JP7466418B2 - ISAR image data learning device, target classification device and radar device - Google Patents

Description

This disclosure relates to an ISAR image data learning device that learns ISAR image data, and a target classification device and radar device that use the same.

Conventionally, inverse synthetic aperture radar (ISAR) has been used to identify targets using ISAR images created from radar echoes. For example, there is a method for simulating and matching a target captured by ISAR with a target shape model stored in a database in advance, and a method for extracting multiple feature points from both the ISAR image and the target shape model.

Conventionally, some target classification devices extract the overall length of a target and the size of a structure from an ISAR image and identify the target based on this in order to consistently identify targets from ISAR images with high accuracy (see, for example, Patent Document 1). In addition, some target classification devices generate pseudo ISAR images from a three-dimensional model of an identified target using the attitude change due to the rotational motion of the identified target as a parameter (see, for example, Patent Document 2).

Furthermore, in classifying ISAR images, there are methods that compensate for the effects of discrete distribution of observation parameters, generate highly compatible filters, and perform highly accurate type determination (see, for example, Patent Document 3). On the other hand, there are methods that create learning models by machine learning using AI (Artificial Intelligence) for use in classifying ISAR images (see, for example, Patent Document 4).

Note that there has been a conventional method of integrating ISAR images over time before performing extraction processing (see, for example, Patent Document 5). When ISAR images are integrated over time, multiple images are superimposed according to the integration time, which has the effect of improving apparent blur. Patent Document 5 discloses a method of identifying a target using a single ISAR image without integration, in order to avoid blurring the overall shape, rather than improving apparent blur caused by superimposing multiple images according to the integration time of the ISAR image.

Of course, to avoid the overall shape becoming blurred, an appropriate integration time (observation time) is set (see, for example, Patent Document 6). Normally, if observation is performed for a long period of time, weak signals can be accumulated due to the integration effect, but in the case of ISAR images, if observation is performed for a long period of time that exceeds the appropriate integration time (observation time), the movement of each reflection point of the target becomes complex, and the change in speed within the integration time (observation time) becomes large. This causes the Doppler spread to become large, causing the image to become blurred, so it is necessary to set an appropriate observation time. The appropriate observation time is determined depending on the relative motion between the ISAR and the target.

JP 2001-221857 A JP 2004-93166 A JP 2012-127920 A WO2020-3586 JP 2016-57164 A JP 2009-162611 A

However, conventional machine learning such as that disclosed in Patent Document 4 has a problem in that it does not consider a learning model that limits the range of learning results used for classification. The methods disclosed in Patent Documents 5 and 6 do not suggest using integral time in the learning model.

The present disclosure has been made to solve the problems described above, and aims to provide an ISAR image data learning device that generates learning results so that the range of learning results used for classification can be limited, and a target classification device and radar device that use the same.

The ISAR image data learning device of the present disclosure is an ISAR image data learning device that generates learning results for classifying targets from ISAR image data obtained by an inverse synthetic aperture radar installed at an observation point , and includes an ISAR image input unit to which the ISAR image data is input, an integral time input unit to which the integral time used to derive the ISAR image data is input, and a learning unit that learns to classify the targets from the ISAR image data corresponding to the integral time input to the integral time input unit.

The target classification device according to the present disclosure includes an ISAR image input unit to which ISAR image data is input, an integral time input unit to which an integral time used to derive the ISAR image data is input, a learning unit that classifies and learns targets from the ISAR image data corresponding to the integral time input to the integral time input unit , a new ISAR image input unit to which newly obtained new ISAR image data is input, a new integral time input unit to which the integral time used to derive the new ISAR image data is input, and a target classification unit that classifies the targets from the new ISAR image data based on the learning results learned by the learning unit.

The radar device according to the present disclosure includes an ISAR image input unit to which ISAR image data is input, an integral time input unit to which an integral time used to derive the ISAR image data is input, a learning unit that classifies and learns targets from the ISAR image data corresponding to the integral time input to the integral time input unit , a new ISAR image input unit to which newly obtained new ISAR image data is input , a new integral time input unit to which an integral time used to derive the new ISAR image data is input, a target classification unit that classifies the targets from the new ISAR image data based on the learning results learned by the learning unit , a tracking radar that tracks the targets, an inverse synthetic aperture radar that transmits a transmission beam to the target tracked by the tracking radar and receives radio waves reflected from the target, and an integration processing unit that integrates a video signal obtained from the radio waves reflected from the target to generate the new ISAR image data and send it to the new ISAR image input unit.

