JP7615719B2 - Denseness estimation device and method - Google Patents

Description

The present invention relates to a density estimation device, a density estimation method, and a density estimation program for estimating the density of individuals.

A method has been proposed for using radar to measure the number of animals in a flock (see, for example, Patent Document 1). Patent Document 1 describes a method for measuring the number of birds in a flock using the reflection intensity of radar radio waves.

In addition, Non-Patent Document 1 shows a model of volume scattering of radar radio waves when targeting a forest.

JP 2016-90409 A

Yoshio Yamaguchi, Toshifumi Moriyama, Motoi Ishido, Hiroyoshi Yamada, “Four-Component Scattering Model for Polarimetric SAR Image Decomposition”, IEEE, IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 43, NO. 8, AUGUST 2005

The technology described in Patent Document 1 measures the number of birds in a flock using only the reflected strength of radio waves. Therefore, with the technology described in Patent Document 1, there is a possibility that objects other than the individual being measured (birds in the case of Patent Document 1) may be recognized as the individual being measured, and as a result, there is a possibility that the number of birds may not be measured accurately.

Here, the number of individuals per unit area is referred to as density.

The present invention aims to provide a density estimation device, a density estimation method, and a density estimation program that can accurately estimate the density of an individual of interest.

The density estimation device according to the present invention is characterized in that it comprises a difference calculation means for calculating a first difference which is the difference between the scattering power of double reflection scattering calculated using polarized radar data when a flock of individuals has occurred and the scattering power of double reflection scattering calculated using polarized radar data when a flock of individuals has not occurred, and a second difference which is the difference between the scattering power of volume scattering calculated using polarized radar data when a flock of individuals has occurred and the scattering power of volume scattering calculated using polarized radar data when a flock of individuals has not occurred, and an estimation means for estimating the density, which is the number of individuals per unit area, based on the first difference and the second difference.

The density estimation method according to the present invention is characterized in that a computer calculates a first difference which is the difference between the scattering power of double reflection scattering calculated using polarized radar data when a flock of individuals has occurred and the scattering power of double reflection scattering calculated using polarized radar data when a flock of individuals has not occurred, and a second difference which is the difference between the scattering power of volume scattering calculated using polarized radar data when a flock of individuals has occurred and the scattering power of volume scattering calculated using polarized radar data when a flock of individuals has not occurred, and estimates the density, which is the number of individuals per unit area, based on the first difference and the second difference.

The density estimation program according to the present invention is characterized in that it causes a computer to execute a difference calculation process that calculates a first difference which is the difference between the scattering power of double reflection scattering calculated using polarized radar data when a flock of individuals has occurred and the scattering power of double reflection scattering calculated using polarized radar data when a flock of individuals has not occurred, and a second difference which is the difference between the scattering power of volume scattering calculated using polarized radar data when a flock of individuals has occurred and the scattering power of volume scattering calculated using polarized radar data when a flock of individuals has not occurred, and an estimation process that estimates the density, which is the number of individuals per unit area, based on the first difference and the second difference.

The present invention makes it possible to accurately estimate the density of the individual of interest.

1 is a block diagram showing an example of a congestion estimation device according to an embodiment of the present invention. 1 is a schematic diagram showing an area irradiated with radio waves when a polarized radar transmits and receives radio waves once. FIG. 1 is a schematic diagram showing a large number of pixels spread out on a plane. FIG. 13 is a schematic diagram showing an example of pixel conversion. FIG. 1 is a schematic diagram showing a probability density function expressed by equation (8). FIG. 1 is a schematic diagram showing a probability density function expressed by equation (15). FIG. 13 is a schematic diagram showing which covariance matrix is used depending on the value of 10 log(<|S _VV | ² >/<|S _HH | ² >) when the individual is an animal. FIG. 13 is a schematic diagram showing a probability density function expressed by equation (18). FIG. 13 is a schematic diagram showing which covariance matrix is used depending on the value of 10 log(<|S _VV | ² >/<|S _HH | ² >) when the individual is a human being. 10 is a flowchart showing an example of a process flow when the density estimation device of the present embodiment learns a neural network. 10 is a flowchart showing an example of a process flow when the density estimation device of the present embodiment estimates the density of individuals belonging to a flock. 13 is a flowchart showing an example of the progress of a process for calculating a scattering power _Pd of double reflective scattering and a scattering power _Pv of volume scattering. 13 is a flowchart showing an example of the progress of a process for calculating a scattering power _Pd of double reflective scattering and a scattering power _Pv of volume scattering. 1 is a schematic block diagram showing an example of the configuration of a computer related to a congestion estimation device 10 according to an embodiment of the present invention. 1 is a block diagram showing an overview of a congestion degree estimation device according to the present invention;

The following describes an embodiment of the present invention with reference to the drawings.

Figure 1 is a block diagram showing an example of a density estimation device according to an embodiment of the present invention. The density estimation device 10 according to this embodiment includes a radar data acquisition unit 11, a density acquisition unit 12, a scattering model power decomposition unit 13, a learning unit 14, an estimation unit 15, a population map generation unit 17, and an output unit 18.

The radar data acquisition unit 11 acquires polarized radar data from a polarized radar (not shown) that uses multiple types of polarized waves. In the following explanation, an example will be given in which the multiple types of polarized waves are horizontal and vertical polarization. An example of such a polarized radar is a synthetic aperture radar that uses horizontal and vertical polarization. Synthetic aperture radars are mounted on moving objects such as aircraft, satellites, and drones. In addition, a polarized radar that uses horizontal and vertical polarization may be a ground-mounted polarized radar that can change the direction or position of irradiation of radio waves.

The radar data acquisition unit 11 is communicatively connected to the polarized radar and acquires polarized radar data from the polarized radar through communication. In the following description, the polarized radar data includes the transmission electric field strength of each of the horizontally polarized wave and the vertically polarized wave, and the reception electric field strength of each of the horizontally polarized wave and the vertically polarized wave.

