JPH07318645A - Target information decision unit - Google Patents

Abstract

PURPOSE:To discriminate between a friend and enemy flying objects highly accurately based on a continuous feature amount representative of the positional relationship between the information related to the object and each sky region. CONSTITUTION:A neural network A in a threshold decision means 10 learns 26 required number of teacher signal being set 25 including a flight prohibition, planar coordinates of each safety sky region, and upper/lower sea level altitudes. In this regard, a feature amount employing a sigmoid function represents the positional relationship of a target with respect to each sky region. A neural network B then learns 29 the relationship between the combination of required number of enemy and friend feature amounts prepared as teacher signals and the decision results of target information. Motion features of a target is then inputted 20 and the planar coordinates and seal level altitude of the target are calculated 21. The network A then calculates the feature amounts for the flight prohibition and each safety sky region based on the calculation results. A feature amount decision means 11 receives two feature amounts and the network B calculates 23 decision results of target information discriminating between enemy and friend.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a target information judging device which inputs information on a target obtained by an observing means and outputs, for example, a kind of target, discrimination of enemy and ally as a judgment result. In the previous Gulf War, ballistic missiles flying from high altitudes, cruise missiles flying at low altitudes according to a designated course, bombers that are difficult to detect by radar, and various other types of attack weapons with high capabilities were used. . In addition, many defense weapons were used to explode and neutralize these attack weapons. An interceptor missile system is one of the defense weapons. This is to capture an attacking weapon as a target by a sensor using radio waves, infrared rays, visible light, etc., and carry out shooting based on information such as the position and speed observed by the sensor to destroy the target. FIG. 3 is a schematic view showing an operating state of the surface-to-air interceptor missile system, where 1 is a target for shooting the interceptor missile system, and 2 is an attacking weapon 1.
The radar device 3 for observing the laser beam, 3 is a tracking beam emitted from the radar device for tracking the attacking weapon 1, and 4 is a shooting beam for controlling the entire missile system by making various judgments and processes based on the information obtained from the radar device. Control device, 5
Is an intercepting missile that destroys the shooting target, and 6 is a launching device that launches the intercepting missile 5. The interception missile system is also required to have high capability as the development of attack weapons. In other words, in addition to improving basic capabilities such as sensor target detection performance, resolution, speed of missiles, flight performance, etc., multiple interception weapons are grouped as one interception system to effectively and efficiently The ability to perform specific fire, for example, the ability to select the type of interception weapon and the interception method depending on the type of target, when intercepting multiple targets at the same time, multiple interception weapons for one target At the same time, it is necessary to have a function to fire quickly so that there is no shooting and no fire. In order for the interception system to have these capabilities, the type and number of targets used to select the type of interception weapon and the interception method for shooting the target, and the target to prevent accidental fire to allies Ability to centrally judge weapons scheduling, such as whether to shoot multiple targets in different orders, in what order and where with the interception weapons installed, based on information about the targets obtained from the sensors is necessary.
In this way, determine various information obtained from the sensor,
The target information determination device embodies the function of outputting it as a determination result.

[0002]

2. Description of the Related Art FIG. 2 is a block diagram of a conventional target information determining apparatus, in which 7 is an observing means for tracking a target and obtaining information about the tracking target, and 8 is input with information about the target as input. Target information determining means for performing threshold value determination and feature amount determination and outputting a determination result of target information, and 9 is a display means for displaying information regarding the target and the information determination result of the target.

In the target information judging means 8, reference numeral 10 is a threshold value judging means for comparing the information about the target with a threshold value and outputting the characteristic quantity, and 11 is the characteristic quantity and the characteristic quantity outputted by the threshold value judging means 12. Feature amount determination means for comparing the determination rules and outputting the determination result of the target information, 12 is a threshold value database for comparing with the information on the target input from the observation means 7 in the 10 threshold value determination means, 13
Is a feature quantity determination rule database for the 11 feature quantity determination means to compare with the feature quantity input from the 10 threshold value determination means.

Next, the operation in the case where the conventional target information judging device outputs whether the target is an enemy or an ally as the judgment result is shown in FIG.
It will be described in accordance with the flowchart of. In the figure, ST is an abbreviation for step.

At ST1, target motion parameters are obtained from the observation means 7.

At ST2, the first threshold value data is obtained from the threshold value database 12. This is, for example, a no-fly zone where the flight of the aircraft is prohibited or a safe zone where a ally aircraft is allowed to fly, and there may be multiple same zones.

