correlation value calculation unit 115C. The correlation matrix calculation unit 115A calculates a correlation matrix Σ represented by Expression (9).

Eigenvalue decomposition unit 115B, as shown in equation (10), decomposed into a correlation matrix sigma, eigenvectors φ (φ _1, φ ₂₎ and eigenvalue lambda (lambda _1, lambda ₂₎ and.

Wherein the angle correlation value calculation unit 115C, based on the eigenvector phi _1, to calculate the characteristic angles correlation value _{r angle} of formula (11).

The characteristic angle correlation determination unit 116 performs a determination process based on the characteristic angle correlation value r _angle from the characteristic angle correlation unit 115 as shown in Expression (12), and outputs a determination result flag _feature . Here, γ _angle is a threshold value at which the signal x from each receiver 10 is considered to include the same signal. The determination result flag _feature is a flag indicating the result of the identity determination. If it is determined that the signal _{x a} and the signal _{x b} is the signal of the same sound is included, the determination result _{flag Description feature} = 1, if it is determined that no signal is included in the same sound, the determination result flag _{It is} assumed that _feature = 0.

FIG. 5 is a diagram for explaining a distribution of features of the signal x according to the first embodiment of the present invention. Next, effects of the signal processing device 100 according to the first embodiment will be described. Receivers 10a and receivers 10b is, when the reception of the same signal, extracts a signal _{x a} and the signal from the _{x b,} the dimensionless feature vector _{f NDa} and dimensionless feature vector _{f NDb} the dimensionless , The feature amount f _{NDa, k} and the feature amount f _{NDb, k} having the same index k have similar values. Therefore, as shown in FIG. 5A, each feature is distributed near a straight line having a slope of 1.

The feature correlation value r _vector calculated by the feature correlation unit 113 is an index indicating the linearity of the feature distribution. When the same signal is included in the signals received by each of the receivers 10, the feature correlation value r _vector becomes a large value. On the other hand, if the same signal is not included in the signals received by the respective receivers 10 and the signals are different, the characteristic correlation value r _vector becomes a small value. Therefore, the correlation determination unit 114 can perform the identity determination by performing the determination on the characteristic correlation value r _{vector based} on the set threshold γ _vector .

Here, as shown in FIG. 5B, even when the straight line derived from the distribution based on the feature has a slope different from 1, if the distribution is linear, the feature correlation value r _vector becomes a large value. Therefore, it is determined that the signals are the same even though they are different signals.

On the other hand, the feature angle correlation value r _angle calculated by the feature angle correlation unit 115 is an index indicating the angle between the slope of the feature distribution and the straight line having the slope 1. When the same signal is included in the signals received by each of the receivers 10, the angle between the slope of the feature distribution and the straight line having the slope 1 is small (the slope of the feature distribution is close to 1). , The feature angle correlation value r _angle is a large value. On the other hand, if the same signal is not included in the signals received by the respective receivers 10 and the signals are different, the characteristic angle correlation value r _angle is a small value. Therefore, the feature angle correlation determination unit 116 performs the determination on the feature angle correlation value r _{angle based} on the set threshold γ _angle to prevent the different signals in the feature determination from being erroneously determined to be the same. it can.

Here, even when the signals received by the respective receivers 10 are different, there is a case where the characteristic angle correlation value r _angle is a large value. For example, if the two signals are uncorrelated, the features are randomly distributed as shown in FIG. For this reason, the inclination is also random, and it may be determined that the signals are the same, even though they are different signals.

Therefore, in the signal processing device 100 of the first embodiment, the feature correlation unit 113 and the feature angle correlation unit 115 calculate both the feature correlation value r _vector and the feature angle correlation value r _angle . Then, when the correlation determination unit 114 and the feature angle correlation determination unit 116 determine that there is a linear relationship between the dimensionless feature vectors f _ND related to the plurality of signals and that the slope of the correlation characteristic is close to 1, It is determined that the signals are the same. Therefore, it is possible to more accurately determine the identity of the signal.

Embodiment 2 FIG.
The signal processing device 100 according to the first embodiment uses the feature amounts f _NDk of the dimensionless feature vector f _ND processed by the dimensionless unit 112 for determining the identity. Here, the effectiveness of the feature amount f _NDk of each feature differs depending on the type of the unsteady sound. If the feature amount f _NDk has _little variation, it is difficult to distinguish the feature f _NDk , so that it is not effective for determination. For example, when performing identity determination, for a signal x having energy in a wide frequency band, a feature amount obtained by quantifying a feature related to the spread of energy is effective. For a signal x having a short duration, a feature quantity obtained by quantifying a characteristic relating to the duration is effective.