According to the present disclosure, it is possible to obtain an ISAR image data learning device that can obtain learning results that enable classification even for ISAR image data obtained with a specific integration time, and a target classification device and radar device that use the same.

FIG. 2 is a functional block diagram of the ISAR image data learning device according to the first embodiment. 1 is a flowchart for explaining the operation of an ISAR image data learning device (ISAR image data learning method) according to the first embodiment. FIG. 2 is a functional block diagram of the ISAR image data learning device according to the first embodiment (including an ISAR image generation unit 13). 1 is a functional block diagram of an ISAR image data learning device, a target classification device, and a radar device according to a first embodiment. 1 is a flowchart explaining the operation of the ISAR image data learning device and the target classification device (target classification method) of embodiment 1.

Embodiment 1.
Hereinafter, an ISAR image data learning device according to a first embodiment, and a target classification device and a radar device using the same (the target classification device according to the first embodiment, and the radar device according to the first embodiment) will be described with reference to Fig. 1 to Fig. 5. In the drawings, the same reference numerals indicate the same or corresponding parts, and detailed description thereof will be omitted. As mentioned above, ISAR is an abbreviation for Inverse Synthetic Aperture Radar, which means an inverse synthetic aperture radar.

The target 1 to be observed is a moving object such as a ship or an aircraft. The body of the target 1 is preferably a linear structure. In other words, it is preferable that the shape of the target 1 changes when the observation point relative to the target 1 is changed. If the structure is linear, it is easy to understand that the shape is different when observed from the side and when observed from the front or rear. The observation point itself may move, or the target 1 may move. In this application, learning is performed on the premise that there are multiple types of targets 1. Therefore, as the amount of learning progresses, even when new ISAR image data shooting described later is of an unknown target 1, it is possible to determine whether there is a similar type or whether there is nothing similar.

In Figures 1 and 3, when classifying corresponding targets 1 from ISAR image data obtained by an inverse synthetic aperture radar 12 installed at an observation point and determining the type corresponding to the target 1, the ISAR image data learning device 2 generates learning results so that the range of learning results used for classification can be limited from the integration time (observation time) during which the ISAR image data was derived (ISAR image data learning device according to embodiment 1). Since ISAR image data is constructed by superimposing multiple ISAR images according to the integration time (observation time), if the integration time used to obtain ISAR image data for the same target 1 changes, the ISAR image data will change.

In Figures 1 and 3, ISAR image input unit 3 receives ISAR image data for each target 1. The integral time input unit 4 receives integral times for each ISAR image data input to ISAR image input unit 3. The learning unit 5 learns by associating the ISAR image data corresponding to the integral time input to integral time input unit 4 with the type of target 1. Information on the type of target 1 may be added to the ISAR image data. The learning unit 5 may also generate learning results for each integral time.

Up to this point, the integral time and the observation time have been described as equivalent, but strictly speaking, they can be treated differently. Here, the integral time is a time delimited by an integral start time and an integral end time. Similarly, the observation time is a time delimited by an observation start time and an observation end time. For example, the integral time input unit 4 may be input with an integral time that is ISAR image data of the same target 1 and has an integral start time and an integral end time between the observation start time and the observation end time of the target 1. Of course, even if the integral time is the same, the time from the observation start time to the observation end time of the target 1 may be different. In the examples included here, even if the integral times are different, the integral time input unit 4 may input integral times with the same integral start time. Similarly, the integral time input unit 4 may input integral times with different integral start times.

Furthermore, the integral time input unit 4 may input an integral time in which the integration start time is the same and the integration end time is the time it takes for the period of the target 1 to make one round trip. Next, the integral time input unit 4 may input an integral time from the integration start time to the integration end time that corresponds to one round trip of the period of the same target 1's vibration. Of course, the ISAR image input unit 3 may input ISAR image data of the same target 1 with different integration times.

The type of target 1 may be information that includes at least one of the model name of target 1, aircraft dimensions, aircraft performance, and the country/organization to which the aircraft belongs. Furthermore, even if the type of target 1 is the same, if the integral time at which the ISAR image data was acquired is different, the ISAR image data will be different. In other words, the ISAR image data input to the ISAR image input unit 3 may have a different integral time input to the integral time input unit 4, even for the same target 1. The ISAR image data may not be data that has actually been acquired by ISAR 12, but may be simulated data. The integral time may not be data that has actually been acquired, but may be simulated data.