In ST3, among the target motion parameters,
The inside / outside determination of the area is performed by comparing the coordinates on the plane and the altitude above sea level with this first area.

In ST4, the determination result of inside or outside determined in ST3 is stored as a target feature amount for the first airspace.

In the loop of ST5, the processes from ST2 to ST4 are performed for all n areas including the flight prohibited area and the safe area.

In ST6, a rule for determination is obtained from the feature amount determination rule database 13. For example, the feature quantity for the first no-fly area is "inside",
When the feature amount for the second safe airspace is "outside", the target is an enemy, or the feature amount for the first flight prohibited airspace is "outside" and the feature amount for the second safe airspace is "inside". , The target is an ally, or
The rule is that when the feature amount for the first prohibited airspace is “outside” and the feature amount for the second safe airspace is also “outside”, the enemy or ally of the target is unknown.

In ST7, the combination of n feature amounts and the feature amount determination rule are compared to obtain the target enemy or ally who matches the rule.

[0012] In ST8, whether the target is an enemy or an ally, which is the result of the determination of the target information, is output.

In ST9, if a new observation means can be obtained from the observation means 7, ST1 to ST8 are repeated, and if not, the operation is ended.

[0014]

Since the conventional target information judging device is configured as described above, even if the aircraft operated by a human in the threshold judging means intends to fly in no air space. Since the deviation occurs and the target motion data obtained from the radar device also has an error or the like, the feature quantity should be represented as a continuous value with ambiguity, but the feature quantity is only binary. If the number and types of air spaces to be determined that cannot be represented or that the feature amount determination means cannot handle consecutive feature amounts increase, the amount of rules in the feature amount determination rule database increases and it becomes difficult to change them. Due to the plurality of reasons, there has been a problem that the determination accuracy of the target information is reduced overall.

The present invention improves the above-mentioned problems,
The object is to obtain a target information determination device that determines whether a flying target is an enemy or an ally.

Further, the present invention improves the above-mentioned problems, and obtains a target information judging device for judging the magnitude of the threat that a flying target poses to the region where the missile system rises. Has an aim.

[0017]

A target information determination device according to the present invention is a flight-prohibited air region where the flight of an aircraft is prohibited,
Relationship between threshold judgment means having a neural network that has learned a safe airspace in which an ally's aircraft is permitted to fly, a feature quantity output from the threshold judgment means, and the target enemy or ally The neural network learning means is provided with a feature amount judging means for obtaining a judgment result of the target information as to whether the target is an enemy or ally, and a learning means for learning the two neural networks.

Further, the target information judging device of the present invention has a threshold judging means having a neural network in which a function for normalizing the time taken for the missile system to defend and the target to reach the region is learned. And a neural network that learns the relationship between the feature amount that is the output of the threshold value determination means and the magnitude of the threat that the target poses to the area protected by the missile system, and the target protects the missile system. A feature amount determination means for obtaining the size of the threat given to the region to be obtained as the determination result of the target information, and a learning means for learning the two neural networks are provided.

[0019]

The target information determining device of the present invention stores the information about the target when the neural network stores the prohibited air space where the flight of the aircraft is prohibited and the safe air space where the aircraft of the friend is permitted to fly. , It is possible to represent the positional relationship with each airspace by a continuous feature amount, and output the target enemy or ally based on the feature amount.

Further, the target information judging device of the present invention is
Whether or not the target is flying toward the defending area and is normalized when the function that normalizes the time taken for the missile system to defend and the target reaching the defending area is stored. Representing the target arrival time with a feature amount,
Based on the feature quantity, the target can output the magnitude of the threat given to the area protected by the missile system.

[0021]

【Example】

First Embodiment FIG. 1 is a block diagram of a target information determination device according to a first embodiment. In FIG. 1, 7 and 9 are exactly the same as those of the conventional device. Reference numeral 8 is a target information determination means in the present invention, 10 is a threshold value determination means for determining a threshold value using a neural network, and 11 is a feature amount determination means for determining a feature amount using a neural network. , 14 are learning means for learning the neural network in 10 and 11.

In the observing means 7, 16 is a radar antenna for radiating and receiving radio waves, and 17 is the antenna 1
6, a transmitter / receiver for supplying power to 6, an indicator for displaying a received wave, and a reference numeral 19 for the received signal.
It is a signal processor for input to.

