Learning model, signal processing device, flying body, and teacher data generation method

Abstract

To enable the detection of a target, such as a ground object, using RAW data.SOLUTION: A learning model causes a computer to function so as to receive a first synthetic aperture radar reception signal based on a reflected electromagnetic wave, which is reflection of an electromagnetic wave with which a learning region is irradiated, to learn using teacher data, where an output is first target information of a target in the learning region corresponding to a feature amount of the first synthetic aperture radar reception signal, and to output, in response to an input of a second synthetic aperture radar reception signal based on a reflected electromagnetic wave which is reflection of an electromagnetic wave with which a detection region is irradiated, second target information of the target in the detection region corresponding to a feature amount of the second synthetic aperture radar reception signal.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a learning model, a signal processing apparatus, an aircraft, and a program.

Background Art

[0002] Observation of the state of the earth's surface including land and sea using aircraft such as artificial satellites, airplanes, or drone devices is widely performed. Observation methods using artificial satellites include observation methods performed by acquiring a radar image so-called SAR image obtained using synthetic aperture radar (SAR: Synthetic Aperture Radar) technology, and observation methods performed by acquiring an optical image and an SAR image and combining the two images. Patent Document 1 describes detecting an object without performing SAR imaging processing using raw data (RAW data) obtained by a synthetic aperture radar. (SAR: Synthetic Aperture Radar) technology, and observing methods such as obtaining a radar image, a so-called SAR image, and observing methods such as obtaining an optical image and an SAR image and combining the two images. Patent Document 1 describes detecting an object without performing SAR imaging processing using raw data (RAW data) obtained by a synthetic aperture radar. using the raw data (RAW data) obtained by a synthetic aperture radar, passing through the SAR imaging process, and detecting the object.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Regarding how to detect an object such as a ground object using RAW data, various methods can be considered. Therefore, an object of the present invention is to provide a learning model, a signal processing apparatus, an aircraft, and a teacher data generation method that enable the detection of an object such as a ground object using RAW data.

Means for Solving the Problems

[0005] ​​ The learning model according to one aspect of the present invention is a learning region that is a reflection electric field generated by reflecting an electromagnetic wave irradiated on the learning region. The feature values of the first synthetic aperture radar received signal based on magnetic waves are used as input, and the target in the learning area is detected. The first object information is learned using training data having the first object information as an output, and the electromagnetic wave irradiated to the detection area is detected. A feature quantity of a second synthetic aperture radar received signal based on a reflected electromagnetic wave is input, The computer is operated to output second object information of the object in the detection region.

[0006] According to this aspect, the target information is output based on the feature amount of the synthetic aperture radar received signal. This makes it possible to detect objects such as ground features using RAW data.

[0007] In the above aspect, the feature amount of the first synthetic aperture radar reception signal is The characteristic quantity of the second synthetic aperture radar received signal includes a frequency spectrum of the signal, and the .... It may include a frequency spectrum of the radar received signal.

[0008] In the above aspect, the feature amount of the first synthetic aperture radar reception signal is The feature amount of the second synthetic aperture radar received signal includes a signal strength of the second synthetic aperture radar received signal. The information may include the signal strength of the received signal.

[0009] In the above aspect, the feature amount of the first synthetic aperture radar reception signal is The feature quantity of the second synthetic aperture radar received signal includes a phase of the signal, and the feature quantity of the second synthetic aperture radar received signal includes a phase of the signal. The phase of the signal may also be included.

[0010] According to these aspects, the characteristic quantities of the synthetic aperture radar received signal include a frequency spectrum, Since object information based on signal strength or phase can be output, RAW data can be used to detect objects such as ground features.

[0011] In the above aspect, the first synthetic aperture radar received signal is a signal corresponding to a part of the learning region that satisfies a predetermined condition for the feature amount, and the feature amount of the first synthetic aperture radar received signal may include a feature amount based on the object in the learning region.

[0012] According to this aspect, the received signal that is the source of the feature amount of the received signal used as the teacher data of the learning model is a signal corresponding to a region where the feature amount satisfies a predetermined condition, so the amount of data required for learning is reduced. Also, since the feature amount of the received signal includes a feature amount based on the object in the learning region, it is possible to detect objects such as ground features using RAW data while reducing the amount of data.

[0013] In the above aspect, the second synthetic aperture radar received signal is a signal corresponding to a part of the detection region that satisfies a predetermined condition for the feature amount, and the feature amount of the second synthetic aperture radar received signal may include a feature amount based on the object in the detection region.

