JP7616758B2 - Identification device, method, and program - Google Patents

Description

The present disclosure relates to an identification device, method, and program, and in particular to an identification device, method, and program that can easily identify flying objects.

In air traffic control, a method is known for identifying flying objects by humans using identification signals emitted by the flying object or position information of the flying object acquired by a sensor. Another method is known for identifying flying objects by comparing image information of the flying object retrieved from radar information of the flying object with the aircraft characteristics of the flying object acquired using an image capturing device, thereby improving the identification probability of the flying object, such as an aircraft.

Patent Document 1 describes that "the learning data generation method includes an acquisition step of acquiring two pieces of data from among appearance data of an aircraft on an image of a specific route, signal data of radio waves emitted from an aircraft on the route, and noise data indicating noise from an aircraft on the route, an identification step of identifying attributes of an aircraft on the route by inputting one of the two pieces of acquired data into a first identification model for identifying attributes of the aircraft, and a generation step of generating learning data used for training a second identification model for identifying attributes of the aircraft by associating the other of the two pieces of data with the attributes of the aircraft on the route identified by the identification step." Patent Document 1 does not describe "generating a learned flying object model by performing regression neural network (RNN) learning, and identifying an approximate flying object that approximates a target flying object from the learned flying object model based on time-series observation information of the target flying object and the learned flying object model."

Patent Document 2 describes that "the position of each aircraft on the airport surface is detected by an airport detection radar, and when any aircraft on the airport surface emits radio waves, a radio wave source region in which the radio wave source is located is determined by direction finders installed in at least two locations, and when only one aircraft is located in this radio wave source region, the identification information of this aircraft is selected from the contents of the radio waves emitted by the aircraft, the aircraft located in the radio wave source region is identified by the airport detection radar, and the aircraft is automatically identified by correlating the position information of the airport detection radar with the identification information of the aircraft identified by the direction finder." Patent Document 2 does not describe "generating a learned flying object model by performing regression neural network (RNN) learning, and identifying an approximate flying object that approximates the target flying object from the learned flying object model based on the time-series observation information of the target flying object and the learned flying object model."

International Publication No. 2019/176058 Japanese Patent Application Publication No. 6-342061

While there is a wide variety of flying objects and the amount of information handled is increasing, the population of workers who identify flying objects is expected to decrease. In this situation, there is a problem in that it is difficult to provide an identification device for identifying multiple flying objects that can easily identify flying objects without the help of a human or with the assistance of a human.

The objective of this disclosure is to provide an identification device, method, and program that solves the above-mentioned problems.

The identification device according to the present disclosure comprises:
An acquisition unit that acquires observation information obtained by observing a flight state of a flying object;
A storage unit that stores time-series observation information obtained by processing the observation information in a time series manner for each of the plurality of flying objects;
A learning unit that uses learning observation information obtained by converting and processing the time-series observation information into a format suitable for learning, and performs the learning using a recurrent neural network (RNN) to generate a learned flying object model;
a processing unit that processes the time-series observation information of a predetermined flying object based on the learning observation information to generate processed information;
and an identification unit that identifies the specified flying object based on the processed information and the learned flying object model.

The method according to the present disclosure comprises:
Obtaining observation information on a flight state of a flying object;
accumulating time-series observation information obtained by processing the observation information in a time series manner for each of the plurality of flying objects;
Using learning observation information obtained by converting and processing the time-series observation information into a format suitable for learning, perform the learning using a recurrent neural network (RNN) to generate a learned flying object model;
generating processed information by processing the time-series observation information of a predetermined flying object based on the learning observation information;
Identifying the predetermined flying object based on the processed information and the learned flying object model;
Equipped with.

The program according to the present disclosure is
Obtaining observation information on a flight state of a flying object;
accumulating time-series observation information obtained by processing the observation information in a time series manner for each of the plurality of flying objects;
Using learning observation information obtained by converting and processing the time-series observation information into a format suitable for learning, perform the learning using a recurrent neural network (RNN) to generate a learned flying object model;
generating processed information by processing the time-series observation information of a predetermined flying object based on the learning observation information;
Identifying the predetermined flying object based on the processed information and the learned flying object model;
to be executed by the computer.