Figure 2 is a schematic diagram showing the area irradiated with radio waves when the polarized radar transmits and receives radio waves once. Here, this area is referred to as "area A." When the polarized radar transmits and receives radio waves once, it transmits radio waves to the band-shaped area A shown in Figure 2 and receives the reflected waves in area A. In Figure 2, the areas shown as squares are areas that correspond to individual parts of area A. The areas that correspond to individual parts are, for example, approximately squares with one side measuring about 1 m to 10 m, depending on the type of polarized radar, etc. The time at which the polarized radar receives the reflected waves differs for each part of area A, so the received electric field strength differs for each part of area A. The transmitted electric field strength is also common to each part of area A.

The density estimation device 10 treats each part in area A as a pixel and estimates the density of individuals for each pixel. The number of pixels corresponding to each part in area A is, for example, several thousand. In addition, the polarized radar changes the irradiation position of the radio waves and provides the polarized radar data to the radar data acquisition unit 11 after each transmission and reception of radio waves or after all transmissions and receptions of radio waves have been completed, so that the density estimation device 10 can estimate the density of individuals for each of the many pixels spreading out in a plane as shown in FIG. 3.

The following describes an example in which the radar data acquisition unit 11 acquires polarized radar data relating to an area corresponding to one pixel and its surrounding pixels, and the density estimation device 10 performs processing on the area corresponding to that one pixel (hereinafter sometimes referred to as the area of interest).

In addition, in this embodiment, the individuals whose density is estimated are "humans" or "non-human animals."

Next, we will explain the processing of the elements of the density estimation device 10 (see Figure 1) when learning a learning model for estimating individual density.

When it is known that a group of individuals has occurred in the region of interest, the density acquisition unit 12 acquires the density in the region of interest at that time from another system that calculates density (hereinafter, simply referred to as another system; not shown). This density is training data used when training a learning model for estimating density, and can be said to be a known density.

The congestion acquisition unit 12 is, for example, connected to another system so that it can communicate with the system, and acquires the congestion level from the other system through communication.

The method by which the separate system calculates the density is not particularly limited. For example, each individual belonging to a flock occurring in the area of interest may carry a GPS (Global Positioning System) terminal, and the separate system may identify the number of individuals in the area of interest by counting the number of GPS terminals in the area of interest, and calculate the density from the number of individuals and the known area of the area of interest. Also, for example, the separate system may photograph the area of interest, count the number of individuals in the area of interest based on the resulting image, and calculate the density from the number of individuals and the known area of the area of interest. Since density is the number of individuals per unit area, it is calculated by dividing the number of individuals by the area of the area in which the flock of individuals occurs.

The radar data acquisition unit 11 also acquires polarized radar data for the area of interest and its surrounding areas from the polarized radar at the time when the above-mentioned swarm occurs.

The radar data acquisition unit 11 also acquires polarized radar data for the area of interest and its surrounding areas from the polarized radar at a time when it is known that no flocks of individuals have occurred in the area of interest.

The scattering model power decomposition unit 13 calculates the scattering power of double reflection scattering and the scattering power of volume scattering by performing scattering model power decomposition on the polarized radar data acquired by the radar data acquisition unit 11. The scattering model power decomposition unit 13 calculates the scattering power of double reflection scattering and the scattering power of volume scattering when a flock occurs in the region of interest, and the scattering power of double reflection scattering and the scattering power of volume scattering when a flock does not occur in the region of interest. The calculation method of these scattering powers will be described later. Furthermore, the scattering model power decomposition unit 13 calculates the difference between the scattering power of double reflection scattering when a flock occurs in the region of interest and the scattering power of double reflection scattering when a flock does not occur in the region of interest. Similarly, the scattering model power decomposition unit 13 calculates the difference between the scattering power of volume scattering when a flock occurs in the region of interest and the scattering power of volume scattering when a flock does not occur in the region of interest. Hereinafter, the difference between the scattering power of double reflection scattering when a flock occurs and the scattering power of double reflection scattering when a flock does not occur in the region of interest is referred to as the first difference. Furthermore, the difference between the scattering power of volume scattering when a school occurs and the scattering power of volume scattering when a school does not occur is referred to as the second difference.

The learning unit 14 learns a learning model for estimating the density by machine learning using a combination of the density acquired by the density acquisition unit 12 (known density) and the first difference and the second difference corresponding to that density. This learning model is a learning model for estimating the density when the first difference and the second difference are given.

In this embodiment, the learning model is described as a neural network. The number of layers of the neural network and the number of neurons in each layer are predetermined. However, the weight between any neuron in any layer and any neuron in the next layer is not determined. If the weight matrix whose elements are the weight between a neuron in the kth layer and a neuron in the next layer (the k+1th layer) is [W ^k ], [W ^k ] can be expressed as the following formula (1).

In the right-hand side of formula (1), w _ij ^k represents the weight between the j-th neuron in the k-th layer and the i-th neuron in the k+1-th layer. Furthermore, formula (1) shows a case where the number of neurons in the k-th layer is n _k and the number of neurons in the k+1-th layer is n _k+1 . In this case, the number of elements (weights) in the weight matrix is n _k ×n _k+1 .

The learning unit 14 learns the neural network (i.e., defines the neural network) by learning a weight matrix for each pair of adjacent layers of the neural network through machine learning using a combination of the density and the first and second differences corresponding to the density.

Next, we will explain the processing of the elements of the density estimation device 10 (see Figure 1) when estimating the density of individuals.

The radar data acquisition unit 11 acquires polarized radar data of a target area for which density is to be estimated at a time when it is known in advance that a flock of individuals does not occur in the target area. The scattering model power decomposition unit 13 then uses the polarized radar data to calculate the scattering power of double reflection scattering and the scattering power of volume scattering when a flock does not occur, and stores each of the scattering powers. The scattering power of double reflection scattering when a flock does not occur, which is calculated and stored in advance, is represented as P _d0 . Furthermore, the scattering power of volume scattering when a flock does not occur, which is calculated and stored in advance, is represented as P _v0 . The density estimation device 10 performs the above process for each area corresponding to each pixel.

