Learning device, target detecting device, learning method, and target detecting method

Abstract

This learning device is provided with: a delay profile generating unit (21) that generates a plurality of first delay profiles from a reception signal in a known environment, and assigns a label indicating a target state to the first delay profiles; an average processing unit (22) that obtains a second delay profile from the first delay profiles to which a label indicating a target absence state has been assigned; a subtraction processing unit (23) that subtracts the second delay profile from each first delay profile to obtain a plurality of third delay profiles; a first machine learning unit (41) that performs machine learning using the plurality of third delay profiles as teacher data to generate a first discriminator that discriminates between the presence or absence of a target; and a second machine learning unit (42) that performs machine learning using, as teacher data, the first delay profiles for which a label indicating a target presence state has been set, to generate a second discriminator that classifies the targets.

Description

Learning device, target detection device, learning method, and target detection method

[0001] The present disclosure relates to target detection technology.

[0002] Conventional wireless sensing technologies obtain RF signals, such as channel state information (CSI) or delay profiles, that include reflected signals from targets and extract information about moving objects through signal processing. One example of signal processing is a method of averaging multiple signals to remove reflected signals from stationary objects and extract time-varying components, i.e., moving object information (see, for example, Patent Document 1).

[0003] Japanese Patent Application Laid-Open No. 2020-008548

[0004] Conventional wireless sensing technology had the problem that when the unwanted scattered waves from structures surrounding the target object are larger than the reflected waves from the target object, the unwanted scattered waves become dominant, reducing the accuracy of target detection.

[0005] The present disclosure has been made to solve such problems, and aims to provide a target detection technology that can remove unnecessary scattered waves and improve the accuracy of target detection.

[0006] One aspect of a learning device according to an embodiment of the present disclosure includes a signal transmitting / receiving unit that transmits and receives wireless signals in an environment where it is known that a target does not exist or where the position of a target is known; a delay profile generating unit that performs AD conversion and time sampling on analog signals received by the signal transmitting / receiving unit to generate a plurality of first delay profiles, and that assigns to each first delay profile a label indicating which of a plurality of target states including a target absent state or a plurality of mutually different target present states the first delay profile is in; and a delay profile generating unit that assigns to each first delay profile a label indicating the target absent state. the first delay profile is set to a label indicating a target presence state, and the second delay profile is set to a label corresponding to the first delay profile; the subtraction processing unit is configured to subtract the second delay profile from each of the plurality of first delay profiles to obtain a plurality of third delay profiles; the first machine learning unit is configured to learn using the plurality of third delay profiles and corresponding labels to generate a first classifier that identifies the presence or absence of a target; and the second machine learning unit is configured to learn using the first delay profile, to which a label indicating a target presence state is set, and the corresponding label to generate a second classifier that classifies the target.

[0007] According to the learning device of the embodiment of the present disclosure, machine learning is performed using a third delay profile obtained by subtracting the second delay profile, which represents a state in which a target is not present, from the first delay profile, thereby enabling machine learning to be performed with unnecessary scattered waves removed, thereby improving the accuracy of target detection.

[0008] 7A is a functional configuration diagram of a target identification device according to embodiment 1. FIG. 7B is a flowchart illustrating the overall operation of the target identification device. FIG. 7C is a flowchart illustrating the second half of the operation of the target identification device according to embodiment 1. FIG. 7D is a diagram illustrating a calculation example of a third delay profile. FIG. 7A is a diagram illustrating a first delay profile to which a label "no target" is set. FIG. 7B is a diagram illustrating a first delay profile to which a label "no target" is set. FIG. 7C is a diagram illustrating a first delay profile to which a label "no target" is set. FIG. 7D is a diagram illustrating a calculation example of a third delay profile. FIG. 7B is a diagram illustrating a first delay profile to which a label "no target" is set. FIG. 7C is a diagram illustrating a second delay profile to which a label "no target" is set. FIG. 7D is a diagram illustrating a calculation example of a third delay profile. FIG. 7C is a diagram illustrating a functional configuration diagram of a target identification device according to embodiment 2. FIG. 7D is a diagram illustrating a functional configuration diagram of a target identification device according to embodiment 2. FIG. 7D is a flowchart illustrating the operation of the target identification device according to embodiment 2. FIG. 12A is a diagram showing an example of a delay profile generated when a time gate is applied to a received signal. FIG. 12B is a diagram showing a delay profile obtained by applying a time gate. FIG. 12B is a diagram showing a functional configuration of a signal processing unit of a target identification device according to embodiment 3. FIG. 14A is a diagram showing how a power threshold is set in a delay profile according to embodiment 1. FIG. 14B is a diagram showing an example of a delay profile generated by setting a power threshold. FIG. 14B is a diagram showing a functional configuration of a label setting unit of a target identification device according to embodiment 4. FIG. 12C is a diagram showing an example of the hardware configuration of a target identification device.