The present disclosure provides an identification device, method, and program that can easily identify flying objects.

FIG. 1 is a block diagram illustrating an identification device according to an embodiment. 10 is a flowchart illustrating an operation of the identification device according to the embodiment. 10 is a flowchart illustrating an operation of the identification device according to the embodiment. 10 is a flowchart illustrating an operation of the identification device according to the embodiment.

Below, an embodiment of the present invention will be described with reference to the drawings. In each drawing, the same or corresponding elements are given the same reference numerals, and duplicate explanations will be omitted as necessary to clarify the explanation.

[Embodiment]
The configuration of an identification device according to an embodiment will be described.
In the embodiment, the conversion to a data format suitable for a regression neural network is taken as an example for explanation. In addition, in the embodiment, the process from information accumulation to identification using a trained flying object model is divided into three phases, an information collection phase, a learning phase, and a utilization phase.
FIG. 1 is a block diagram illustrating an identification device according to an embodiment.

As shown in FIG. 1, the identification device 11 according to the embodiment includes an acquisition unit 111, a storage unit 112, a learning unit 113, a processing unit 114, and an identification unit 115.

<Information gathering phase>
The acquisition unit 111 acquires observation information obtained by observing the flight state of a flying object, which is an object in flight. The observation information is measured, for example, by a sensor that observes the flying object. The acquisition unit 111 may acquire observation information of a simulated flying object that simulates the flying object, instead of the observation information of the flying object.

The observation information includes at least one of the following: time-series three-dimensional position information of the flying object, the moving direction of the flying object, and the moving speed of the flying object. The observation information may be information obtained by observing the flight state from a specified time in the past to the present.

The storage unit 112 stores time-series observation information obtained by processing observation information in a time series manner for each of a plurality of flying objects. The acquisition unit 111 may acquire observation information at a predetermined cycle, and the storage unit 112 may store the time-series observation information at a predetermined cycle. The time-series observation information includes observation information from the past to the present. The time-series observation information is sometimes referred to as current status information.

In other words, the acquisition unit 111 manages time-series observation information about flying objects acquired and calculated in time series from observation information of flying objects observed by a sensor (not shown), or time-series observation information of simulated flying objects that simulate flying objects, and periodically accumulates and stores the time-series observation information in the accumulation unit 112.

The identification device 11 may further include a display unit that displays time-series observation information. The acquisition unit and the display unit may be collectively referred to as a current status display unit.

The accumulation unit 112 may accumulate time-series observation information when the identification of a flying object is successful. Specifically, when the identification device 11 identifies a flying object managed by the identification device 11 and the identification is successful, the accumulation unit 112 may store the time-series observation information together with information related to the identification. In addition, when an operator identifies a flying object managed by the operator and the identification is successful, the accumulation unit 112 may store the time-series observation information together with information related to the identification.

The storage unit may also be referred to as a data storage unit. Since the observation information is information measured by a sensor that observes the flying object, the observation information may also be referred to as sensor information.

<Learning Phase>
The learning unit 113 uses the learning observation information obtained by converting and processing the time-series observation information into a format suitable for learning, and performs learning using a recurrent neural network (RNN) to generate a learned flying object model. That is, the learning unit 113 acquires the time-series observation information accumulated from the storage unit 112, processes the acquired time-series observation information into a format suitable for learning, and performs learning using the recurrent neural network to generate a learned flying object model. Processing into a format suitable for learning refers to, for example, standardizing the time-series observation information or converting it into a dummy variable in machine learning.

The learning unit 113 stores the used learning observation information and the generated learned flying object model. Since the learning observation information is information that has been converted and processed into a format suitable for learning, it is sometimes referred to as data processing information.