[0014] According to this aspect, regarding the feature amount of the second synthetic aperture radar received signal used in the inference using the learning model, it can also be based on a signal corresponding to a part of the region where the feature amount satisfies a predetermined condition, so the amount of data required for inference is reduced. Also, since the feature amount of the second synthetic aperture radar received signal includes a feature amount based on the object in the learning region, it is possible to detect objects such as ground features using RAW data while reducing the amount of data. .

[0015] In the above aspect, the first object information may include information indicating the attributes of the object in the learning region, and the second object information may include information indicating the attributes of the object in the detection region. According to this aspect, since information indicating the attributes of the object can be obtained, it is possible to detect the object while obtaining more detailed information about the object.

[0016] According to this aspect, since information indicating the attributes of the object can be obtained, it is possible to detect the object while obtaining more detailed information about the object. According to this aspect, since information indicating the attributes of the object can be obtained, it is possible to detect the object while obtaining more detailed information about the object.

[0017] An information processing apparatus according to another aspect of the present invention includes a storage unit that stores a learning model, a signal acquisition unit that acquires a second synthetic aperture radar reception signal, and an estimation unit that inputs the second synthetic aperture radar reception signal to the learning model and infers second object information. An information processing apparatus according to another aspect of the present invention includes a storage unit that stores a learning model, a signal acquisition unit that acquires a second synthetic aperture radar reception signal, and an estimation unit that inputs the second synthetic aperture radar reception signal to the learning model and infers second object information. An information processing apparatus according to another aspect of the present invention includes a storage unit that stores a learning model, a signal acquisition unit that acquires a second synthetic aperture radar reception signal, and an estimation unit that inputs the second synthetic aperture radar reception signal to the learning model and infers second object information.

[0018] An aircraft according to another aspect of the present invention includes a storage unit that stores a learning model, a signal acquisition unit that acquires a second synthetic aperture radar reception signal, an estimation unit that inputs the second synthetic aperture radar reception signal to the learning model and infers second object information, and a signal output unit that outputs an output signal based on the second object information to the outside. An aircraft according to another aspect of the present invention includes a storage unit that stores a learning model, a signal acquisition unit that acquires a second synthetic aperture radar reception signal, an estimation unit that inputs the second synthetic aperture radar reception signal to the learning model and infers second object information, and a signal output unit that outputs an output signal based on the second object information to the outside. An aircraft according to another aspect of the present invention includes a storage unit that stores a learning model, a signal acquisition unit that acquires a second synthetic aperture radar reception signal, an estimation unit that inputs the second synthetic aperture radar reception signal to the learning model and infers second object information, and a signal output unit that outputs an output signal based on the second object information to the outside. An aircraft according to another aspect of the present invention includes a storage unit that stores a learning model, a signal acquisition unit that acquires a second synthetic aperture radar reception signal, an estimation unit that inputs the second synthetic aperture radar reception signal to the learning model and infers second object information, and a signal output unit that outputs an output signal based on the second object information to the outside.

[0019] A teacher data generation method according to another aspect of the present invention includes a computer acquiring a first synthetic aperture radar reception signal based on a reflected electromagnetic wave obtained by reflecting an electromagnetic wave irradiated on a learning region, extracting a feature amount of a learning signal corresponding to a region where an object exists in the learning region from the first synthetic aperture radar reception signal, acquiring object information of the object, and using a pair of the feature amount of the learning signal and the object information as an input to the feature amount of a second synthetic aperture radar reception signal based on a reflected electromagnetic wave obtained by reflecting an electromagnetic wave irradiated on a detection region. A teacher data generation method according to another aspect of the present invention includes a computer acquiring a first synthetic aperture radar reception signal based on a reflected electromagnetic wave obtained by reflecting an electromagnetic wave irradiated on a learning region, extracting a feature amount of a learning signal corresponding to a region where an object exists in the learning region from the first synthetic aperture radar reception signal, acquiring object information of the object, and using a pair of the feature amount of the learning signal and the object information as an input to the feature amount of a second synthetic aperture radar reception signal based on a reflected electromagnetic wave obtained by reflecting an electromagnetic wave irradiated on a detection region. A teacher data generation method according to another aspect of the present invention includes a computer acquiring a first synthetic aperture radar reception signal based on a reflected electromagnetic wave obtained by reflecting an electromagnetic wave irradiated on a learning region, extracting a feature amount of a learning signal corresponding to a region where an object exists in the learning region from the first synthetic aperture radar reception signal, acquiring object information of the object, and using a pair of the feature amount of the learning signal and the object information as an input to the feature amount of a second synthetic aperture radar reception signal based on a reflected electromagnetic wave obtained by reflecting an electromagnetic wave irradiated on a detection region. A teacher data generation method according to another aspect of the present invention includes a computer acquiring a first synthetic aperture radar reception signal based on a reflected electromagnetic wave obtained by reflecting an electromagnetic wave irradiated on a learning region, extracting a feature amount of a learning signal corresponding to a region where an object exists in the learning region from the first synthetic aperture radar reception signal, acquiring object information of the object, and using a pair of the feature amount of the learning signal and the object information as an input to the feature amount of a second synthetic aperture radar reception signal based on a reflected electromagnetic wave obtained by reflecting an electromagnetic wave irradiated on a detection region. A teacher data generation method according to another aspect of the present invention includes a computer acquiring a first synthetic aperture radar reception signal based on a reflected electromagnetic wave obtained by reflecting an electromagnetic wave irradiated on a learning region, extracting a feature amount of a learning signal corresponding to a region where an object exists in the learning region from the first synthetic aperture radar reception signal, acquiring object information of the object, and using a pair of the feature amount of the learning signal and the object information as an input to the feature amount of a second synthetic aperture radar reception signal based on a reflected electromagnetic wave obtained by reflecting an electromagnetic wave irradiated on a detection region. A teacher data generation method according to another aspect of the present invention includes a computer acquiring a first synthetic aperture radar reception signal based on a reflected electromagnetic wave obtained by reflecting an electromagnetic wave irradiated on a learning region, extracting a feature amount of a learning signal corresponding to a region where an object exists in the learning region from the first synthetic aperture radar reception signal, acquiring object information of the object, and using a pair of the feature amount of the learning signal and the object information as an input to the feature amount of a second synthetic aperture radar reception signal based on a reflected electromagnetic wave obtained by reflecting an electromagnetic wave irradiated on a detection region. Output second object information of an object in a detection area according to a feature amount of a radar reception signal And store it as teacher data used for learning of a learning model that causes a computer to function including that.