[0009] Various embodiments of the present disclosure will be described in detail below with reference to the drawings. In the drawings, identical or similar parts are designated by identical or similar reference numerals, and redundant explanations of such parts will be omitted. In addition, in this disclosure, the term "or" is used to mean an inclusive logical OR unless otherwise specified.

[0010] Embodiment 1. <Configuration> A target identification device according to embodiment 1 of the present disclosure will be described with reference to Fig. 1. Fig. 1 is a functional configuration diagram of the target identification device according to embodiment 1. As shown in Fig. 1, the target identification device includes a signal transmitting / receiving unit 10, a signal processing unit 20, a label setting unit 30, a machine learning unit 40, a database device 50, and a target identification result display control unit 60. The overall operation of the target identification device is controlled by a control unit (not shown) included in the target identification device.

[0011] The signal processing unit 20 includes a delay profile generating unit 21, an averaging unit 22, and a subtraction unit 23.

[0012] The machine learning unit 40 includes a first machine learning unit 41 and a second machine learning unit 42 .

[0013] The database device 50 includes a first database device 51 and a second database device 52 .

[0014] (Signal Transmitter / Receiver) The signal transmitter / receiver 10 includes an antenna and a transceiver (not shown) and transmits and receives radio signals. The signal transmitter / receiver 10 transmits and receives radio signals in an environment where there are no targets or in an environment where any target is located in any state (target presence state A, target presence state B, ...). The signal transmitter / receiver 10 supplies the received analog signal to the signal processor 20.

[0015] (Delay Profile Generator) The delay profile generator 21 performs AD conversion on the analog signal received from the signal transmitter / receiver 10 to a digital signal, performs time sampling, and generates a delay profile. The "delay profile" here includes, for example, discrete data with the horizontal axis representing delay time and the vertical axis representing power. The horizontal axis may represent propagation distance, or the vertical axis may represent amplitude or complex power.

[0016] The delay profiles generated by the delay profile generator 21 include delay profiles with known target states and delay profiles with unknown target states. In the present disclosure, a delay profile with a known target state is referred to as a "first delay profile," and a delay profile with an unknown target state is referred to as a "fourth delay profile." Note that in the present disclosure, the "target state" includes, for example, a case where there is no target, one or more stationary targets, one or more moving targets, or a case where stationary targets and moving targets are mixed.

[0017] In one aspect, the delay profile generating unit 21 supplies to the averaging processing unit 22 the delay profile from the first delay profile to which the label "no target" indicating the absence of a target has been set by the label setting unit 30.

[0018] In one aspect, the delay profile generator 21 supplies the first delay profile to the subtraction processor 23 .

[0019] In one aspect, the delay profile generator 21 supplies the fourth delay profile to the subtraction processor 23 .

[0020] In one aspect, the delay profile generator 21 supplies the fourth delay profile to the database device 50 .

[0021] (Averaging Processing Unit) The averaging processing unit 22 calculates the average value of multiple first delay profiles labeled "no target" and received from the delay profile generating unit 21, and obtains an average delay profile consisting of the average values. In the present disclosure, the average delay profile may be referred to as a "second delay profile." The averaging processing unit 22 supplies the obtained average delay profile to the subtraction processing unit 23.

[0022] (Subtraction Processing Unit) The subtraction processing unit 23 subtracts the average delay profile obtained from the averaging processing unit 22 from the delay profile obtained from the delay profile generating unit 21 to obtain a differential delay profile consisting of the differences at each time.

[0023] In one aspect, the subtraction processing unit 23 subtracts the average delay profile (second delay profile) acquired from the averaging processing unit 22 from the first delay profile labeled "no target" acquired from the delay profile generation unit 21 to acquire a differential delay profile. In the present disclosure, the differential delay profile acquired based on the first delay profile and the second delay profile may be referred to as a "third delay profile." The subtraction processing unit 23 supplies the acquired third delay profile to the machine learning unit 40.