In the threshold value judging means 10 of the target information judging means 8, 20 is a signal input device for receiving the output from the observing means 7, and 21 is for generating motion parameters such as the target position and speed. The motion specification generator 22 is a feature quantity calculation element that is composed of the neural network A and calculates a feature quantity representing the feature of the target from the motion specification of the target.

A neural network is a neural network of the brain of an animal, and here refers to a circuit simulating its function by computer software or a circuit simulating its structure.

FIG. 5 is a diagram for explaining an example of the structure of the neural network in the case of the three-layer structure, and 15 is
It is a neuron corresponding to a nerve cell, and there are a total of 7 neurons.

In FIG. 5, each neuron is
It has a function that adds a weight called synapse weight to one or more inputs and adds it, and an input / output function that converts it into an output to the next neuron.

The first layer, which is the input layer of the neural network A, has three neurons, which are the coordinates on the plane of the target and the altitude of the target sea level. The coordinates on the plane are, for example, latitude and longitude. The output layer has two neurons, and is a feature quantity indicating the positional relationship of the target with respect to the no-flying airspace and a feature quantity indicating the positional relationship of the target with respect to the safe airspace, both of which are continuous values between 0 and 1, for example. take. The neural network is, for example, back propagation that is one of neural networks, and is composed of four or more layers. When the neural network used has four layers, the output layer is the fourth layer.

When the neural network A is composed of four layers, the first layer is an input layer for inputting the target position, and the second layer is the target position where the flight prohibited area is
It is a layer that determines whether it is in the area inside or outside the side for each side that constitutes the area such as the safety airspace,
The third layer is a layer that integrates the inside and outside of each side to determine whether or not the target position is within the airspace, and the fourth layer integrates the inside and outside of each side and the target position is , A layer that outputs the inside or outside of each airspace.

In the i-th neuron of the k-th layer of the neural network A, a function for adding a weight called a synapse weight to one or more inputs and adding it, and an output of the neuron are shown in Equation 1.

[0030]

[Equation 1]

In the equation 1, the input / output function f (x)
Uses, for example, a sigmoid function. Equation 2 shows the sigmoid function.

[0032]

[Equation 2]

In the feature amount determination means 11 of the target information determination means 8, 23 is a neural network B.
Is a determination result calculation element for calculating the determination result of the target information from the target feature amount obtained by the threshold value determination means 10.

The first layer, which is the input layer of the neural network B, has two neurons when there are two types of airspace, namely, a no-flying airspace and a safe airspace. It is the feature amount for the flight prohibited airspace and the feature amount for the target safe airspace,
The result of the judgment regarding the target enemy and ally is 3 for enemy, ally, and unknown.
In one case, the output layer has three neurons, and this neural network is, for example, backpropagation which is one of the forms of the neural network, and is composed of four or more layers. Input and output according to the formula.

In the learning means 14 of the above-mentioned target evaluation means 8, 24 is an airspace database which is a database of a flight prohibited airspace and a safe airspace, and 25 is a teacher signal for generating a teacher signal for learning the neural network A. A generation element, 26 is a feature amount calculation learning element for learning the neural network A, 27 is a feature amount determination function database which is a database of the relation between the feature amount and the determination result of the target information, and 28 is the neural network B A teacher signal generation element for generating a teacher signal for causing the learning signal 29 to be a determination result calculation learning element for learning the neural network B.

In the display means 9, 30 is a display signal generator for converting into a signal for displaying the determination result of the target information calculated by the determination result calculation element 23, and 31 is a display of the determination result of the target information. It is an indicator that does.

Next, in the above embodiment, for example, the neural network used for the threshold value judging means 10 and the characteristic amount judging means 11 is back propagation, and the target information is judged as enemy, ally, or unknown. The operation of the information judging means 8 will be described with reference to the flowchart shown in FIG. ST10 to 15 indicate the operation of the learning means 14. ST 1, 16 and 17 show the operation of the threshold value judging means 10. ST18 and ST8 indicate the operation of the feature amount determination means 11.

At ST10, a required number of flight prohibited areas are set in order to use them as the teacher signals of the neural network A of the threshold value determination means, and the altitude on the plane indicating the coordinates of the flight prohibited areas on the plane and the upper and lower limits of the air areas are set. The altitude is the teacher signal.