The learning unit 113 saves the learning observation information and the learned flying object model, and then evaluates the learned flying object model generated by learning. If the learned flying object model passes the evaluation, the learning unit 113 outputs the learning observation information at the time of learning to the processing unit 114, and outputs the learned flying object model to the identification unit 115. If the learned flying object model passes the evaluation, for example, the learning error by the regression neural network is equal to or less than the error threshold, in which case the learning unit 113 ends the learning.

The learning unit may also be referred to as an RNN learning unit. The processing unit may also be referred to as a data processing unit. The identification unit may also be referred to as an identification processing unit.

<Utilization phase>
The processing unit 114 periodically acquires time-series observation information on a predetermined flying object acquired and calculated from the observation information managed by the acquisition unit 111, or time-series observation information of a simulated predetermined flying object that simulates the predetermined flying object. Here, the predetermined flying object is a flying object that is a target to be identified, for example, a flying object newly observed by a sensor.

The processing unit 114 acquires the learning observation information during learning from the learning unit 113. The processing unit 114 generates processed information by processing the time-series observation information acquired from the acquisition unit 111 based on the learning observation information, and outputs the generated processed information to the identification unit 115. In other words, the processing unit 114 generates processed information by processing the time-series observation information of a specified flying object based on the learning observation information. The processing unit 114 stores the processed information acquired from the learning unit 113.

The processed information is the output when learning observation information (data processing information) and time series observation information (information on the three-dimensional position, direction and speed of a flying object, etc.) are input to the processing unit 114.

The identification unit 115 performs identification using the processing information acquired from the processing unit 114 and the learned flying object model acquired from the learning unit 113, and outputs the identification result to the acquisition unit 111. That is, the identification unit 115 identifies the specified flying object based on the processing information and the learned flying object model.

The identification unit 115 also stores the learned flying object model acquired from the learning unit 113. The acquisition unit 111 displays the identification result acquired from the identification unit 115, i.e., information about the identified specific flying object, on a display unit (not shown).

The identification device 11 according to the embodiment identifies a specific flying object newly observed by a sensor based on a learned flying object model learned from observation information (time-series observation information) of flying objects from the past to the present. This makes it possible to provide an identification device, method, and program that can easily identify flying objects without relying on the skill of an individual.

In addition, the identification device 11 according to the embodiment can reduce the workload of people compared to when people identify flying objects, and therefore the number of workers required can be reduced.

In addition, when the observation information acquired from the sensor is used for learning in the identification device 11, the observation information required for learning can be easily stored in the storage unit 112. Therefore, there is no need to add new work to collect observation information.

In addition, the observational information used during learning can be not only that obtained from sensors, but also simulated observational information that simulates the observational information obtained from sensors. This can shorten the time required to collect the information necessary for learning, and reduce the costs of building equipment to collect new information, the costs of maintaining the equipment, and the personnel costs involved.

The operation of the identification device according to the embodiment will be described.
FIG. 2 is a flowchart illustrating the operation of the identification device according to the embodiment.
FIG. 2 shows the process flow of the information gathering phase.

As shown in FIG. 2, the storage unit 112 stores current status information (time-series observation information) (step S101). The storage unit 112 periodically stores time-series observation information about flying objects acquired and calculated from the observation information managed by the acquisition unit 111, or time-series observation information about simulated flying objects that simulate flying objects.

The acquisition unit 111 checks whether or not to perform identification on the flying object managed by the acquisition unit 111 (step S102).

If the flying object is identified in step S102 (step S102: Yes), information regarding the identification of the flying object is stored in the storage unit 112 (step S103).

If the flying object is not identified in step S102 (step S102: No), an end determination is made (step S104).

If it is not the end in step S104 (step S104: No), return to step S101.

In step S104, if the process is to end (step S104: Yes), i.e., if the monitoring of the current situation is to end, the display of the current situation information is terminated.

FIG. 3 is a flowchart illustrating the operation of the identification device according to the embodiment.
FIG. 3 shows the process flow of the learning phase.