[0024] In one aspect, the subtraction processing unit 23 subtracts the average delay profile (second delay profile) acquired from the averaging processing unit 22 from the fourth delay profile acquired from the delay profile generation unit 21 to acquire a differential delay profile. In the present disclosure, the differential delay profile acquired based on the fourth delay profile and the second delay profile may be referred to as a "fifth delay profile." The subtraction processing unit 23 supplies the acquired fifth delay profile to the database device 50.

[0025] (Label Setting Unit) The label setting unit 30 sets a label (target absent, target present state A, target present state B, ...) for each delay profile for which the presence or absence of a target or the target state is known, among the first delay profiles generated by the delay profile generating unit 21. The label setting unit 30 accepts user input via an input device (not shown), and sets a label to the first delay profile in accordance with the accepted user input.

[0026] (First Machine Learning Unit) The first machine learning unit 41 is a machine learning unit that performs machine learning using a plurality of third delay profiles and labels set to the plurality of third delay profiles as training data to generate a first classifier that identifies the presence or absence of a target. Algorithms used for machine learning include, for example, a convolutional neural network (CNN). Other machine learning algorithms may also be used.

[0027] (Second Machine Learning Unit) The second machine learning unit 42 is a machine learning unit that performs machine learning using, as training data, a first delay profile, of a plurality of first delay profiles, to which a label indicating a target presence state is set, and the corresponding label indicating the target presence state, to generate a second classifier that classifies targets. Algorithms used for machine learning include, for example, CNN. Other machine learning algorithms may also be used.

[0028] (First Database Device) The first database device 51 is a database device that stores the first classifier generated by the first machine learning unit 41 .

[0029] (Second Database Device) The second database device 52 is a database device that stores the second classifier generated by the second machine learning unit 42 .

[0030] (Classification Identification Result Display Control Unit) The classification identification result display control unit 60 is a functional unit that performs display control so that the judgment results from the first database device 51 and the judgment results from the second database device 52 are displayed on a display device (not shown).

[0031] <Hardware Configuration> Here, with reference to Fig. 16 , an example of the hardware configuration of the signal processing unit 20, the label setting unit 30, the machine learning unit 40, the database device 50, and the class identification result display control unit 60 will be described. The functions realized by these functional units or the database device 50 may be realized by a processor 101 that executes programs stored in a memory 102 as shown in Fig. 16 . The functions realized by these functional units or the database device 50 are realized by software, firmware, or a combination of software and firmware. The software and firmware are written as programs and stored in the memory 102. The processor 101 realizes various functions by reading and executing the programs stored in the memory 102. Here, examples of memory 102 include non-volatile or volatile semiconductor memory such as RAM (random access memory), ROM (read-only memory), flash memory, EPROM (erasable programmable read-only memory), EEPROM (electrically erasable programmable read-only memory), etc., as well as magnetic disks, flexible disks, optical disks, compact disks, minidisks, and DVDs.

[0032] <Operation> Next, the overall operation of the target identification device will be described with reference to Fig. 2. The processing in Fig. 2 is performed by a control unit (not shown) provided in the target identification device.

[0033] (Step S1) In step S1, it is confirmed whether a valid classifier is stored in the database device 50. Here, the "valid classifier" includes a state in which a classifier that classifies target types in the current propagation environment has been generated. If the propagation environment changes significantly and the class classification accuracy of the existing classifier decreases, it is necessary to generate a "valid classifier" again.

[0034] (Step S2) If a valid classifier is not stored in the database device 50 in step S1, a "classifier generation phase" is executed in step S2. The "classifier generation phase" here includes generating a classifier that classifies the type of target in the current propagation environment. Specific operations in this phase will be described later in <Classifier generation phase>.

[0035] (Step S3) If a valid classifier is stored in the database device 50 in step S1, a "target class identification phase" is executed in step S3. The "target class identification phase" here includes estimating a target from an unknown signal using a valid classifier in the current propagation environment. Specific operations in this phase will be described later in <Target class identification phase>.