The radar data acquisition unit 11 acquires polarized radar data at a time when a flock of individuals is generated and the density of the individuals is to be estimated. The scattering model power decomposition unit 13 then calculates the scattering power of double scattering and the scattering power of volume scattering at that time. The scattering power of double scattering is represented as _Pd1 , and the scattering power of volume scattering is represented as _Pv1 .

The scattering model power decomposition unit 13 calculates the first difference (the difference between the scattering power of double-reflection scattering when a school has occurred and the scattering power of double-reflection scattering when a school has not occurred) by the following formula (2), where _ΔPd represents the first difference.

ΔP _d = P _d1 - P _d0 ...(2)

Furthermore, the scattering model power decomposition unit 13 calculates a second difference (the difference between the scattering power of volume scattering when a school occurs and the scattering power of volume scattering when a school does not occur) by the following formula (3), where ΔP _v represents the second difference.

ΔP _v = P _v1 - P _v0 (3)

The estimation unit 15 estimates the density of individuals by applying the first difference ΔP _d and the second difference ΔP _v to the neural network trained in advance by the training unit 14 .

The density estimation device 10 performs the above process for each area corresponding to each pixel.

As a result, the density of individuals is estimated for each pixel.

The population map generating unit 17 generates a population map. The population map is information that shows the number of individuals for each area in a map form. Note that it is assumed that the density is estimated for each pixel. The population map generating unit 17 generates the population map by converting the density estimated for each pixel into the number of individuals. The conversion from density to number of individuals can be performed by multiplying the density by the area of the area corresponding to one pixel. It is assumed that the area of the area corresponding to one pixel is known.

Here, if the area corresponding to one pixel is an approximate square with a side of about 1 m to 10 m, one pixel in the population map may be divided too finely. In such a case, the population map generating unit 17 generates the population map by converting the pixel so as to combine multiple pixels into one pixel. In this case, the area of the area corresponding to one pixel after conversion is larger than the area corresponding to one pixel before conversion. FIG. 4 is a schematic diagram showing an example of pixel conversion. The upper part of FIG. 4 is a map showing the density of each pixel before conversion, and shows a map for 25 pixels. Here, the density of each pixel is expressed in a manner according to the density (for example, a pattern, or color, etc.). It is assumed that the 25 pixels shown in the upper part of FIG. 4 are converted into one pixel. In this example, it is assumed that the 25 pixels shown in the upper part of FIG. 4 are converted into one pixel in the first row and second column of the lower part of FIG. 4. In this case, the population map generating unit 17 calculates the average density of the 25 pixels before conversion (see the upper part of FIG. 4). The population map generating unit 17 then calculates the number of individuals for the pixel in the first row and second column of the lower part of Fig. 4 by multiplying the area of the region corresponding to one pixel after conversion (one pixel in the lower part of Fig. 4) by the average value. The population map generating unit 17 similarly calculates the number of individuals for each of the other pixels in the lower part of Fig. 4. The population map generating unit 17 then generates, as a population map, information that represents the number of individuals for each pixel after conversion (each pixel in the lower part of Fig. 4) in a manner corresponding to the number of individuals (for example, pattern, color, etc.).

The output unit 18 is an output device for outputting the population map. For example, the output unit 18 may be a display device. In this case, the population map generation unit 17 causes the display device (output unit 18) to display the generated population map (e.g., see the lower part of FIG. 4). Note that a display device is one example of the output unit 18, and the output unit 18 is not limited to a display device. For example, the output unit 18 may be a printer device, and the population map generation unit 17 may cause the printer device (output unit 18) to print the generated population map (e.g., see the lower part of FIG. 4).

Next, the process in which the scattering model power decomposition unit 13 calculates the scattering power of double reflection scattering and the scattering power of volume scattering will be described. Note that the radar data acquisition unit 11 is assumed to have acquired polarized radar data of the region of interest and its surrounding areas from a polarized radar.

If the received electric field strengths of the horizontally polarized waves and the vertically polarized ^waves are _{E Hs} and E _Vs , respectively, the received electric field strength of the polarized radar can be expressed as [E ^Hs , E _Vs ^{] T} . Note that ^T means transposition. Similarly, if the transmitted electric field strengths of the horizontally polarized ^waves and the vertically polarized waves are E _Ht ^and E _Vt , ^respectively , the transmitted electric field strength of the polarized radar can be expressed as [E _Ht ^, E _Vt ] ^T . The subscript H represents a horizontally polarized wave, and the subscript V represents a vertically polarized wave. This also applies to the other symbols shown below.

Here, the following equation (4) holds:

In equation (4), the first matrix on the right side is a scattering matrix. The scattering model power _{decomposition unit 13 obtains a scattering matrix from equation (4), the received electric field strength [E Hs, E Vs] T, and the transmitted electric field strength [E Ht, E Vt ] T. Furthermore} ^{, the scattering model power} decomposition unit 13 obtains a ^covariance matrix expressed as in equation (5) below from the scattering matrix.

Here, "<>" means the ensemble average of the pixel in question and its surrounding pixels. Also, "*" represents the complex conjugate.

The covariance matrix representing the surface scattering model is given by the following equation (6). Note that β in equation (6) is an unknown quantity.

The covariance matrix representing the model of two-times reflection scattering is given by the following formula (7). Note that α in formula (7) is an unknown quantity.

Next, we will explain the covariance matrix that represents the model of volume scattering. First, we will explain the covariance matrix that represents the model of volume scattering when the individual is an "animal other than a human (hereinafter simply referred to as an animal)." When the individual is an animal, it can be considered similar to the case where a group of individuals is a forest. This is because when the individual is an animal, the orientation of the dipole is considered to be close to random. The model of volume scattering in the case of a forest is described in Non-Patent Document 1.

If we consider a group of animals to be similar to a forest, with the dipoles pointing in random directions, the probability density function of the dipole orientation as seen by the polarimetric radar is expressed by equation (8).

Figure 5 is a schematic diagram showing the probability density function expressed by equation (8). Figure 5 shows that the probability of a dipole's existence is constant regardless of the dipole's orientation.