Advantages of the Invention

[0020] According to the present invention, it is possible to detect an object such as a ground object using RAW data.

Brief Description of the Drawings

[0021]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Embodiments for Carrying Out the Invention

[0022] With reference to the accompanying drawings, preferred embodiments of the present invention will be described. In each figure , those denoted by the same reference numerals have the same or similar configurations.

[0023] FIG. 1 shows a block diagram of an observation system 10 according to the present embodiment. The observation system 10 includes a flying object 100 and an observation device 200. The flying object 100 is arranged in the space above the earth and the observation device 200 is arranged on the earth. In the present embodiment, the flying object 100 observes the detection area D on the earth's surface by radar, and the observation signal O processed in the flying object 100 is transmitted to the observation device 200. The observation signal O is, for example, the received signal obtained by the flying object 100 as described later or the received signal corresponding thereto, and the object information such as a ground object or a ship on the sea in the detection area D. It is a signal indicating object information.

[0024] The flying object 100 includes a communication antenna 101, a radar device 102, and a signal processing device 103. The flying object 100 is an artificial satellite capable of acquiring and processing received signals, and is arranged in space and orbits around the earth. Note that the flying object 100 may be a geostationary satellite. Also , the flying object 100 may be a device capable of being located above the earth, such as an aircraft, a helicopter, or a drone device.

[0025] The communication antenna 101 is an antenna for the flying object 100 to communicate with an external device provided on the earth or in space.

[0026] The radar device 102 irradiates an electromagnetic wave EM1, which is, for example, a microwave, onto a detection area D on the Earth's surface. The reflected electromagnetic wave EM2 reflected by the object to be observed in the detection area D is acquired. The radar device 102 is, for example, a synthetic aperture radar (SAR). The reflected electromagnetic wave EM2 is processed and recorded by the radar device 102 as a received signal (RAW data) based on the fluctuation of the electromagnetic wave that can be handled by the flying object 100. The received signal is recorded, for example, as a complex number signal corresponding to each coordinate of the detection area D. The radar device 102 transmits the observation signal O to the observation device 200 through the communication antenna 101. The radar device 102 includes a processor for controlling the acquisition process of the received signal and a storage device that stores the programs necessary for the control. The signal processing device 103 is an information processing device that processes the received signal acquired by the radar device 102. The signal processing device 103 has a storage area such as a memory, and is a computer that performs predetermined processing by the processor executing the programs stored in the storage area.

[0027] The signal processing device 103 has a storage unit 104 and a control unit 105. The storage unit 104 is, for example, a semiconductor memory such as a RAM or an optical disk. The storage unit 104 stores various information used for the processing in the signal processing device 103.

[0028] The signal processing device 103 has a storage unit 104 and a control unit 105. The storage unit 104 stores a learning model 1041. The learning model 1041 is a program learned to take the received signal as an input and output object information according to the feature amount of the received signal. The storage unit 104 stores a learning model 1041. The learning model 1041 is a program learned to take the received signal as an input and output object information according to the feature amount of the received signal.