[0036] <Classifier Generation Phase> The target identification device in the classifier generation phase will be described mainly with reference to Figs. 3 to 5. The target identification device in the classifier generation phase has an aspect as a learning device that performs machine learning. Fig. 3 is a functional configuration diagram of the target identification device in the classifier generation phase. Fig. 4 is a flowchart showing the first half of the operation of the target identification device in the classifier generation phase. Fig. 5 is a flowchart showing the second half of the operation of the target identification device in the classifier generation phase.

[0037] 3 , the target class identification device in the classifier generation phase includes a signal transmitting / receiving unit 10, a signal processing unit 20, a label setting unit 30, a machine learning unit 40, and a database device 50. The signal processing unit 20 includes a delay profile generation unit 21, an averaging processing unit 22, and a subtraction processing unit 23. The machine learning unit 40 includes a first machine learning unit 41 and a second machine learning unit 42. The database device 50 includes a first database device 51 and a second database device 52.

[0038] <Operation> (Step S11) In step S11 of FIG. 4, the signal transmitting / receiving unit 10 transmits and receives wireless signals and receives signals in which the target state is known in an environment in which it is known that a target does not exist or in an environment in which the position of a target is known.

[0039] (Step S12) Next, in step S12, the delay profile generator 21 acquires the signal received by the signal transmitter / receiver 10 and generates a first delay profile of the acquired signal. An example of the first delay profile is shown in Fig. 6. The horizontal axis may represent the propagation distance, and the vertical axis may represent the amplitude or complex power.

[0040] (Step S13) Next, in step S13, the label setting unit 30 sets a label of a target state for the first delay profile. Here, the "label" includes, for example, data for associating and referencing which target state the delay profile corresponds to.

[0041] (Step S14) Next, in step S14, the averaging processing unit 22 acquires from the delay profile generating unit 21 a plurality of delay profiles that have labels indicating the absence of a target, among the first delay profiles, and stacks the acquired delay profiles. When a certain number of delay profiles have been stacked, the averaging processing unit 22 calculates the average value of the stacked delay profiles to acquire a second delay profile. Examples of calculation of the second delay profile are shown in Figures 7A to 7D.

[0042] (Step S15) Next, in step S15, the subtraction processing unit 23 acquires a third delay profile from the first delay profile and the second delay profile. Specifically, the subtraction processing unit 23 acquires the first delay profile from the delay profile generation unit 21 and the second delay profile from the averaging processing unit 22. The first delay profile acquired by the subtraction processing unit 23 from the delay profile generation unit 21 includes a delay profile labeled "target present" or a delay profile labeled "target absent." The subtraction processing unit 23 then subtracts the second delay profile from the first delay profile to calculate the difference, thereby acquiring a third delay profile. The "subtraction" here may be, for example, power subtraction, amplitude subtraction, or complex power subtraction. An example of calculating the third delay profile is shown in FIG. 8.

[0043] (Step S21) In step S21 of FIG. 5, the first machine learning unit 41 receives an arbitrary number of third delay profiles from the subtraction processing unit 23 and generates a first classifier with a label of “target present” or “target absent”.

[0044] (Step S22) In step S22, the first machine learning unit 41 transfers the first classifier generated in step S21 to the first database device 51, and the first database device 51 stores the first classifier.

[0045] (Step S23) Next, in step S23, the second machine learning unit 42 receives an arbitrary number of first delay profiles from the delay profile generation unit 21, and generates a second classifier using a label indicating a target presence state.

[0046] (Step S24) Next, in step S24, the second machine learning unit 42 transfers the second classifier generated in step S23 to the second database device 52, and the second database device 52 stores the second classifier.

[0047] <Target Identification Phase> The target identification device in the target identification phase will be described with reference to Fig. 9 and Fig. 10. The target identification device in the target identification phase has an aspect as a target detection device that identifies targets using the results of machine learning. Fig. 9 is a functional configuration diagram of the target identification device in the target identification phase. Fig. 10 is a flowchart showing the operation of the target identification device in the target identification phase.

[0048] 9 , the target class identification device in the target class identification phase includes a signal transmitting / receiving unit 10, a signal processing unit 20, a database device 50, and a class identification result display control unit 60. The signal processing unit 20 includes a delay profile generating unit 21, an averaging processing unit 22, and a subtraction processing unit 23. The database device 50 includes a first database device 51 and a second database device 52.

[0049] <Operation> (Step S31) In step S31 of FIG. 10, the signal transmitting / receiving unit 10 transmits and receives a signal indicating an unknown target state (step S31).