The scattering matrix rotated by θ is expressed as the following equation (9).

The covariance matrix representing the volume scattering model is a 3-row, 3-column matrix with 9 elements. Using equation (9), the 9 elements are expressed as shown in equation (10) below.

In formula (10), Re( ) is a function that extracts the real part of a complex number. Furthermore, a, b, c, and I _i (i: 1, 2, ..., 10) in formula (10) are expressed as formula (11) shown below.

Using equations (8), (10) and (11), a covariance matrix representing a volume scattering model is obtained. Here, the scattering matrix of a dipole oriented in the horizontal direction is written as [S] _h-dipole ^HV , and the scattering matrix of a dipole oriented in the vertical direction is written as [S] _v-dipole ^HV . [S] _h-dipole ^HV and [S] _v-dipole ^HV are expressed as the following equations (12) and (13), respectively.

The covariance matrix when [S] _h-dipole ^HV shown in formula (12) is substituted into the right side of the first formula of formula (11), and the covariance matrix when [S] _v-dipole ^HV shown in formula (13) is substituted into the right side of the first formula of formula (11) are obtained. Here, even if [S] _h-dipole ^HV is substituted into the right side of the first formula of formula (11), or even if [S] _v-dipole ^HV is substituted into the right side of the first formula of formula (11), the same covariance matrix is obtained. The covariance matrix can be expressed as formula (14) shown below, and can be said to be a covariance matrix representing a model of volume scattering.

As mentioned above, let us think of a group of animals as being similar to a forest. In actual forests, the dominant structure is one in which dipoles are oriented vertically, and Non-Patent Document 1 presents a more accurate probability density function of a forest (probability density function of the dipole orientation). This probability density function is expressed as shown in Equation (15) below.

Figure 6 is a schematic diagram showing the probability density function expressed by equation (15). Figure 6 shows that the probability of the existence of a dipole pointing upward is higher than the probability of the existence of a dipole pointing in any other direction.

A covariance matrix representing a model of volume scattering when the probability density function represented by formula (15) is used is also obtained. In this case, a covariance matrix representing a model of volume scattering is obtained using formula (15), formula (10), and formula (11). As in the above case, a covariance matrix when [S] _h-dipole ^HV shown in formula (12) is substituted into the right side of the first formula of formula (11), and a covariance matrix when [S] _v-dipole ^HV shown in formula (13) is substituted into the right side of the first formula of formula (11) are obtained. When [S] _h-dipole ^HV is substituted into the right side of the first formula of formula (11), a covariance matrix of formula (16) shown below is obtained, and when [S] _v-dipole ^HV is substituted into the right side of the first formula of formula (11), a covariance matrix of formula (17) shown below is obtained.

The matrices shown in formula (14), formula (16), and formula (17) are all covariance matrices that represent a model of volume scattering when the individual is an animal. However, which covariance matrix is used is determined according to the value of 10log(<|S _VV | ² >/<|S _HH | ² >). FIG. 7 is a schematic diagram showing which covariance matrix is used according to the value of 10log(<|S _VV | ² >/<|S _HH | ² >) when the individual is an animal. That is, when 10log(<|S _VV | ² >/<|S _HH | ² >) is less than −2 dB, the covariance matrix shown in formula (17) is used. Furthermore, when 10log(<|S _VV | ² >/<|S _HH | ² >) is between −2 dB and 2 dB, the covariance matrix shown in formula (14) is used. When 10log(<|S _VV | ² >/<|S _HH | ² >) is greater than 2 dB, the covariance matrix shown in formula (16) is used.

Next, we will explain the covariance matrix that represents the model of volume scattering of a human flock when the individuals are humans. In this case, too, the covariance matrix (covariance matrix shown in formula (14)) derived using the probability density function expressed by formula (8) is used.

In addition, the arms of people in a group often point vertically downward. Therefore, if we think of an arm as a dipole, the probability density function of the dipole orientation is expressed by the following equation (18).

Figure 8 is a schematic diagram showing the probability density function expressed by equation (18). Figure 8 shows that the probability of existence of a dipole pointing downward is particularly higher than the probability of existence of dipoles pointing in other directions.

A covariance matrix representing a model of volume scattering when the probability density function represented by formula (18) is used is also obtained. In this case, a covariance matrix representing a model of volume scattering is obtained using formula (18), formula (10), and formula (11). As in the above case, a covariance matrix when [S] _h-dipole ^HV shown in formula (12) is substituted into the right side of the first formula of formula (11), and a covariance matrix when [S] _v-dipole ^HV shown in formula (13) is substituted into the right side of the first formula of formula (11) are obtained. When [S] _h-dipole ^HV is substituted into the right side of the first formula of formula (11), a covariance matrix of formula (19) shown below is obtained, and when [S] _v-dipole ^HV is substituted into the right side of the first formula of formula (11), a covariance matrix of formula (20) shown below is obtained.

The matrices shown in formula (14), formula (19), and formula (20) are all covariance matrices that represent a model of volume scattering when the individual is a human. However, which covariance matrix is used is determined according to the value of 10log(<|S _VV | ² >/<|S _HH | ² >). FIG. 9 is a schematic diagram showing which covariance matrix is used according to the value of 10log(<|S _VV | ² >/<|S _HH | ² >) when the individual is a human. That is, when 10log(<|S _VV | ² >/<|S _HH | ² >) is less than −3.5 dB, the covariance matrix shown in formula (20) is used. Furthermore, when 10log(<|S _VV | ² >/<|S _HH | ² >) is between −3.5 dB and 3.5 dB, the covariance matrix shown in formula (14) is used. When 10log(<|S _VV | ² >/<|S _HH | ² >) is greater than 3.5 dB, the covariance matrix shown in formula (19) is used.

As described above, whether the individual is an animal or a human, the covariance matrix representing the volume scattering model depends on the value of 10 log(<|S _VV | ² >/<|S _HH | ² >).

Next, we will explain how to calculate the scattering power of double reflection scattering and the scattering power of volume scattering.