In ST11, the coordinates on the plane of the no-fly area created in ST10 and the altitude above sea level indicating the upper and lower limits of the air area are used as teacher signals in the neural network A of the threshold value judging means, for example, as shown in FIG. The flight-prohibited airspace 32 as shown in is learned. FIG. 7B shows an example of the distribution of the feature quantity in the D-E cross section of FIG. 7A when the sigmoid function is used as the input / output function of the neuron of the neural network. The feature quantity in this case is
It is a value indicating the positional relationship of the target with respect to the no-fly area. The higher the value, the higher the certainty that you are flying in a no-fly zone. In order to express the smooth feature quantity as shown in FIG. 7B, the sigmoid function shown in the above-mentioned Expression 2 is used as the input / output function of the neuron of the neural network.

In ST12, the threshold value judging means 10
In order to use the neural network A as a teacher signal,
Create the required number of safety airspace, and coordinate on the plane of the safety airspace,
The altitude above and below the airspace is used as the teacher signal.

In ST13, the coordinates on the plane of the safe airspace created in ST12 and the altitude above sea level indicating the upper and lower limits of the airspace are used as teacher signals in the neural network A of the threshold value judging means 10, for example, as shown in FIG. The distribution of the feature quantity corresponding to the safe airspace 33 as shown in (4) is learned.

At ST14, the required number of enemy feature amounts, ally feature amounts, and target information determination results shown in Table 1 are created as teacher signals. In Table 1, (Large) and (Small) are actually numerical values between 1 and 0 indicating the feature amount.

[0043]

[Table 1]

In ST15, the relationship between the combination of each feature amount and the determination result of the target information is learned by the teacher signal created in ST14.

At ST1, the target motion parameters are input from the observation means 7.

At ST16, the coordinate of the target on the plane and the altitude above sea level are calculated.

In ST17, with the value obtained in ST16 as an input, the neural network A learned in ST11 and ST13 is used to calculate the characteristic amount for the flight prohibited airspace and the characteristic amount for the safe airspace.

In ST18, 2 calculated in ST17
With one feature amount input as it is, the neural network B learned in ST15 calculates the determination result of the target information as to whether the target is an enemy or an ally.

In ST8, the determination result of the target information obtained in ST18 is output to the display means 9.

In ST9, when new target motion parameters are obtained from the observation means 7, ST1, 16, 17, 1
8 and 8 are repeated, and if not obtained, the operation ends.

Next, the operation of the learning ST11 of the neural network A of FIG. 6 will be described with reference to the flowchart of FIG. The learning of the neural network B in ST13 and ST15 is the same as in ST11.

At ST19, 1 is input to the neuron of the feature quantity of the flight prohibition air space in the output layer of the neural network A as the feature quantity of the flight prohibition air space.

In ST20, 0 is input to the neuron of the feature amount of the safe airspace in the output layer of the neural network as the feature amount of the safe airspace.

In ST21, the neural network performs an operation in the direction opposite to the normal direction, that is, an operation for obtaining the coordinates on the target plane and the target altitude above sea level from the feature amount to obtain a value.

In ST22, using the coordinates on the plane of the target and the altitude above sea level obtained in ST21, and the coordinates on the plane of the target and the altitude above sea level created as a teacher signal,
The synapse weight in the neural network is adjusted so that the evaluation function, Equation 3, is minimized.

[0056]

[Equation 3]

1 of learning from ST19 to ST22
It is possible to make the neural network learn the no-fly area by repeating the cycle and learning the necessary number of times.

In the above-mentioned embodiment, the case where the judgment regarding the flight prohibited area and the safe air area is made has been described. However, even if the present invention is applied to other air areas, for example, an air area where an enemy may fly. It is valid. Although the case where back propagation is used as the neural network has been described, it is also useful to apply other various types of neural networks to the present invention.

Third Embodiment FIG. 9 is a block diagram of a target information judging device according to the second embodiment. In FIG. 9, 7 to 11, 14, 16 to 21, 23
25 to 31 are exactly the same as those in FIG. 1, which is a block diagram of the target information determining apparatus of the first embodiment.

In the threshold value judging means 10, 37
Is a target arrival time generator that predicts the time required for the target to reach the area to be protected by the missile system, and 38 is a database of the defense area that stores the coordinates of the defense area.