[0050] (Step S32) Next, in step S32, the delay profile generating unit 21 acquires the signal received by the signal transmitting / receiving unit 10, and generates a fourth delay profile of the acquired signal (step S32).

[0051] (Step S33) Next, in step S33, the subtraction processing unit 23 obtains a fifth delay profile from the fourth delay profile and the second delay profile. Specifically, the subtraction processing unit 23 obtains the fourth delay profile from the delay profile generation unit 21 and the second delay profile from the averaging processing unit 22. Thereafter, the subtraction processing unit 23 obtains the fifth delay profile by subtracting the second delay profile from the fourth delay profile to calculate the difference.

[0052] (Step S34; Step S35) Next, in step S34, the subtraction processing unit 23 inputs the fifth delay profile to the first classifier stored in the first database device 51, and the first database device 51 determines the presence or absence of a target based on the determination result of the first classifier. The processes of steps S34 and S35 may be performed by a control unit (not shown).

[0053] (Step S36) If it is determined that a target is present (YES in step S35), in step S36, the delay profile generator 21 inputs the fourth delay profile to the second classifier stored in the second database device 52, and the second database device 52 classifies the state of the target (step S36). The processing of step S36 may be performed by a control unit (not shown).

[0054] (Step S37) In step S37, the class identification result display control unit 60 controls the display to display the determination result in step S34 and the categorization result in step S36 on a display device (not shown). Similarly, even if it is determined in step S35 that there is no target, the class identification result display control unit 60 controls the display to display the determination result in step S35 on a display device (not shown).

[0055] <Effect> As described above, the third delay profile is generated by subtracting the second delay profile from the first delay profile, so that unnecessary scattered waves can be removed, improving the accuracy of determining whether or not a target is present (first effect).

[0056] Furthermore, in the first embodiment, a two-stage classification method is adopted, in which the first classifier determines whether or not a target is present, and the second classifier classifies the target state. In the latter classification of the target state, the fourth delay profile is used as a reference before components due to unnecessary scattered waves are removed, which solves the problem that the reflected waves from the target are also removed when unnecessary scattered waves are removed, making it difficult to determine the target state (second effect).

[0057] Therefore, by combining the first and second effects, the accuracy of classifying target states is improved overall.

[0058] Embodiment 2 In the first embodiment, all signals received by the signal transmitting / receiving unit 10 are transmitted to the delay profile generating unit 21, and a delay profile is generated. In the second embodiment, as shown in FIG. 11 , the signal processing unit 20A includes a time gate unit 24 in front of the delay profile generating unit 21. The time gate unit 24 extracts a time gate signal only within an arbitrary time range from the received signal and transmits the extracted time gate signal to the delay profile generating unit 21. FIG. 12 shows an example of a delay profile generated when the time gate unit is applied to the received signal. By providing a time gate, as shown in FIGS. 12A and 12B, only a portion of the delay profile obtained by the first embodiment is extracted. The configuration other than the signal processing unit 20A is the same as that of the first embodiment, and therefore a description thereof will be omitted.

[0059] By providing the time gate unit 24, it is possible to prevent the target class identification device from using signals that are not related to the reflected waves from the target and that are close or far away from the target class identification device in the first place. This makes it possible to remove unnecessary scattered waves to a certain extent, and is expected to improve the class identification accuracy.

[0060] Embodiment 3 In embodiment 1, all signals received by the signal transmitting / receiving unit 10 are transmitted to the delay profile generating unit 21, which generates a delay profile. In embodiment 3, as shown in FIG. 13, the delay profile generating unit 21B provided in the signal processing unit 20B includes a power threshold unit 25. The power threshold unit 25 is a functional unit that sets a threshold for power. The delay profile generating unit 21B generates a delay profile by inserting a zero value into sample values ​​that are less than the threshold set by the power threshold unit 25. An example of a delay profile generated by setting a power threshold is shown in FIG. 14. The configuration other than the signal processing unit 20B is the same as in embodiment 1, so a description thereof will be omitted.

[0061] By providing the power threshold unit 25, it is possible to avoid using profiles with small values ​​that are not related to the reflected waves from the target, which improves the accuracy of generating a classifier and is expected to improve the accuracy of class classification.