Figure 1 (A) shows an ISAR image data learning device 2 in which the learning unit 5 does not learn information about the distance from the observation point to the target 1. On the other hand, Figure 1 (B) shows an ISAR image data learning device 2 in which the learning unit 5 learns information about the distance from the observation point to the target 1. The distance information to be learned by the learning unit 5 may be input via the ISAR image input unit 3 or via the integral time input unit 4, as shown in Figure 1 (B). In other words, the ISAR image input unit 3 may receive ISAR image data associated with information about the distance from the observation point to the target 1. The integral time input unit 4 may receive an integral time associated with information about the distance from the observation point to the target 1. If there is information about the distance from the observation point to the target 1, the learning unit 5 can learn ISAR image data at a predetermined scale from the distance information.

Figure 3 illustrates a case where the ISAR image input unit 3 and the integral time input unit 4 respectively receive the ISAR image data and integral time output from the ISAR image generation unit 13. Figure 3(A) corresponds to Figure 1(A), and Figure 3(B) corresponds to Figure 1(B). That is, Figure 3(A) illustrates an ISAR image data learning device 2 in which the learning unit 5 does not learn information about the distance from the observation point to the target 1. Figure 3(B) illustrates an ISAR image data learning device 2 in which the learning unit 5 learns information about the distance from the observation point to the target 1. Note that the ISAR image data learning device 2 may include the ISAR image generation unit 13, or the ISAR image generation unit 13 may be provided externally.

A radar device 10 (tracking radar 11, ISAR 12) is installed at an observation point that observes the target 1. The observation point may be mobile. In other words, the radar device 10 may be mounted on a moving body such as a ship or aircraft. The scanning range is the range that can be scanned from the radar device 10 (tracking radar 11, ISAR 12) centered on the observation point. The radar device 10 may be capable of scanning 360 degrees in the azimuth direction. The observation point (radar device 10) is assumed to be a ship or aircraft that itself moves, but the observation point may also be a fixed point.

Next, the operation of the ISAR image data learning device according to the first embodiment (the ISAR image data learning method according to the first embodiment) will be described with reference to FIG. 2. In FIG. 2, step 1 is a processing step in which ISAR image data is input to the ISAR image input unit 3 for each target 1. Step 2 is a processing step in which the integral time for each ISAR image data input to the ISAR image input unit 3 is input to the integral time input unit 4. The order of processing of steps 1 and 2 does not matter. They may be performed simultaneously. Step 3 is a processing step in which the learning unit 5 learns by associating the ISAR image data corresponding to the integral time input to the integral time input unit 4, which is the basis for classification (determination), with the type of target 1 based on the ISAR image data and the integral time. Information on the type of target 1 may be added to the ISAR image data when performing step 1.

In FIG. 4, the target classification device 6 uses the learning results (learning model) of the ISAR image data learning device 2 shown in FIG. 1 and FIG. 4 (target classification device according to embodiment 1). The new ISAR image input unit 7 is a unit to which newly obtained ISAR image data is input. The new integral time input unit 8 is a unit to which the integral time for each newly obtained ISAR image data input to the new ISAR image input unit 7 is input. The new ISAR image data (newly obtained ISAR image data) and new integral time (newly obtained integral time) referred to here may be obtained by the radar device 10 (radar device according to embodiment 1) described later or by the calculation of a subsequent circuit. The calculation of a subsequent circuit may, for example, use the ISAR image generation unit 13 for the new calculation or the integral processing unit 14 described later.

In FIG. 4, the target classification unit 9 classifies the target 1 corresponding to the newly obtained ISAR image data based on the learning results learned by the learning unit 5, and determines the type corresponding to the target 1. In addition, if the learning unit 5 has also learned information on the distance from the observation point to the target 1, the target classification unit 9 may convert the newly obtained ISAR image data to a predetermined scale based on the learning results learned by the learning unit 5, and then classify the corresponding target 1 and determine the type corresponding to the target 1.

In order to speed up processing, the target classification unit 9 may limit the range of learning results used for classification based on the integral time for each newly obtained ISAR image data. Note that when the learning unit 5 generates learning results for each integral time, the target classification unit 9 can classify target 1 based on the learning results that the learning unit 5 has learned and generated for each integral time, and determine the type that corresponds to target 1.

As the learning unit 5 progresses, when the integral time for each newly obtained ISAR image data does not match the integral time input to the integral time input unit 4, or when the error from the integral time input to the integral time input unit 4 is not within a predetermined range, the target classification unit 9 is able to classify target 1 based on the learning result of the closest angle and determine the type corresponding to target 1.