[0029] The signal processing device 103 has a storage unit 104 and a control unit 105. The storage unit 104 is, for example, a semiconductor memory such as a RAM or an optical disk. The storage unit 104 stores various information used for the processing in the signal processing device 103. For example, the storage unit 104 stores various information used for the processing in the signal processing device 103. The storage unit 104 stores various information used for the processing in the signal processing device 103.

[0030] The learning model 1041 is stored in the storage unit 104. The learning model 1041 is a program learned to take the received signal as an input and output object information according to the feature amount of the received signal. The learning model 1041 is a program learned to take the received signal as an input and output object information according to the feature amount of the received signal. It is mu. Details of the feature amount and the learning model 1041 in this embodiment will be described later. .

[0031] The control unit 105 performs signal processing in the signal processing device 103. Further, the control unit 105 controls the transmission of the processing result of the received signal through the flying object 100. The control unit 105 includes a signal acquisition unit 1051, an estimation unit 1052, and a signal output unit 1053.

[0032] The signal acquisition unit 1051 acquires a received signal from the radar device 102.

[0033] The estimation unit 1052 inputs the feature amount of the received signal acquired by the signal acquisition unit 1051 into the learning model 1041 and acquires object information from the learning model 1041.

[0034] The signal output unit 1053 outputs the object information acquired by the estimation unit 1052 as an observation signal O to the observation device 200 through the communication antenna 101. Further, the signal output unit 1053 may output the received signal corresponding to the object information as the observation signal O together with the object information. .

[0035] The observation device 200 is a device that transmits a control signal for controlling the observation of the detection area D by the flying object 100 to the flying object 100 and acquires the observation signal O from the flying object 100. The observation device 200 has a communication unit 201 including an antenna and a control unit for controlling communication by the antenna. The transmission and reception of information with the flying object 100 are performed through the communication unit 201.

[0036] The signal processing unit 202 processes the observation signal O from the flying object 100. The signal processing unit 202, based on the observation signal O acquired from the flying object 100, for example, the observation result in the detection area D ​ Perform processing to visualize the fruit by an image.

[0037] With reference to FIGS. 2 and 3, the learning of the learning model 1041 according to the present embodiment will be described. Herein.

[0038] FIG. 2 is a diagram schematically illustrating the learning and inference of the learning model 1041. The learning model 1041 is learned using the learning data LD1 as teacher data. The learning data LD includes a set of a feature amount of a reception signal R0 (first synthetic aperture radar reception signal) acquired by irradiating electromagnetic waves to a certain region (learning region) and object information TI0 (first object information) corresponding to the feature amount of the reception signal R0. The learning model 1041 is learned with the feature amount of the reception signal R0 as an input and the object information TI0 as an output. Here, the feature amount in the present embodiment is information including the frequency spectrum, signal intensity, or phase of the reception signal R0. The learning model 1041 is learned using the learning data LD1 as teacher data. The learning data LD includes a set of a feature amount of a reception signal R0 (first synthetic aperture radar reception signal) acquired by irradiating electromagnetic waves to a certain region (learning region) and object information TI0 (first object information) corresponding to the feature amount of the reception signal R0. The learning model 1041 is learned with the feature amount of the reception signal R0 as an input and the object information TI0 as an output. Here, the feature amount in the present embodiment is information including the frequency spectrum, signal intensity, or phase of the reception signal R0. a feature amount of a reception signal R0 (first synthetic aperture radar reception signal) acquired by irradiating electromagnetic waves to a certain region (learning region) and object information TI0 (first object information) corresponding to the feature amount of the reception signal R0. a feature amount of a reception signal R0 (first synthetic aperture radar reception signal) acquired by irradiating electromagnetic waves to a certain region (learning region) and object information TI0 (first object information) corresponding to the feature amount of the reception signal R0. The learning model 1041 is learned with the feature amount of the reception signal R0 as an input and the object information TI0 as an output. Here, the feature amount in the present embodiment is information including the frequency spectrum, signal intensity, or phase of the reception signal R0. The learning model 1041 is learned with the feature amount of the reception signal R0 as an input and the object information TI0 as an output. Here, the feature amount in the present embodiment is information including the frequency spectrum, signal intensity, or phase of the reception signal R0. Here, the feature amount in the present embodiment is information including the frequency spectrum, signal intensity, or phase of the reception signal R0.

[0039] FIG. 3 shows the reception signal R0 and the feature amount of the reception signal when a learning region D0 in which two objects, object O1 and object O2, exist is observed. FIG. 3 shows the reception signal R0 and the feature amount of the reception signal when a learning region D0 in which two objects, object O1 and object O2, exist is observed.