Here, the error threshold for determining when learning is to be completed in the learning performed by the learning unit 113 is defined in advance as L.

As shown in FIG. 3, the learning unit 113 acquires the accumulated time-series observation information from the accumulation unit 112, and converts and processes it into a format suitable for learning (step S201). The processed information is referred to as learning observation information.

The learning unit 113 uses the processed learning observation information to perform learning using a recurrent neural network (RNN) (step S202).

The learning unit 113 checks the evaluation result. That is, the learning unit 113 compares the learning error e by the regression neural network with the error threshold L (step S203).

If the error e is less than or equal to the error threshold L (step S203: Yes), the learning unit 113 stores the learning observation information in the processing unit 114 (step S204).

If the error e is not less than the error threshold L (step S203: No), the learning unit 113 returns to step S201.

After step S204, the learned flying object model generated in step S202 is stored in the identification unit 115 (step S205).

FIG. 4 is a flowchart illustrating the operation of the identification device according to the embodiment.
FIG. 4 shows the process flow of the utilization phase.

As shown in FIG. 4, the processing unit 114 acquires time-series observation information about a flying object acquired and calculated from the observation information managed by the acquisition unit 111, or time-series observation information of a simulated flying object that simulates a flying object, and processes this to generate processed information (step S301).

The identification unit 115 identifies the flying object using the processing information acquired from the processing unit 114 and the learned flying object model acquired from the learning unit 113 (step S302).

The acquisition unit 111 displays the flying object identification result on the display unit (step S303).

After step S303, a termination determination is made (step S304).

If it is not the end in step S304 (step S304: No), return to step S301.

In step S304, if the end is determined (step S304: Yes), i.e., if monitoring of the current situation is to be ended, the display of the current situation is ended.

The features of the identification device 11 according to the embodiment will be described below.
The identification device 11 identifies a target flying object using a regression neural network based on time-series observation information regarding the flying object acquired and calculated from observation information (sensor information) or time-series observation information simulating the flying object.
The identification device 11 includes an acquisition unit 111 , a storage unit 112 , a learning unit 113 , a processing unit 114 , and an identification unit 115 .
The acquisition unit (current status display unit) 111 displays and manages time-series observation information regarding flying objects acquired and calculated from observation information or time-series observation information simulating flying objects, and also inputs, displays, and manages information on identified flying objects.
The storage unit 112 stores the observation information managed by the acquisition unit 111 .
The learning unit 113 uses the observation information stored in the storage unit 112 to convert the time-series observation information into learning observation information required for learning, and generates a learned flying object model.
The processing unit 114 acquires the observation information managed by the acquisition unit 111 , and outputs processed information that is processed based on the learned observation information to the identification unit 115 .
The identification unit 115 uses the processed information and the learned flying object model to output the result of identifying the flying object to the acquisition unit 111.
The learning unit 113 processes the observation information obtained from the storage unit 112, and then learns the identification rules by using a regression neural network, and stores the used learning observation information and the generated learned flying object model.

In addition, in the embodiment, since learning is performed using only time series information and information on the identification results, it can be applied not only to flying objects, but also to identifying moving objects on land, on water, or underwater, such as vehicles, ships, and submarines.

In the above embodiment, the present invention has been described as a hardware configuration, but the present invention is not limited to this. The present invention can also be realized by having a CPU (Central Processing Unit) execute a computer program to process each component.

In the above embodiment, the program can be stored and supplied to the computer using various types of non-transitory computer readable media. The non-transitory computer readable media includes various types of tangible storage media. Examples of the non-transitory computer readable media include magnetic recording media (specifically, flexible disks, magnetic tapes, and hard disk drives), magneto-optical recording media (specifically, magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-R/Ws, semiconductor memories (specifically, mask ROMs, PROMs (Programmable ROMs), EPROMs (Erasable PROMs)), flash ROMs, and RAMs (Random Access Memory). The program may also be supplied to the computer by various types of transitory computer readable media. Examples of the transitory computer readable media include electrical signals, optical signals, and electromagnetic waves. The transitory computer readable media can supply the program to the computer via wired communication paths such as electric wires and optical fibers, or wireless communication paths.