[0062] Embodiment 4 In Embodiment 1, a signal received by the signal transmitting / receiving unit 10 is transmitted to the delay profile generating unit 21 to generate a delay profile. In Embodiment 4, as shown in FIG. 15 , the target identification device includes a label setting unit 30C including a radio wave propagation analysis unit 31, instead of including the signal transmitting / receiving unit 10. The radio wave propagation analysis unit 31 executes generation of a simulated signal in the classifier generation phase. The environment set by the radio wave propagation analysis unit 31 needs to simulate the current propagation environment (actual environment). The other configurations are the same as those in Embodiment 1 (classifier generation phase), and therefore description thereof will be omitted.

[0063] By providing the radio wave propagation analysis unit 31, it becomes possible to generate signals to be input to the delay profile generation unit 21 in the discriminator generation phase offline, eliminating the need to transmit and receive actual signals in this phase.

[0064] <Supplementary Notes> Some aspects of the various embodiments described above are summarized below.

[0065] (Supplementary Note 1) The learning device according to Supplementary Note 1 includes a signal transmitting / receiving unit (10) that transmits and receives a radio signal in an environment where it is known that a target does not exist or where the position of a target is known, a delay profile generating unit (21) that performs AD conversion and time sampling on an analog signal received by the signal transmitting / receiving unit to generate a plurality of first delay profiles, and assigns to each first delay profile a label indicating which of a plurality of target states including a target absent state or a plurality of mutually different target present states the first delay profile is in, and a delay profile generating unit (21) that assigns to each first delay profile a label indicating the target absent state. a subtraction processing unit (23) that subtracts the second delay profile from each of the plurality of first delay profiles to obtain a plurality of third delay profiles; a first machine learning unit (41) that learns using the plurality of third delay profiles and corresponding labels to generate a first classifier that identifies the presence or absence of a target; and a second machine learning unit (42) that learns using the first delay profile, to which a label indicating a target presence state is set, and the corresponding label to generate a second classifier that classifies the target.

[0066] (Supplementary Note 2) The learning device according to Supplementary Note 2 is the learning device described in Supplementary Note 1, wherein the delay profile generation unit has a time gate of a predetermined time range, acquires a time gate signal within the time gate from an analog signal received by the signal transmission / reception unit, converts the acquired time gate signal into a digital signal, and obtains the plurality of first delay profiles from the converted digital signal.

[0067] (Supplementary Note 3) The learning device according to Supplementary Note 3 is the learning device described in either Supplementary Note 1 or 2, wherein the delay profile generation unit has a predetermined power threshold, acquires a time gate signal within the time gate from an analog signal received by the signal transmission / reception unit, converts the acquired time gate signal into a digital signal, replaces sample values ​​in the converted digital signal that are less than the power threshold with zero values, and obtains the plurality of first delay profiles from the replaced digital signal.

[0068] (Supplementary Note 4) The learning device according to Supplementary Note 4 is the learning device described in any one of Supplements 1 to 3, further comprising a label setting unit (30) that sets a label indicating the target state, and the delay profile generation unit assigns the label set by the label setting unit to each first delay profile.

[0069] (Supplementary Note 5) The learning device according to Supplementary Note 5 is the learning device described in Supplementary Note 4, wherein the label setting unit has a radio wave propagation analysis unit (31) that holds the results of radio wave propagation analysis that simulates a real environment, and the delay profile generation unit obtains the plurality of first delay profiles using the results of radio wave propagation analysis held by the radio wave propagation analysis unit instead of a digital signal obtained from an analog signal received by the signal transmission / reception unit.

[0070] (Appendix 6) The target detection device according to Supplementary Note 6 includes a signal transmitting / receiving unit (10) that transmits and receives a radio signal in an environment where the presence or absence of a target is unknown, a delay profile generating unit (21) that performs AD conversion and time sampling on an analog signal received by the signal transmitting / receiving unit to generate a fourth delay profile, a subtraction processing unit (23) that obtains a fifth delay profile by subtracting a second delay profile that is calculated as an average of a plurality of first delay profiles that are acquired in an environment where the presence or absence of a target is known and that are labeled to indicate a target absence state from the fourth delay profile, a first database device (51) that holds a first classifier that identifies the presence or absence of a target and receives the fifth delay profile as an input to determine the presence or absence of a target, a second database device (52) that holds a second classifier that classifies targets and receives the fourth delay profile as an input to classify the targets, and a class identification result display control unit (60) that controls display of a determination result by the first database device and a determination result by the second database device on a display device.