[0040] When the learning region D0 is observed, the reception signal R0 is obtained. Based on the reception signal R0, the feature amount (frequency spectrum, signal intensity, or phase) of the reception signal R0 is calculated. For associating the object information TI0 with the feature amount of the reception signal R0, information processing is performed to make the object information TI0 corresponding to the reception signal R0 understandable to the user. Based on the reception signal R0, the feature amount (frequency spectrum, signal intensity, or phase) of the reception signal R0 is calculated. For associating the object information TI0 with the feature amount of the reception signal R0, information processing is performed to make the object information TI0 corresponding to the reception signal R0 understandable to the user. For associating the object information TI0 with the feature amount of the reception signal R0, information processing is performed to make the object information TI0 corresponding to the reception signal R0 understandable to the user. For associating the object information TI0 with the feature amount of the reception signal R0, information processing is performed to make the object information TI0 corresponding to the reception signal R0 understandable to the user.

[0041] For example, based on the reception signal R0, the feature amount of the reception signal R0 is generated as a graph, and the user may associate the object information TI0 based on the graph. Further, the reception signal R0 and the object For example, based on the reception signal R0, the feature amount of the reception signal R0 is generated as a graph, and the user may associate the object information TI0 based on the graph. Further, the reception signal R0 and the object The received signal R0 is then processed by a computer using a learning model that associates the received signal R0 with the object information T0. The feature quantity of the received signal R0 may be associated with object information. The information may be converted to information that can be understood by the user through a predetermined conversion process for the purpose of the feature. The object information TI0 associated with the quantity is obtained by observing the learning area D0 without going through the above-mentioned imaging. For example, when detecting a ship, the target information may be obtained from other devices. The report TI0 includes information obtained by processing the received signal R0, as well as the Automatic Identification System (AIS). IS) may be considered as information obtained using the

[0042] The target information TI0 is the feature quantity (frequency spectrum, signal strength, or phase) of the received signal R0. ) The received signal R0 is, for example, I(t)+jQ(t) (j is The signal is obtained in the form of a complex number, with t being the imaginary unit and t being the time. When written as a formula, the received signal R0 is A(t)e jθ(t) In this case, The signal strength of the received signal R0 is A(t), the phase of the received signal R0 is θ(t), and the detection For each coordinate in the domain D0, the signal strength and phase are obtained. By performing frequency conversion on the received signal R0, the frequency spectrum of the received signal R0 is Torque X(ω) can be obtained.

[0043] The learning model 1041 responds to the input of these features calculated from the received signal R0. For example, the learning model 1041 is trained to output object information for the received signal R0. The control unit 105 calculates the feature amount based on the feature amount, and outputs the object information using the feature amount as an input. It is learned to do so. Note that the learning model 1041 has the received signal R0 itself inputted , and may be learned to output object information. In this case, in the learning model 1041 calculation of feature amounts based on the received signal R0 may be performed

[0044] The object information T10 includes information on the probability distribution of the probability that an object exists in each pixel of the learning region . Further, the object information TI0 includes, for example, the number of objects existing in the learning region and the information on the attributes of the objects . The attributes of an object are, for example, various information regarding the object, such as the type of the object (building, moving body, terrain, natural object, etc.), the size of the object, the moving speed of the object, etc . The information on the attributes of an object may be the probability of classifying the object into an item of a certain attribute . .

[0045] The learning model 1041 is learned by a general machine learning method such as a method using a neural network, for example, using the learning data LD1 prepared as described above . Note that the learning model 1041 may be configured as a single learning model, or may be configured as a learning model combining a plurality of learning models . .

[0046] Based on the electromagnetic wave irradiated to the detection region by the flying object 100, when the feature amount of the received signal R1 (second synthetic aperture radar received signal) acquired by the flying object 100 is input to the learned learning model 104 1, the learning model 1041 outputs object information TI1 (second object information) . .

[0047] With reference to FIGS. 4 and 5, the processing by the flying object 100 will be described. In the processing in FIGS. 4 and 5 , the object information output by the learning model 1041 is the frequency spectrum of the received signal R1 It is assumed that the information is according to the spectrum.

[0048] In step S401 of FIG. 4, the radar device 102 emits an electromagnetic wave EM1 to the detection area D1. The timing of the irradiation may be the timing controlled by the observation device 200, or may be the timing specified in advance in the flying object 100. As shown in FIG. 5, objects O3, O4, and O5 exist in the detection area D1.

[0049] In step S402, the signal acquisition unit 1051 acquires, from the radar device 102, a reception signal R1 based on the reflected electromagnetic wave EM2 detected by the radar device 102.