The present invention has been described above with reference to the embodiment, but the present invention is not limited to the above. Various modifications that can be understood by a person skilled in the art can be made to the configuration and details of the present invention within the scope of the invention.

The present invention is not limited to the above embodiment, and can be modified as appropriate without departing from the spirit and scope of the invention.

11: Recognition device 111: Acquisition unit 112: Storage unit 113: Learning unit 114: Processing unit 115: Recognition unit e: Error L: Error threshold

Claims (9)

In the information gathering phase,
An acquisition unit that acquires observation information obtained by observing a flight state of a flying object;
A storage unit that stores, for each of the plurality of flying objects, time-series observation information obtained by processing , in time series, simulated observation information that simulates the observation information together with information regarding the identification of the flying object;
During the learning phase,
A learning unit that uses learning observation information obtained by converting and processing the time-series observation information into a format suitable for learning, and performs learning using a recurrent neural network (RNN) to generate a learned flying object model;
In the utilization phase,
A processing unit that generates processed information processed by the same processing as that used when generating the trained flying object model in order to match an information format of a predetermined time series observation information of a predetermined flying object with an information format of the trained observation information;
An identification unit that identifies the predetermined flying object based on the processing information and the learned flying object model;
An identification device comprising: The observation information includes at least one of time-series three-dimensional position information of the flying object, a moving direction of the flying object, and a moving speed of the flying object;
The information regarding the identification of the flying object includes the type of the flying object, the flight number, the route, or information regarding whether the flying object is an aircraft;
The identification device according to claim 1 . The acquisition unit acquires the observation information at a predetermined period.
The identification device according to claim 1 or 2 . The storage unit stores the predetermined time series observation information when the identification of the predetermined flying object is successful,
A case where the flying object is successfully identified indicates a case where the observation information of the flying object is one of the learned flying object models.
An identification device according to any one of claims 1 to 3 . the learning unit terminates the learning when an error in the learning by the regression neural network is equal to or smaller than an error threshold.
An identification device according to any one of claims 1 to 4 . A display unit that displays the time-series observation information and information about the identified predetermined flying object is further provided.
An identification device according to any one of claims 1 to 5 . The observation information is information obtained by observing the flight state from a predetermined time in the past to the present.
An identification device according to any one of claims 1 to 6 . In the information gathering phase,
Obtaining observation information on a flight state of a flying object;
accumulating, for each of the plurality of flying objects, time-series observation information obtained by processing , in a time series manner, simulated observation information that simulates the observation information together with information regarding the identification of the flying object;
In the learning phase,
Using learning observation information obtained by converting and processing the time-series observation information into a format suitable for learning, a recurrent neural network (RNN) is used to generate a learned flying object model;
In the utilization phase,
generating processed information that is processed in the same manner as the processing performed when the learned flying object model was generated, so that the information format of the predetermined time series observation information of the predetermined flying object coincides with the information format of the learned observation information;
Identifying the predetermined flying object based on the processed information and the learned flying object model;
A method for providing the above. In the information gathering phase,
Obtaining observation information on a flight state of a flying object;
accumulating, for each of the plurality of flying objects, time-series observation information obtained by processing , in a time series manner, simulated observation information that simulates the observation information together with information regarding the identification of the flying object;
In the learning phase,
Using learning observation information obtained by converting and processing the time-series observation information into a format suitable for learning, a recurrent neural network (RNN) is used to generate a learned flying object model;
In the utilization phase,
generating processed information that is processed in the same manner as the processing performed when the learned flying object model was generated, so that the information format of the predetermined time series observation information of the predetermined flying object coincides with the information format of the learned observation information;
Identifying the predetermined flying object based on the processed information and the learned flying object model;
A program that causes a computer to execute the following.

Images