[0071] (Supplementary Note 7) The learning method according to Supplementary Note 7 is a learning method performed by a learning device including a signal transmitting / receiving unit (10), a delay profile generating unit (21), an averaging processing unit (22), a subtraction processing unit (23), a first machine learning unit (41), and a second machine learning unit (42), and includes a step (S11) in which the signal transmitting / receiving unit transmits and receives a radio signal in an environment in which it is known that a target does not exist or in an environment in which a position where a target is located is known, and a step (S12) in which the delay profile generating unit performs AD conversion and time sampling on an analog signal received by the signal transmitting / receiving unit to generate a plurality of first delay profiles, and assigns to each first delay profile a label indicating which of a plurality of target states including a target absence state or a plurality of mutually different target presence states the first delay profile represents. the first machine learning unit performs learning using the third delay profiles and the corresponding labels to generate a first classifier that identifies the presence or absence of a target (S21); and the second machine learning unit performs learning using the first delay profiles to which labels that indicate the target presence state are set and the corresponding labels to generate a second classifier that classifies a target (S23).

[0072] (Supplementary Note 8) A target detection method according to Supplementary Note 8 is a target detection method performed by a target detection device including a signal transmitting / receiving unit (10), a delay profile generating unit (21), a subtraction processing unit (23), a first database device (51), a second database device (52), and a class identification result display control unit (60), the method including: a step (S31) in which the signal transmitting / receiving unit transmits and receives a radio signal in an environment in which the presence or absence of a target is unknown; a step (S32) in which the delay profile generating unit performs AD conversion and time sampling on the analog signal received by the signal transmitting / receiving unit to generate a fourth delay profile; and a step (S33) in which the subtraction processing unit generates a fourth delay profile from a plurality of first delay profiles acquired in an environment in which the presence of a target is known and labeled with a label indicating the absence of a target. The method includes a step (S33) of subtracting a second delay profile calculated as an average of delay profiles from the fourth delay profile to obtain a fifth delay profile; a step (S34) of the first database device holding a first classifier that identifies the presence or absence of a target, receiving the fifth delay profile as an input, and determining the presence or absence of a target; a step (S36) of the second database device holding a second classifier that classifies targets, receiving the fourth delay profile as an input, and classifying the targets; and a step (S37) of the class classification result display control unit controlling the display of the determination results by the first database device and the determination results by the second database device on a display device.

[0073] It is possible to combine the embodiments, and to modify or omit each embodiment as appropriate.

[0074] The learning device of the present disclosure can be used as a device for generating a classifier for identifying targets by wireless sensing.

[0075] 10 Signal transmitting / receiving unit, 20 Signal processing unit, 20A Signal processing unit, 20B Signal processing unit, 21 Delay profile generation unit, 21B Delay profile generation unit, 22 Average processing unit, 23 Subtraction processing unit, 24 Time gate unit, 25 Power threshold unit, 30 Label setting unit, 30C Label setting unit, 31 Radio wave propagation analysis unit, 40 Machine learning unit, 41 First machine learning unit, 42 Second machine learning unit, 50 Database device, 51 First database device, 52 Second database device, 60 Classification result display control unit, 101 Processor, 102 Memory.

Claims

1. A learning device comprising: a signal transmitting / receiving unit that transmits and receives wireless signals in an environment where it is known that a target is not present or an environment where the position of a target is known; a delay profile generating unit that performs AD conversion and time sampling on analog signals received by the signal transmitting / receiving unit to generate a plurality of first delay profiles, and assigns to each first delay profile a label indicating which of a plurality of target states including a target absent state or a plurality of different target present states the first delay profile represents; an averaging processing unit that obtains a second delay profile that is an average of the first delay profiles that have been assigned a label indicating the target absent state; a subtraction processing unit that obtains a plurality of third delay profiles by subtracting the second delay profile from each of the plurality of first delay profiles; a first machine learning unit that trains using the plurality of third delay profiles and their corresponding labels to generate a first classifier that distinguishes between the presence and absence of a target; and a second machine learning unit that trains using the first delay profiles to which a label indicating a target present state has been assigned and the corresponding label to generate a second classifier that classifies targets.