[0050] In step S403, the estimation unit 1052 inputs the feature amount of the reception signal R1 into the learning model 1 041. At this time, the estimation unit 1052 performs an operation to calculate the feature amount (for example, frequency spectrum) of the reception signal R1.

[0051] In step S404, the estimation unit 1052 acquires object information MD1 corresponding to the feature amount of the reception signal R1 from the learning model 1041. The object information includes, as an example, the object information TI1a and the object information TI1b shown in FIG. 5. The object information TI1a is a probability distribution of the probability that an object exists in the detection area. In the object information TI1a, areas A1, A2, and A3 are shown as areas where the probability of the existence of an object is high. The object information TI 1b is the number of objects and the attributes of the objects in the detection area. In this example, three objects are detected. 1b is the number and attributes of the objects in the detection area. In this example, three objects are detected.

[0052] In step S405, the signal output unit 1053 outputs an output signal based on the object information TI1. Output the number to the observation device 200. The output signal based on the object information TI1 is a signal that transmits all of the object information TI 1 or a part of the object information TI1. Alternatively, the output signal may be a signal that transmits information that is the result of information processing performed on the object information.

[0053] Referring to FIGS. 6 and 7, a case will be described in which the object information output by the learning model 1041 is information based on the signal strength of the received signal R1. As shown in FIG. 7, it is assumed that objects O3, O4, and O5 exist in the detection area D1 in the same manner as in the case of FIG. 5.

[0054] In step S601 of FIG. 6, the radar device 102 irradiates the detection area D1 with the electromagnetic wave EM1. In step S602, the signal acquisition unit 1051 acquires, from the radar device 10 2, the received signal R1 based on the reflected electromagnetic wave EM2 detected by the radar device 102.

[0055] In step S603, the estimation unit 1052 inputs the signal strength of the received signal R1 as a feature amount of the received signal R1 to the learning model 1041. The received signal R1 is, for example, a signal capable of generating the image IG1 shown in FIG. 7. The image IG1 can take various forms according to the signal. As an example, in the image IG1, the signal strength obtained by detecting each object may be plotted, and the change in the signal strength may be shown like a pattern. The estimation unit 1052 acquires from the learning model 1041 the object information MD1 corresponding to the signal strength of the received signal R1. The object information is, as an example, as shown in FIG. 7, and includes the object information TI 1a and the object information TI1b in the same manner as the case shown in FIG. 5. 1.

[0056] ​​​​​​In step S605, the signal output unit 1053 outputs an output signal based on the object information TI1 to the observation device 200.

[0057] With reference to FIG. 8, a case where the object information output by the learning model 1041 is information based on the phase of the received signal R1 will be described. In step S801 of FIG. 8, the radar device 102 irradiates the detection area D1 with the electromagnetic wave EM1. In step S602, the signal acquisition unit 1051 acquires the received signal R1 based on the reflected electromagnetic wave EM2 detected by the radar device 102 from the radar device 102. In step S803, the estimation unit 1052 inputs the phase of the received signal R1 to the learning model 1041. In step S804, the estimation unit 1052 acquires object information MD1 corresponding to the signal intensity of the received signal R1 from the learning model 1041 . In step S605, the signal output unit 1053 outputs an output signal based on the object information TI1 to the observation device 200.

[0058] As described so far, when inferring object information using the learning model 1041, the learning model 1041 is trained to output object information by using the frequency spectrum, intensity, or phase as a feature amount of the received signal, and inference using the learning model 1041 is performed . At this time, it is possible to combine the frequency spectrum, intensity, or phase as the feature amount of the received signal.

[0059] With reference to FIGS. 9 and 10, another learning method of the learning model 1041 according to the present embodiment will be described.

[0060] FIG. 9 is a diagram schematically illustrating the learning and inference of the learning model 1041. The learning model ​​​​​​1041 is trained using the training data LD2 as the teacher data. The training data LD2 includes , among the received signals R0 (first synthetic aperture radar received signals) obtained by irradiating an electromagnetic wave to a certain area (learning area), a received signal R0a whose feature amount satisfies a predetermined condition, and an object information TI0 (first object information) corresponding to the feature amount of the received signal R 0a. A set of the received signal R0a and the object information TI0 corresponding to the feature amount of the received signal R0a is included.