  Therefore, the signal processing device 100 according to the second embodiment calculates in advance the validity of each feature from the variation of each feature stored in the feature database 121. A weight based on the validity is assigned to each feature value to determine the identity.

FIG. 6 is a diagram showing a configuration of a signal processing system centered on a signal processing device 100 according to Embodiment 2 of the present invention. 6, the same reference numerals as those in FIG. 1 perform the same operations as those described in the first embodiment. 6, the weighting unit 117 (117a, 117b) performs a process of weighting each element of the dimensionless feature vector f _ND obtained by dimensionlessly converting the signal received by each receiver 10. In addition, the feature database 121 according to the second embodiment stores data relating to weights for correcting variations related to the respective feature amounts f _k in association with the feature data.

Next, the weight assigned by the weight assigning unit 117 will be described. As for the weight, a weight vector w is calculated in advance from the feature data in the feature database 121 based on Expression (13). Here, α is an arbitrary coefficient. Further, when calculating the average value μ _k and the standard deviation σ _k of the feature value f _k stored as the feature data, only the signal of the same type as the signal received by the receiver 10 is used to obtain the signal. , The validity of the feature according to the type can be calculated.

The weight assigning unit 117 performs a process of assigning the weight vector w stored as data in the feature database 121 to the dimensionless feature vector f _ND using Expression (14). Here, ○ means an element product (Hadamard product).

FIG. 7 is a diagram for explaining an effect according to the second embodiment of the present invention. When all the features are treated equally and the identity determination is performed despite the large variation of a certain feature _fk , there is a possibility that an appropriate determination result cannot be obtained. FIG. 7A shows that although the same signal is received by each of the receivers 10, an appropriate determination result cannot be obtained due to the influence of handling the feature value f _k that is easily affected by the SN ratio. An example is shown.

Therefore, in the signal processing device 100 according to the second embodiment, the weighting unit 117 weights the dimensionless feature vector f _ND in accordance with the validity of each feature f _k , thereby obtaining the weights shown in FIG. As shown in b), the distribution of the feature amount f _k can be corrected. Therefore, the accuracy of the identity determination can be improved.

Embodiment 3 FIG.
Although not particularly shown in the above-described first and second embodiments, the signal processing device 100 performs processing on the signals received by each of the receivers 10 in the receiver array having the plurality of receivers 10. May perform a process of determining the identity.

  In the above-described first and second embodiments, identity determination processing is performed on signals received by a plurality of receivers 10 arranged at different positions. However, the present invention is not limited to this. Absent. The signal processing device 100 may perform processing for determining the identity of signals received at different times in a single receiver 10 or a receiver array having a plurality of receivers 10.

  In the first embodiment described above, the frequency-related features include the frequency at which the energy peaks, the spread of the energy related to the frequency, the center of gravity of the energy related to the frequency, and the like. In addition, as the characteristics relating to time, the duration of a signal, the time at which energy peaks, the spread of energy over time, and the like have been exemplified. Then, as a feature related to the distribution of the energy value, moments such as skewness and kurtosis are exemplified.

In addition, the feature value _fk can be adopted for various features. As for the frequency, a characteristic amount _fk characterized by frequency variation, the lowest frequency, the skewness of the energy distribution related to the frequency, the kurtosis of the energy distribution related to the frequency, and the like can be adopted. As for time, the feature amount f _k is characterized by the center of gravity of energy with respect to time, the skewness of energy distribution with respect to time, the kurtosis of energy distribution with respect to time, the rate of increase of energy with respect to time, the rate of decrease of energy with respect to time, and the like. Can be adopted. Other than these feature values f _k , it is possible to adopt features inherent in the signal received by the receiver 10.

In the first embodiment described above, the characteristic angle correlation unit 115 when calculating a characteristic angular correlation value r _angle, has been described a method using a principal component analysis and least squares method, the distribution of the feature amount f _k other than those A method of calculating the inclination can be used.

In the above-described first embodiment, after the identity determination based on the linearity using the feature correlation value r _vector by the correlation determination unit 114 is performed, the feature angle correlation value r _angle by the feature angle correlation determination unit 116 is used. Was determined. However, the order of the determination may be reversed.