In the feature quantity calculating element 22 of the threshold value judging means 10, the first layer which is the input layer of the neural network A has five neurons, which have two coordinates on the target plane and the normalized velocity. The two vectors and the target arrival time predicted by the target arrival time generator 37. The coordinates on the plane are, for example, latitude and longitude. Output layer is 2
It has two neurons and is a feature quantity indicating whether or not the target is flying toward the area to be defended and a feature quantity indicating the size of the target arrival time, both of which are, for example, in back propagation which is one of neural networks. Yes 4
It is composed of more than one layer. When the neural network used has four layers, the output layer is the fourth layer.

The addition and input / output functions used in the neural network A are exactly the same as in the first embodiment.

Next, in the above embodiment, for example, the neural network used for the threshold value judgment means 10 and the feature amount judgment means 11 is back propagation, and, for example,
Operation of the target information determination unit 8 when the determination result of the target information is set as large threat, medium threat, and small threat will be described with reference to the flowchart shown in FIG. ST23, 11, 24, 25,
26 shows the operation of the learning means 14. ST1, 26, 1
7 shows the operation of the threshold value judging means 10. ST18,
Reference numeral 8 indicates the operation of the feature amount determination means 11.

In ST23, in order to use the teacher signal of the neural network A of the threshold value judging means, an example of the coordinates on the target plane and the velocity vector pointing in the direction of the area to be protected is created and used as the teacher signal. To do.

In ST11, the neural network A of the threshold value judging means is learned by the teacher signal created in ST23. This is an operation in which the neural network A learns the area to be protected by learning which direction the target at a certain position is facing to the area to be protected. From Example 1 shown in FIG. By using the position and velocity vector as the teacher signal as in Example 5, the region 39 to be protected can be learned. By using this neural network, it is possible to determine whether or not the target is flying toward the protected area, using the coordinates and the velocity vector of the target as inputs.

At ST24, the neural network A relating to the target arrival time is learned by using the target arrival time and the feature amount as the teacher signal as shown in FIG. By using this neural network, the time required for the target to reach the defense area can be normalized to a value of 1 to 0.

In ST25, as shown in Table 2, the feature amount of whether or not the target is flying toward the area to be protected, the feature amount regarding the target arrival time, and the size of the threat as the result of the determination of the target information, It is created as a teacher signal of the neural network B of the feature quantity determination means 11. In Table 2, a large threat, a medium threat, and a small threat are actually numerical values 1 to 0 indicating the feature amount.

[0068]

[Table 2]

In ST26, the neural network B of the feature amount determination means 11 is made to learn the relationship between the combination of each feature amount and the determination result of the target information by the teacher signal created in ST25.

At ST1, target movement parameters are input from the observation means 7.

In ST26, the coordinates on the target plane,
Calculate the velocity vector and the target arrival time, which is the predicted time for the target to reach the defense area.

At ST17, using the value obtained at ST26 as an input, the neural network A learned at ST11 and ST24 is used to calculate the feature amount and the target arrival time regarding whether or not the target is flying toward the defense area. A feature amount regarding

In ST18, 2 calculated in ST17
The two feature quantities are input as they are, and the size of the threat is calculated as the determination result of the target information by the neural network B learned in ST26.

In ST8, the determination result of the target information obtained in ST18 is output to the display means 9.

At ST9, if new target motion parameters are obtained from the observation means 7, ST1, 26, 17, 1
8 and 8 are repeated, and if not obtained, the operation ends.

In the above embodiment, the case where back propagation is used as the neural network has been described. However, other various types of neural networks are also effective when applied to the present invention.

[0077]

Since the present invention is constructed as described above, the following effects can be obtained.

By inputting the coordinates of the target on the plane of the target and the target altitude above sea level, a neural network that stores the prohibited airspace where the flight of the aircraft is prohibited and the safe airspace where the aircraft of the friend is permitted to fly is stored. By using
It becomes possible to calculate the continuous feature amount considering the error of the target information and the ambiguity of the airspace, and based on the continuous feature amount, it is possible to accurately distinguish between the enemy and the ally of the tracking target. Can be determined. Further, by dividing the neural network into a plurality, the scale of the neural network can be reduced.