Using a covariance matrix <[C]> _surface ^hv representing a model of surface scattering, a covariance matrix <[C]> _double ^hv representing a model of double reflection scattering, and a covariance matrix <[C]> _vol ^hv representing a model of volume scattering, the covariance matrix <[C]> ^HV is expressed as shown in Formula (21) below.

In formula (21), _fs , _fd , and _fv are expansion coefficients and are unknowns. As described above, α and β are also unknowns. Here, the following formula (22) is obtained using _fs , _fd , _fv , α, and β.

In the coefficient of _fv in each equation shown in equation (22), _mij is the element in the i-th row and j-th column of the covariance matrix representing the model of volume scattering. As described above, the covariance matrix representing the model of volume scattering differs depending on whether the individual is an animal or a human and the value of 10log(<|S _VV | ² >/<|S _HH | ² >). Specific examples of _m11 , _m22 , _m33 , and _m13 shown in equation (22) are shown below.

When the individuals belonging to the group are animals, the values of m ₁₁ , m ₂₂ , m ₃₃ , and m ₁₃ are shown in the following formula (23).

When the individuals belonging to the group are humans, the values of m ₁₁ , m ₂₂ , m ₃₃ , and m ₁₃ are shown in the following formula (24).

Furthermore, the scattering power of surface scattering is denoted as _Ps , the scattering power of double reflection scattering is denoted as _Pd , and the scattering power of volume scattering is denoted as _Pv . These scattering powers _Ps , _Pd , and _Pv can be expressed by the following formula (25).

Moreover, the sum of the scattered powers P _s , P _d , and P _v is expressed by the following equation (26).

The scattering model power decomposition unit 13 obtains a scattering matrix (the first matrix on the right side of the formula (4)) from the polarized radar data (specifically, the received electric field strength and the transmitted electric field strength) obtained by the radar data acquisition unit 11 and the formula (4). The scattering model power decomposition unit 13 obtains a covariance matrix <[C]> ^HV expressed by the formula (5) from the scattering matrix. As a result, <|S _HH | ² >, <|S _HV | ² >, <|S _VV | ² >, <|S _HH S _VV ^* |> are obtained. The scattering model power decomposition unit 13 substitutes these values into the formula (22) and solves the simultaneous equations using the eight formulas shown in the formulas (22), (25), and (26) to obtain the unknowns P _s , P _d , P _v , f _s , f _d , f _v , α, and β. This results in the scattering power _Pd of double reflective scattering and the scattering power _Pv of volume scattering.

In addition, it is assumed that whether the individual to be estimated for density is an animal or a human is specified in advance. If the individual is an animal, the values of _m11 , _m22 , _m33 , and _m13 corresponding to the value of 10log(<|S _VV | ² >/<|S _HH | ² >) are identified based on formula (23), and the values are substituted into formula (22). If the individual is a human, the values of _m11 , _m22 , _m33 , and _m13 corresponding to the value of 10log(<|S _VV | ² >/<|S _HH | ² >) are identified based on formula (24), and the values are substituted into formula (22).

A flock of individuals has occurred in the target area for estimating the density, and P _d and P _v calculated as described above using polarized radar data at the time when the density of the individuals is estimated correspond to the above-mentioned P _d1 and P _v1 , respectively.

Moreover, _Pd and _Pv calculated as above using polarized radar data at a time when it is known that no flocks of individuals have occurred in the target area correspond to the above-mentioned _Pd0 and _Pv0 , respectively. Here, when calculating _Pd and _Pv using polarized radar data at a time when it is known that no flocks of individuals have occurred in the target area, it may be specified whether an individual belonging to the flock in the case where a flock has occurred is an animal or a human. In this case, as in the above case, if the specified individual is an animal, the values of _m11 , _m22 , _m33 , and _m13 corresponding to the value of 10log(<|S _VV | ² >/<|S _HH | ² >) are identified based on formula (23), and the values are substituted into formula (22). Furthermore, when the designated individual is a human being, the values of _m11 , _m22 , _m33 , and _m13 corresponding to the value of 10log(<|S _VV | ² >/<|S _HH | ² >) are identified based on formula (24), and the values are substituted into formula (22). In this case, _Pd0 and _Pv0 when the designated individual is an animal may differ from _Pd0 and _Pv0 when the designated individual is a human being.

In addition, when calculating Pd and _Pv using polarized radar data at a time when it is known that no flock of individuals has occurred in the target area, _Pd and _Pv may be calculated regardless of whether the specified individual is an animal or a human. When no flock of individuals has occurred in the target area, the value of 10log(<|S _VV | ² >/<|S _HH | ² >) is considered to be _a value near 0 dB. In this case, whether the individual is a human or an animal, the values of (m ₁₁ , m ₂₂ , m ₃₃ , m ₁₃ ) are the same, i.e., (3/8, 1/4, 3/8, 1/8). Therefore, _Pd and _Pv may be calculated by substituting 3/8, 1/4, 3/8, and 1/8 as the values of m ₁₁ , m ₂₂ , m ₃₃ , and m ₁₃ into equation (22), respectively. In this case, P _d0 and P _v0 do not change depending on whether the individual is an animal or a human being.

As described above, the scattering model power decomposition unit 13 calculates the first difference ΔP _d (the difference between the scattered power of double reflection scattering when a school has occurred and the scattered power of double reflection scattering when a school has not occurred) using equation (2). Also, the scattering model power decomposition unit 13 calculates the second difference ΔP _v (the difference between the scattered power of volume scattering when a school has occurred and the scattered power of volume scattering when a school has not occurred) using equation (3).

The estimation unit 15 estimates the density of individuals by applying the first difference ΔP _d and the second difference ΔP _v obtained as a result to a neural network.

The radar data acquisition unit 11 and the density acquisition unit 12 are realized, for example, by a central processing unit (CPU) of a computer that operates according to a density estimation program, and a communication interface of the computer. For example, the CPU may read the density estimation program from a program recording medium such as a program storage device of the computer, and operate as the radar data acquisition unit 11 and the density acquisition unit 12 using a communication interface according to the density estimation program. In addition, the scattering model power decomposition unit 13, the learning unit 14, the estimation unit 15, and the population map generation unit 17 are realized, for example, by a CPU of a computer that operates according to a density estimation program. For example, the CPU may read the density estimation program from a program recording medium as described above, and operate as the scattering model power decomposition unit 13, the learning unit 14, the estimation unit 15, and the population map generation unit 17 according to the density estimation program.