[0061] The predetermined condition for the feature amount is a condition that enables extraction of a received signal including a feature amount based on an object in the learning area, and can be changed according to the purpose of observation or the like. For example, when extracting the received signal R0a from the received signal R0, a received signal whose signal intensity satisfies a predetermined condition can be extracted. For example, the predetermined condition for the signal intensity is that the signal intensity is greater than a predetermined threshold value. Also, for example, the predetermined condition for the frequency spectrum is that the frequency is within a predetermined frequency band. Also, for example, the predetermined condition for the phase is that the phase is within a predetermined range. For example, when extracting the received signal R0a from the received signal R0, a received signal whose signal intensity satisfies a predetermined condition can be extracted. For example, the predetermined condition for the signal intensity is that the signal intensity is greater than a predetermined threshold value. Also, for example, the predetermined condition for the frequency spectrum is that the frequency is within a predetermined frequency band. Also, for example, the predetermined condition for the phase is that the phase is within a predetermined range. For example, the predetermined condition for the signal intensity is that the signal intensity is greater than a predetermined threshold value. Also, for example, the predetermined condition for the frequency spectrum is that the frequency is within a predetermined frequency band. Also, for example, the predetermined condition for the phase is that the phase is within a predetermined range. For example, the predetermined condition for the frequency spectrum is that the frequency is within a predetermined frequency band. Also, for example, the predetermined condition for the phase is that the phase is within a predetermined range. For example, the predetermined condition for the phase is that the phase is within a predetermined range. For example, the predetermined condition for the phase is that the phase is within a predetermined range.

[0062] The learning model 1041 is trained with the feature amount of the received signal R0a as the input and the object information TI0 as the output.

[0063] With reference to FIG. 10, generation of the training data LD2 will be described. In the example of FIG. 10, the case of generating the training data LD2 using the observation results of a learning area D0 in which two objects, object O1 and object O2, exist is shown. When the learning area D0 is observed, the received signal R0 is obtained. When the received signal R0 is plotted as a graph or converted into a SAR image and imaged, an image IG0

[0064] When the learning area D0 is observed, the received signal R0 is obtained. When the received signal R0 is plotted as a graph or converted into a SAR image and imaged, an image IG0 is obtained. ​​is obtained. The image IG0 can be associated with the signal strength of the received signal R0. Among the received signals R0, the received signals whose signal strength is greater than a predetermined threshold value are extracted from the received signal R0. For example, a received signal R0a including a received signal R0a1 corresponding to a part of the region in the image IG0 and a received signal R0a2 corresponding to another part of the region in the image IG0 is extracted. When the received signal R0a1 is imaged, an image IG2 is obtained, and when the received signal R0a2 is imaged, an image IG3 is obtained. The images IG2 and IG3 can be associated with the respective signal strengths of the received signals R0a1 and R0a2.

[0065] Feature amounts of each received signal of the received signal R0a including the received signal R0a1 and the received signal R0a2 are each associated with object information TI0 based on the image IG0. The object information includes, as an example, object information TI0a and object information TI0b. The object information TI0a is a probability distribution of the probability that an object exists in the detection region. In the object information TI0a, regions A4 and A5 are shown as regions where the probability that an object exists is high. The object information TI0b is the number of objects and the attributes of the objects in the detection region. In this example, two objects are detected. The learning data LD2 is generated as a set of the received signal R0a and the object information TI0.

[0066] The learning model 1041 is learned by using the learning data LD2 prepared as described above by a general machine learning method such as a method using a neural network.

[0067] With reference to FIGS. 11 and 12, the processing by the flying object 100 will be described.

[0068] In step S1101 of FIG. 11, the radar device 102 irradiates the detection area D1 with the electromagnetic wave EM1. As shown in FIG. 12, objects O3, O4, O5 exist in the detection area D1.

[0069] In step S1102, the signal acquisition unit 1051 acquires, from the radar device 102, a reception signal R1 based on the reflected electromagnetic wave EM2 detected by the radar device 102.

[0070] In step S1103, the estimation unit 1052 extracts a reception signal R1a that satisfies a condition of a predetermined signal strength from the reception signal R1. The feature amount of the extracted reception signal R1a corresponds to, for example, an image IG4 in which a part of the image IG1 based on the reception signal R1 is cut out, as shown in FIG. 12.

[0071] In step S1104, the estimation unit 1052 inputs the reception signal R1a into the learning model 104 1.

[0072] In step S1105, the estimation unit 1052 acquires object information MD1 based on the feature amount of the reception signal R1a from the learning model 1041. The object information includes, as an example, object information TI1a and object information TI1b shown in FIG. 12. In the object information TI1a, regions A1, A2, A3 are shown as regions where the probability of the existence of an object is high, similar to the case of FIG. 5. The object information TI1b includes the number of objects and the attributes of the objects in the detection area. In this example, three objects are detected.