10, 10a, 10b Receiver 100 Signal processing unit 110 Arithmetic processing unit 111, 111a, 111b Feature extraction unit 112, 112a, 112b Non-dimensionalization unit 113 Feature correlation unit 114 Correlation determination unit 115 Feature angle correlation unit 115A Correlation matrix calculation Unit 115B eigenvalue decomposition unit 115C feature angle correlation value calculation unit 116 feature angle correlation determination unit 117, 117a, 117b weighting unit 120 storage device 121 feature database

Claims (10)

A signal processing device that performs identity determination processing to determine whether the same signal is included in a plurality of signals,
A feature extraction unit that extracts a plurality of types of feature amounts from the plurality of signals;
A dimensionless unit for dimensionlessly converting the plurality of types of feature amounts,
A feature correlation unit that calculates a feature correlation value indicating linearity between the plurality of types of feature amounts with respect to the correlation between the plurality of types of feature amounts that have been made dimensionless,
A correlation determination unit that determines whether the same signal is included based on the feature correlation value,
A feature angle correlation unit that calculates a feature angle correlation value representing a gradient of a distribution in the plurality of types of feature amounts,
A signal processing device comprising: a feature angle correlation determination unit that determines whether or not the same signal is included based on the feature angle correlation value.   2. The feature extraction unit according to claim 1, wherein the feature extraction unit extracts one or a plurality of the feature amounts from the plurality of signals, among the feature amounts related to time, the feature amount related to the frequency, and the feature amount related to the moment of the energy value distribution. Signal processing device. A feature database that stores data related to the feature amount obtained by analyzing a signal of a sample obtained in advance as feature data,
The dimensionless section,
Based on the feature data stored in the feature database, calculate an average and a standard deviation of each feature amount, divide a difference between the value of the feature amount and the average by a standard deviation, and calculate the plurality of types of the features. 3. The signal processing device according to claim 1, wherein the amounts are made dimensionless.   The signal processing device according to claim 1, wherein the feature angle correlation unit calculates the feature angle correlation value using principal component analysis.   4. The signal processing device according to claim 1, wherein the feature angle correlation unit calculates the feature angle correlation value using a least squares method. 5.   The signal processing device according to any one of claims 1 to 5, further comprising: a weighting unit configured to weight each of the plurality of types of feature amounts that have been made dimensionless based on a distribution of the feature amounts. . A feature database that stores data related to the feature amount obtained by analyzing a signal of a sample obtained in advance as feature data,
7. The signal processing device according to claim 6, wherein the weight is calculated for each of the feature amounts from an average and a standard deviation of the feature amounts based on the stored feature data, and is associated with the feature data.   8. The feature angle correlation unit according to claim 1, wherein the feature angle correlation unit calculates the feature angle correlation value for the plurality of types of feature amounts determined by the correlation determination unit to include the same signal. The signal processing device according to claim 1. A feature extraction step of extracting a plurality of types of feature amounts from a plurality of signals;
A dimensionless step of dimensionlessly converting the plurality of types of feature quantities,
A feature correlation value calculating step of calculating a feature correlation value indicating linearity between the plurality of types of feature amounts, for the correlation between the plurality of types of feature amounts that have been made dimensionless;
A correlation determination step of temporarily determining whether the same signal is included based on the feature correlation value,
A feature angle correlation value calculation step of calculating a feature angle correlation value representing a gradient of a distribution in the plurality of types of feature amounts, when it is provisionally determined that the same signal is included,
A characteristic angle correlation determining step of determining whether or not the same signal is included based on the characteristic angle correlation value. A feature extraction step of extracting a plurality of types of feature amounts from a plurality of signals;
A dimensionless step of dimensionlessly converting the plurality of types of feature quantities,
A feature correlation value calculating step of calculating a feature correlation value indicating linearity between the plurality of types of feature amounts, for the correlation between the plurality of types of feature amounts that have been made dimensionless;
A correlation determination step of temporarily determining whether the same signal is included based on the feature correlation value,
A feature angle correlation value calculation step of calculating a feature angle correlation value representing a gradient of a distribution in the plurality of types of feature amounts, when it is provisionally determined that the same signal is included,
A program of a signal processing method for causing a computer to perform a feature angle correlation determination step of determining whether or not the same signal is included based on the feature angle correlation value.

Images