Next, we'll explain the process.

Figure 10 is a flowchart showing an example of the process when the density estimation device 10 of this embodiment learns a neural network. Explanations of matters that have already been explained will be omitted as appropriate.

It is assumed that whether the individuals belonging to the flock are humans or animals is specified in advance. It is also assumed that a flock of individuals occurs in an area where another system calculates the density. It is also assumed that the scattering power P _d (P _d0 ) of double reflection scattering and the scattering power P _v (P _v0 ) of volume scattering when no flock of individuals occurs are calculated in advance.

The radar data acquisition unit 11 acquires polarized radar data from the polarized radar (step S1). This polarized radar data is data obtained by the polarized radar irradiating radio waves onto an area for which another system calculates the density.

Next, the scattering model power decomposition unit 13 uses the polarized radar data obtained in step S1 to calculate the scattering power P _d (P _d1 ) of double reflection scattering when a flock of individuals occurs, and the scattering power P _v (P _v1 ) of volume scattering (step S2).

Next, the scattering model power decomposition unit 13 calculates a first difference ΔP _d (the difference between the scattering power P _d1 of double reflection scattering when a swarm has occurred and the scattering power P _d0 of double reflection scattering when a swarm has not occurred) and a second difference ΔP _v (the difference between the scattering power P _v1 of volume scattering when a swarm has occurred and the scattering power P _v0 of volume scattering when a swarm has not occurred) (step S3).

The congestion acquisition unit 12 also acquires the congestion calculated by the other system from the other system (step S4).

Next, the learning unit 14 learns the neural network by learning a weight matrix for each pair of adjacent layers of the neural network using the density obtained in step S4 and the first difference ΔP _d and the second difference ΔP _v calculated in step S3 (step S5). As already described, the number of layers of the neural network and the number of neurons in each layer are predetermined.

The density estimation device 10 performs steps S1 to S5 for each case in which an animal is specified as an individual and each case in which a human is specified as an individual, and generates a neural network corresponding to the animal and a neural network corresponding to the human. In the process of step S2, branching processing is performed between the case in which the individual is a human and the case in which the individual is an animal. The process of step S2 will be described later.

Figure 11 is a flowchart showing an example of the process when the density estimation device 10 of this embodiment estimates the density of individuals belonging to a flock. Explanations of matters that have already been explained will be omitted as appropriate.

It is assumed that whether the individuals belonging to the flock are humans or animals is specified in advance. Also, it is assumed that the scattering power P _d (P _d0 ) of double reflection scattering and the scattering power P _v (P _v0 ) of volume scattering when no flock of individuals occurs in the area in which the density of individuals is to be estimated are calculated in advance.

The radar data acquisition unit 11 acquires polarized radar data from the polarized radar (step S11). This polarized radar data is data obtained by the polarized radar irradiating radio waves onto the area for which the density of individuals is to be estimated.

Next, the scattering model power decomposition unit 13 calculates the scattering power P _d (P _d1 ) of double reflection scattering and the scattering power P _v (P _v1 ) of volume scattering using the polarized radar data obtained in step S11 (step S12).

Next, the scattering model power decomposition unit 13 calculates the first difference ΔP _d and the second difference ΔP _v (step S13).

Steps S12 and S13 are the same processes as steps S2 and S3 described above (see Figure 10).

Next, the estimation unit 15 estimates the density of individuals by applying the first difference ΔP _d and the second difference ΔP _v to a neural network (step S14). If an animal is specified as an individual, the estimation unit 15 may use a neural network corresponding to the animal in step S14. If a human is specified as an individual, the estimation unit 15 may use a neural network corresponding to the human in step S14.

The population density estimation device 10 performs the processes of steps S11 to S14 for each area corresponding to each pixel. As a result, the population density is obtained for each pixel. After the population density for each pixel is obtained, the population map generation unit 17 generates a population map (see, for example, the lower part of Figure 4) based on the density for each pixel. The method by which the population map generation unit 17 generates the population map has already been explained, so the explanation will be omitted here. The population map generation unit 17 displays the population map on, for example, a display device (output unit 18).

As described above, step S2 (see FIG. 10) and step S12 (see FIG. 11) are similar processes, and both are processes for calculating the scattering power _Pd of double reflection scattering and the scattering power _Pv of volume scattering. Below, the process progress of the processes (steps S2, S12) for calculating the scattering power _Pd of double reflection scattering and the scattering power _Pv of volume scattering will be described. FIGS. 12 and 13 are flowcharts showing an example of the process progress of the processes for calculating the scattering power _Pd of double reflection scattering and the scattering power _Pv of volume scattering. Descriptions of matters that have already been described will be omitted as appropriate.

In the following description, 10 log(<|S _VV | ² >/<|S _HH | ² >) will be represented by the symbol L.

The scattering model power decomposition unit 13 calculates the scattering matrix (the first matrix on the right-hand side of equation (4)) using equation (4) and the received electric field strength and the transmitted electric field strength contained in the polarized radar data (step S21).

Next, the scattering model power decomposition unit 13 calculates the covariance matrix <[C]> ^HV shown in equation (5) from the scattering matrix (step S22).

Next, the scattering model power decomposition unit 13 determines whether the pre-specified individual is a human or an animal (step S23).

If the pre-specified individual is a human, proceed to step S24.

In step S24, scattering model power decomposition unit 13 determines the category to which the value of L (that is, the value of 10 log(<|S _VV | ² >/<|S _HH | ² >)) belongs.

If the value of L is less than -3.5 dB, scattering model power decomposition section 13 substitutes 5/8, 1/4, 1/8, and 1/8 into _m11 , _m22 , _m33 , and _m13 in equation (22), respectively (step S25).After step S25, the process proceeds to step S32.