[0073] In step S1106, the signal output unit 1053 outputs an output signal based on the object information TI1 to the observation device 200. ​​​​​​​​​​​

[0074] In this way, the feature of the entire received signal is used to estimate the object from the feature of the partial received signal. By making it possible to output information, the learning model 1041 can derive the data of the input data used for inference. Furthermore, the amount of data can be reduced. This improves the real-time observation. This makes it possible to reduce the computational load of the target object information using a lighter processing device. It will be possible to obtain information.

[0075] FIG. 13 is a block diagram showing an observation system 10A according to another embodiment. As shown in FIG. 1, the signal processing device 103 is provided so as to be included in the observation device 200A. The control unit 1301 of the flying object 100A controls the reception information acquired by the radar device 102. The received signal is transmitted to the observation device 200A. It is also possible to perform the above-mentioned processing.

[0076] The above-described embodiment is intended to facilitate understanding of the present invention and does not limit the present invention. The elements and conditions of the embodiments are merely examples. The above is not limited to the above, and can be changed as appropriate. They may be partially substituted or combined. [Explanation of symbols]

[0077] 10...observation system, 100...aircraft vehicle, 101...communication antenna, 102...radar device, 103: signal processing device, 104: memory unit, 1041: learning model, 105: control unit, 10 51… Signal acquisition unit, 1052… Estimation unit, 1053… Signal output unit, 200… Observation device, 20 1… Communication unit, 202… Signal processing unit

Claims

1. A learning model that is trained using teacher data where the input is a feature amount of a first synthetic aperture radar reception signal based on a reflected electromagnetic wave obtained by reflecting an electromagnetic wave irradiated on a learning area, and the output is first object information of an object in the learning area, and that functions a computer to output second object information of an object in the detection area for an input of a feature amount of a second synthetic aperture radar reception signal based on a reflected electromagnetic wave obtained by reflecting an electromagnetic wave irradiated on the detection area.

2. The learning model according to Claim 1, wherein the feature amount of the first synthetic aperture radar reception signal includes a frequency spectrum of the first synthetic aperture radar reception signal, and the feature amount of the second synthetic aperture radar reception signal includes a frequency spectrum of the second synthetic aperture radar reception signal.

3. The learning model according to Claim 1, wherein the feature amount of the first synthetic aperture radar reception signal includes a signal intensity of the first synthetic aperture radar reception signal, and the feature amount of the second synthetic aperture radar reception signal includes a signal intensity of the second synthetic aperture radar reception signal.

4. The learning model according to Claim 1, wherein the feature amount of the first synthetic aperture radar reception signal includes a phase of the first synthetic aperture radar reception signal, and the feature amount of the second synthetic aperture radar reception signal includes a phase of the second synthetic aperture radar reception signal.

5. The learning model according to Claim 1, wherein the first synthetic aperture radar reception signal is a signal corresponding to a partial area in the learning area that satisfies a predetermined condition for the feature amount, and the feature amount of the first synthetic aperture radar reception signal includes a feature amount based on an object in the learning area.

6. The learning model according to Claim 5, wherein the second synthetic aperture radar reception signal is a signal corresponding to a partial area in the detection area that satisfies the predetermined condition for the feature amount, and the feature amount of the second synthetic aperture radar reception signal includes a feature amount based on an object in the detection area.

7. The learning model according to Claim 1, wherein the first object information includes information indicating an attribute of an object in the learning area, and the second object information includes information indicating an attribute of an object in the detection area.

8. A storage unit storing the learning model according to any one of Claims 1 to 7, and a signal acquisition unit that acquires the second synthetic aperture radar reception signal. ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ An estimation unit that inputs the second synthetic aperture radar reception signal to the learning model and infers the second object information A signal processing apparatus including the same. **Claim 9** A storage unit storing the learning model according to any one of Claims 1 to 7, A signal acquisition unit that acquires the second synthetic aperture radar reception signal, An estimation unit that inputs the second synthetic aperture radar reception signal to the learning model and infers the second object information And a signal output unit that outputs an output signal based on the second object information to the outside. An aircraft **Claim 10** 。 A computer Acquiring a first synthetic aperture radar reception signal based on a reflected electromagnetic wave obtained by reflecting an electromagnetic wave irradiated on a learning area; Extracting a learning signal corresponding to a region where an object exists in the learning region from the first synthetic aperture radar reception signal; Obtaining object information of the object; A set of a feature amount of the learning signal and the object information is input to a feature amount of a second synthetic aperture radar reception signal based on a reflected electromagnetic wave obtained by reflecting an electromagnetic wave irradiated on a detection region, and the second Storing, as teacher data, for use in learning a learning model that causes a computer to function so as to output second object information of an object in the detection region according to the feature amount of the synthetic aperture radar reception signal And a teacher data generation method including the above. ​ ​ ​ ​ ​ ​

Images