2. The learning device described in claim 1, wherein the delay profile generation unit has a time gate of a predetermined time range, acquires a time gate signal within the time gate from an analog signal received by the signal transmission / reception unit, converts the acquired time gate signal into a digital signal, and obtains the plurality of first delay profiles from the converted digital signal.

3. A learning device according to claim 1 or 2, wherein the delay profile generation unit has a predetermined power threshold, acquires a time gate signal within the time gate from the analog signal received by the signal transmission / reception unit, converts the acquired time gate signal into a digital signal, replaces sample values ​​in the converted digital signal that are less than the power threshold with zero values, and obtains the plurality of first delay profiles from the replaced digital signal.

4. A learning device according to any one of claims 1 to 3, further comprising a label setting unit that sets a label indicating the target state, and the delay profile generation unit assigns the label set by the label setting unit to each first delay profile.

5. A learning device as described in claim 4, wherein the label setting unit has a radio wave propagation analysis unit that holds the results of radio wave propagation analysis that simulates a real environment, and the delay profile generation unit obtains the plurality of first delay profiles using the results of radio wave propagation analysis held by the radio wave propagation analysis unit instead of a digital signal obtained from an analog signal received by the signal transmission / reception unit.

6. A target detection device comprising: a signal transmitting / receiving unit that transmits and receives wireless signals in an environment where the presence or absence of a target is unknown; a delay profile generating unit that performs AD conversion and time sampling on analog signals received by the signal transmitting / receiving unit to generate a fourth delay profile; a subtraction processing unit that obtains a fifth delay profile by subtracting from the fourth delay profile a second delay profile that is calculated as the average of multiple first delay profiles acquired in an environment where the absence of targets is known and that are labeled to indicate a target absence state; a first database device that holds a first classifier that identifies the presence or absence of a target and receives the fifth delay profile as input to determine the presence or absence of a target; a second database device that holds a second classifier that classifies targets and receives the fourth delay profile as input to classify the targets; and a class identification result display control unit that controls the display of the determination results by the first database device and the determination results by the second database device on a display device.

7. A learning method performed by a learning device including a signal transmitting / receiving unit, a delay profile generating unit, an averaging processing unit, a subtraction processing unit, a first machine learning unit, and a second machine learning unit, comprising: a step in which the signal transmitting / receiving unit transmits and receives a radio signal in an environment where it is known that a target does not exist or an environment where a position where a target is located is known; a step in which the delay profile generating unit performs AD conversion and time sampling on an analog signal received by the signal transmitting / receiving unit to generate a plurality of first delay profiles, and assigns to each first delay profile a label indicating which of a plurality of target states including a target absent state or a plurality of mutually different target present states the first delay profile represents; a step in which the averaging processing unit obtains a second delay profile that is an average of the first delay profiles assigned with the label indicating the target absent state; and a step in which the subtraction processing unit obtains a plurality of third delay profiles by subtracting the second delay profile from each of the plurality of first delay profiles. a step in which the first machine learning unit learns using the plurality of third delay profiles and corresponding labels to generate a first classifier that identifies the presence or absence of a target; and a step in which the second machine learning unit learns using the first delay profile, in which a label indicating a target presence state is set, and the corresponding label to generate a second classifier that classifies a target.

8. A target detection method performed by a target detection device comprising a signal transmitting / receiving unit, a delay profile generating unit, a subtraction processing unit, a first database device, a second database device, and a class identification result display control unit, comprising: a step in which the signal transmitting / receiving unit transmits and receives a radio signal in an environment where the presence or absence of a target is unknown; a step in which the delay profile generating unit performs AD conversion and time sampling on an analog signal received by the signal transmitting / receiving unit to generate a fourth delay profile; a step in which the subtraction processing unit subtracts a second delay profile, which is calculated as the average of a plurality of first delay profiles acquired in an environment where the absence of a target is known and which are labeled to indicate a target absence state, from the fourth delay profile to obtain a fifth delay profile; a step in which the first database device holds a first classifier that identifies the presence or absence of a target, receives the fifth delay profile as input, and determines the presence or absence of a target; and a step in which the second database device holds a second classifier that classifies targets, receives the fourth delay profile as input, and classifies the target. a step in which the class identification result display control unit controls a display device to display the determination result by the first database device and the determination result by the second database device.

Images