A neural network in which a function of normalizing the time when the target reaches the area protected by the missile system and the area where the target defends is input is input with the coordinates of the target on the plane of the target, the target altitude above sea level, and the velocity vector. By using it, it becomes possible to calculate a continuous feature amount that takes into consideration the ambiguity of the target information error and the area to be protected, and the tracking target can be calculated based on the continuous feature amount. It is possible to determine the magnitude of the threat that the missile system flies toward the defensive area, and increases when the target reaches the defensive area in a short time. Also,
By dividing the neural network into a plurality, it is possible to reduce the scale of the neural network.

[Brief description of drawings]

FIG. 1 is a block diagram of a target information determination device according to a first embodiment of the present invention.

FIG. 2 is a block diagram of a conventional target information determination device.

FIG. 3 is a diagram illustrating an outline of operation of a general interceptor missile system.

FIG. 4 is a flowchart showing the operation of a conventional target information determination device.

FIG. 5 is a diagram illustrating a general neural network.

FIG. 6 is a flowchart showing the operation of the target information determination device in the first embodiment of the present invention.

FIG. 7A is a diagram for explaining the operation of the threshold value judging means 10 according to the second embodiment of the present invention, and FIG.
FIG. 7 is a diagram showing a distribution of feature amounts of 0 to 1 which is the certainty when the target in the flight section D-E of FIG.

FIG. 8 is a diagram illustrating a detailed operation of learning ST19 of the neural network in the flowchart of FIG. 6 in the first embodiment of the present invention.

FIG. 9 is a block diagram of a target information determination device according to a second embodiment of the present invention.

FIG. 10 is a flowchart showing the operation of the target information determination device in the second embodiment of the present invention.

FIG. 11 is a diagram illustrating a learning method of a protected area according to the second embodiment of the present invention.

FIG. 12 is a diagram illustrating a function for normalizing the time required for the target to reach the protected area according to the second embodiment of this invention.

[Explanation of symbols]

 1 Target 2 Radar Device 3 Tracking Beam 4 Shooting Control Device 5 Intercepting Missile 6 Launching Device 7 Observing Means 8 Target Information Determining Means 9 Displaying Means 10 Threshold Determining Means 11 Feature Quantifying Means 12 Threshold Database 13 Feature Quantifying Rule database 14 Learning means 15 Neuron 16 Antenna 17 Transceiver 18 Indicator 19 Signal processor 20 Signal input device 21 Motion parameter generator 22 Feature amount calculation element 23 Judgment result calculation element 24 Airspace database 25 Teacher signal generation element 26 Feature amount Calculation learning element 27 Feature amount determination function database 28 Teacher signal generation element 29 Determination result calculation learning element 30 Display signal generator 31 Display 32 Flight prohibited airspace 33 Safe airspace 34 Target distribution of feature amount against flight prohibited airspace 35 Target safety Distribution of feature quantity for airspace 3 6 Observation range of radar equipment 37 Target arrival time generator 38 Database of defense area 39 Defense area

Claims (2)

[Claims] 1. A radar device which is an observing means for observing a flying object as a tracking target and obtaining motion parameters such as a target position and velocity and other target information, and motion parameters of the target obtained from the observing device. , A flight-prohibited airspace where the flight of the aircraft is prohibited, and a neural network trained with the safe airspace where the aircraft of the friend is allowed to fly as a threshold, and the target position and the flight-prohibited airspace and Threshold value determining means for outputting the positional relationship with the safe airspace as a continuous feature value, and the feature value for each airspace output by the threshold value determining means are input, and a pattern of feature values for each airspace and A feature quantity determination unit that has a neural network that has learned the relationship with the determination result of the target information, and outputs the target enemy or ally as the determination result of the target information, and the respective neural networks. A target information determination device, comprising: a learning unit for learning a network. 2. A radar device, which is an observing means for observing a flying object as a tracking target, and obtaining motion parameters such as a target position and velocity, and other target information, and motion parameters of the target obtained from the observing device. , And the interception missile system has a neural network trained with a threshold that is a function that normalizes the time that the interception missile system defends and the target reach the defending region, and the target moves toward the defending region. Whether or not the target is flying and whether or not the target is flying, and a threshold determination unit that outputs a normalized time until the target reaches the area to be protected as a continuous feature amount, and a feature that the threshold determination unit outputs With the amount as an input, and having a neural network that learned the relationship between the pattern of the feature amount for each area and the determination result of the target information, flying toward the area where the target defends, And target information characterized by including a feature amount determination unit that outputs the size of a threat that increases when the arrival time to the defense area is short as a determination result, and a learning unit that learns each neural network. Judgment device.