When the value of L is -3.5 dB or more and 3.5 dB or less, scattering model power decomposition section 13 substitutes 3/8, 1/4, 3/8, and 1/8 into m ₁₁ , m ₂₂ , m ₃₃ , and m ₁₃ in equation (22) respectively (step S26). After step S26, the process proceeds to step S32.

If the value of L exceeds 3.5 dB, scattering model power decomposition section 13 substitutes 1/8, 1/4, 5/8, and 1/8 into m ₁₁ , m ₂₂ , m ₃₃ , and m ₁₃ in equation (22) respectively (step S27).After step S27, the process proceeds to step S32.

Also, if the pre-specified individual is an animal, the process proceeds to step S28 (see FIG. 13).

In step S28, scattering model power decomposition unit 13 determines the category to which the value of L (that is, the value of 10 log(<|S _VV | ² >/<|S _HH | ² >)) belongs.

If the value of L is less than -2 dB, scattering model power decomposition section 13 substitutes 8/15, 4/15, 1/5, and 2/15 into _m11 , _m22 , _m33 , and _m13 in equation (22), respectively (step S29). After step S29, the process proceeds to step S32 (see FIG. 12).

When the value of L is -2 dB or more and 2 dB or less, scattering model power decomposition section 13 substitutes 3/8, 1/4, 3/8, and 1/8 into m ₁₁ , m ₂₂ , m ₃₃ , and m ₁₃ in equation (22) respectively (step S30). After step S30, the process proceeds to step S32.

When the value of L exceeds 2 dB, scattering model power decomposition section 13 substitutes 1/5, 4/15, 8/15, and 2/15 into m ₁₁ , m ₂₂ , m ₃₃ , and m ₁₃ in equation (22) respectively (step S31).After step S31, the process proceeds to step S32.

At the time of proceeding to step S32 (see FIG. 12 ), values have been substituted for _m11 , _m22 , _m33 , and _m13 in equation (22). In step S32, the scattering model power decomposition unit 13 calculates the scattering power _Pd of second-order reflective scattering and the scattering power _Pv of volume scattering by solving simultaneous equations using the eight equations shown in equations (22), (25), and (26).

Steps S21 to S32 shown in Figures 12 and 13 are examples of the processing progress of the scattering model power decomposition unit 13 in step S2 (see Figure 10) and step S12 (see Figure 11).

According to this embodiment, the scattering model power decomposition unit 13 calculates the difference (first difference) between the scattering power of double reflection scattering when a swarm occurs and the scattering power of double reflection scattering when a swarm does not occur, and the difference (second difference) between the scattering power of volume scattering when a swarm occurs and the scattering power of volume scattering when a swarm does not occur. Then, the estimation unit 15 applies the first difference and the second difference to a neural network to estimate the density of individuals. Therefore, the density estimation device 10 can estimate the density of individuals while eliminating the influence of stationary objects.

The estimation unit 15 also estimates the density of individuals without using data on the scattering power of surface scattering. Therefore, the density estimation device 10 can estimate the density of individuals while eliminating the influence of objects that cause surface scattering. Note that humans and animals do not fall under the category of objects that cause surface scattering.

This allows for accurate estimation of the density of individuals.

Furthermore, when the individual is an animal, the scattering model power decomposition unit 13 substitutes values obtained from a covariance matrix (see FIG. 7) representing a model of animal volume scattering into _m11 , _m22 , _m33 , and _m13 in equation (22). When the individual is a human, the scattering model power decomposition unit 13 substitutes values obtained from a covariance matrix (see FIG. 9) representing a model of human volume scattering into _m11 , _m22 , _m33 , and _m13 in equation (22). Therefore, whether estimating the density of animals or the density of humans, the density can be estimated with high accuracy.

In the above embodiment, the learning model is a neural network. The learning model may be a learning model obtained by other machine learning techniques. For example, the learning model may be a learning model obtained based on linear regression or log-linear regression.

In the above embodiment, the estimation unit 15 estimates the density of individuals by applying the first difference _ΔPd and the second difference _ΔPv to a learning model. The estimation unit 15 may estimate the density of individuals without using a learning model. In this case, the learning unit 14 does not need to be provided. Hereinafter, a case will be described in which the estimation unit 15 estimates the density of individuals without using a learning model.

In human and animal flocks, double scattering occurs when the density of individuals is low to moderate, and as density increases, the scattering power of double scattering increases.

Furthermore, when the density of individuals is high, such as in crowds of people and animals, the scattering power of double reflection scattering becomes saturated and volume scattering occurs. As the density increases, the scattering power of volume scattering increases.

Therefore, when the density is low to medium, the density can be expressed by a function of the first difference ΔP _d . Hereinafter, this function will be referred to as the first function. When the density is high, the density can be expressed by a function of the second difference ΔP _v . Hereinafter, this function will be referred to as the second function.

The first function and the second function may be determined in advance. Then, the estimation unit 15 may determine whether to estimate the density using the first function or the second function based on the value of ΔP _d /ΔP _v . That is, the estimation unit 15 calculates the value of ΔP _d /ΔP _v . Then, if the value of ΔP _d /ΔP _v is equal to or greater than a predetermined threshold value (represented as Th), the estimation unit 15 may estimate the density from the value of the first function and the value of the first difference ΔP _d . Also, if the value of ΔP _d /ΔP _v is less than the threshold value Th, the estimation unit 15 may estimate the density from the value of the second function and the value of the first difference ΔP _v .

FIG. 14 is a schematic block diagram showing an example of the configuration of a computer relating to the congestion estimation device 10 of an embodiment of the present invention. The computer 1000 includes, for example, a CPU 1001, a main memory device 1002, an auxiliary memory device 1003, an interface 1004, a display device 1005, and a communication interface 1006.

The congestion estimation device 10 according to an embodiment of the present invention is realized by a computer 1000. The operation of the congestion estimation device 10 is stored in the auxiliary storage device 1003 in the form of a program (density estimation program). The CPU 1001 reads the program from the auxiliary storage device 1003, deploys the program in the main storage device 1002, and executes the processing described in the above embodiment according to the program.

The auxiliary storage device 1003 is an example of a non-transient tangible medium. Other examples of non-transient tangible media include a magnetic disk, a magneto-optical disk, a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), and a semiconductor memory that are connected via the interface 1004. In addition, when a program is distributed to the computer 1000 via a communication line, the computer 1000 that receives the program may load the program into the main storage device 1002 and execute the processing described in the above embodiment in accordance with the program.

Furthermore, some or all of the components may be realized by general-purpose or dedicated circuits, processors, etc., or a combination of these. These may be configured by a single chip, or by multiple chips connected via a bus. Some or all of the components may be realized by a combination of the above-mentioned circuits, etc., and programs.

When some or all of the components are realized by multiple information processing devices, circuits, etc., the multiple information processing devices, circuits, etc. may be centrally located or distributed. For example, the information processing devices, circuits, etc. may be realized as a client-server system, cloud computing system, etc., in a form in which each is connected via a communication network.

Next, an overview of the present invention will be described. FIG. 15 is a block diagram showing an overview of a congestion estimation device of the present invention. The congestion estimation device of the present invention includes a difference calculation means 73 and an estimation means 75.

The difference calculation means 73 (e.g., the scattering model power decomposition unit 13) calculates a first difference which is the difference between the scattering power of double reflection scattering when a flock of individuals occurs and the scattering power of double reflection scattering when a flock of individuals does not occur, and a second difference which is the difference between the scattering power of volume scattering when a flock of individuals occurs and the scattering power of volume scattering when a flock of individuals does not occur.

The estimation means 75 (e.g., the estimation unit 15) estimates the density, which is the number of individuals per unit area, based on the first difference and the second difference.

This configuration allows for accurate estimation of the density of the individual in question.

The difference calculation means 73 may be configured to calculate a first difference and a second difference when the individual is a human using a model of double-reflection scattering and a model of volume scattering of a human flock, and the estimation means 75 may be configured to estimate the density when the individual is a human based on the first difference and the second difference.

The system may also be configured to include a population map generating means (e.g., population map generating unit 17) that generates a population map, which is information that shows the number of individuals in each area in a map format based on the density of each area.

The estimation means 75 may also be configured to estimate the density by applying the first difference and the second difference to a pre-trained learning model.

The configuration may also include a learning means (e.g., learning unit 14) that learns a learning model by machine learning using a known density and a first difference and a second difference corresponding to the known density.

The learning model may also be a neural network.

The present invention is suitable for use in a density estimation device that estimates the density of individuals.

REFERENCE SIGNS LIST 10 Density estimation device 11 Radar data acquisition unit 12 Density acquisition unit 13 Scattering model power decomposition unit 14 Learning unit 15 Estimation unit 17 Population map generation unit 18 Output unit

Claims (10)

a difference calculation means for calculating a first difference which is the difference between the scattering power of double reflection scattering calculated using the polarized radar data when a swarm of individuals has occurred and the scattering power of double reflection scattering calculated using the polarized radar data when a swarm of individuals has not occurred , and a second difference which is the difference between the scattering power of volume scattering calculated using the polarized radar data when a swarm of individuals has occurred and the scattering power of volume scattering calculated using the polarized radar data when a swarm of individuals has not occurred;
and an estimation means for estimating a density, which is the number of individuals per unit area, based on the first difference and the second difference. the difference calculation means calculates a first difference and a second difference when the individual is a human , using a double-reflection scattering model calculated using the polarized radar data and a volume scattering model of a human group calculated using the polarized radar data ;
The crowding degree estimation device according to claim 1 , wherein the estimation means estimates the crowding degree when the individual is a human being, based on the first difference and the second difference. 3. The density estimation device according to claim 1, further comprising: a population map generation means for generating a population map, which is information that represents the number of individuals in each area in a map form, based on the density of each area. The congestion degree estimation device according to claim 1 , wherein the estimation means estimates the congestion degree by applying the first difference and the second difference to a learning model that has been learned in advance. The congestion estimation device according to claim 4 , further comprising: a learning means configured to learn a learning model by machine learning using a known congestion degree and a first difference and a second difference corresponding to the known congestion degree. The congestion estimation device according to claim 4 or 5, wherein the learning model is a neural network. The computer
calculating a first difference which is the difference between the scattering power of double reflection scattering calculated using the polarized radar data when a flock of individuals has occurred and the scattering power of double reflection scattering calculated using the polarized radar data when a flock of individuals has not occurred , and a second difference which is the difference between the scattering power of volume scattering calculated using the polarized radar data when a flock of individuals has occurred and the scattering power of volume scattering calculated using the polarized radar data when a flock of individuals has not occurred;
and estimating a density, which is the number of individuals per unit area, based on the first difference and the second difference. The computer
when calculating the first difference and the second difference, a model of double reflection scattering calculated using the polarized radar data and a model of volume scattering of a human group calculated using the polarized radar data are used to calculate the first difference and the second difference when the individual is a human;
The method of claim 7 , further comprising estimating a crowding level when the individual is a human being based on the first difference and the second difference. On the computer,
a difference calculation process for calculating a first difference which is the difference between the scattering power of double reflection scattering calculated using the polarized radar data when a flock of individuals has occurred and the scattering power of double reflection scattering calculated using the polarized radar data when a flock of individuals has not occurred , and a second difference which is the difference between the scattering power of volume scattering calculated using the polarized radar data when a flock of individuals has occurred and the scattering power of volume scattering calculated using the polarized radar data when a flock of individuals has not occurred; and
a density estimation program for executing an estimation process for estimating a density, which is the number of individuals per unit area, based on the first difference and the second difference. On the computer,
in the difference calculation process, a first difference and a second difference are calculated when the individual is a human being , using a model of double reflection scattering calculated using the polarized radar data and a model of volume scattering of a human group calculated using the polarized radar data ;
The computer-readable storage medium according to claim 9 , further comprising: a computer-readable storage medium for storing a computer-readable storage medium that stores a computer-readable storage medium and ...

Images