The new ISAR image input unit 7 may also be configured to input multiple newly obtained ISAR image data having the same target 1 but different integral times. The new ISAR image input unit 7 may also be configured to input multiple newly obtained ISAR image data having the same target 1 and integral time but different integral start times and integral end times for the integral times. In these cases, the target classification unit 9 can classify the corresponding target 1 for each of the multiple newly obtained ISAR image data based on the learning results learned by the learning unit 5, and determine the type corresponding to the target 1, thereby improving the accuracy of the determination.

Of course, in these cases (when multiple newly obtained ISAR image data with the same target 1 but different integral times are input, or when multiple newly obtained ISAR image data with the same target 1 and integral times but different integral start and end times for the integral times are input), the target classification unit 9 may determine the type corresponding to the target 1 if the results of classifying the corresponding target 1 for each of the multiple newly obtained ISAR image data all match. Also, when the results of classifying the corresponding target 1 for each of the multiple newly obtained ISAR image data include ones that do not match and the number of ones that do not match is small compared to the number of ones that match, the target classification unit 9 may determine the type corresponding to the target 1 as a result of classifying the ones that match.

If the learning unit 5 performs comprehensive learning using multiple ISAR image data with different integration times, the target classification unit 9 can also perform comprehensive classification. Since the integration time can be freely used to generate data for various cases through signal processing, it is clearly more advantageous than using natural environmental conditions both when the learning unit 5 is learning and when the target classification unit 9 is classifying.

Similarly, in FIG. 4, the radar device 10 having the target classification device 6 (ISAR image data learning device 2 and target classification device 6) has the following configuration. The tracking radar 11 is the same radar as described above, and tracks the target 1. The inverse synthetic aperture radar 12 (ISAR 12) transmits a transmission beam to the target 1 tracked by the tracking radar 11 and receives radio waves reflected from the target 1. The integration processing unit 14 integrates the video signal obtained from the radio waves reflected from the target 1 in order to generate newly obtained ISAR image data and send it to the new ISAR image input unit 7 together with the inverse synthetic aperture radar 12. In other words, the integration processing unit 14 (similar to the ISAR image generation unit 13) can be said to generate ISAR image data by superimposing multiple ISAR images.

As described above, multiple newly obtained ISAR image data having the same target 1 but different integration times are input to the new ISAR image input unit 7. Therefore, the target classification unit 9 can classify the corresponding target 1 for each of the multiple newly obtained ISAR image data based on the learning results learned by the learning unit 5, and determine the type corresponding to the target 1, thereby improving the accuracy of the determination.

Finally, the operation of the target classification device according to embodiment 1 (the target classification method according to embodiment 1) will be described with reference to FIG. 5. In FIG. 5, step 11 is a processing step in which newly obtained ISAR image data is input to the new ISAR image input unit 7. Step 12 is a processing step in which newly obtained ISAR image data is input to the new integral time input unit 8. The order of processing of steps 11 and 12 does not matter. They may be performed simultaneously. The new ISAR image data (newly obtained ISAR image data) and new integral time (newly obtained integral time) referred to here may be obtained by the radar device 10 (the radar device according to embodiment 1).

Step 13 is a processing step in which new ISAR image data (newly obtained ISAR image data) and new integral time (newly obtained integral time) are input from the new ISAR image input unit 7 and the new integral time input unit 8 to the learning unit 5, and the learning model is used. Step 14 is a processing step in which target 1 corresponding to the newly obtained ISAR image data is classified based on the learning result (learning model) learned by the learning unit 5, and the type corresponding to target 1 is determined. Other target classification methods according to embodiment 1 are mainly the same as the operation of the target classification device according to embodiment 1 described above, so their explanations are omitted.

In the ISAR image data learning device of embodiment 1, and the target classification device and radar device using the same, the ISAR image data and integral time (observation time) learned by the learning unit 5 may be simulated data obtained by simulation. Similarly, the new ISAR image data and integral time (observation time) classified by the target classification unit 9 may also be simulated data obtained by simulation. By having the target classification unit 9 classify using this simulated data, the performance of the learning unit 5 and the target classification unit 9 can be confirmed. In other words, the ISAR image data and integral time (observation time) generated by the ISAR image generation unit 13 may also be simulated data obtained by simulation.

As described above, the ISAR image data learning device of embodiment 1, and the target classification device and radar device using the same learn by associating the ISAR image data corresponding to the integral time with the type of target 1, or use the learning results, and can therefore classify target 1 using the integral time.

1 Target 1, 2 ISAR image data learning device, 3 ISAR image input unit,
4 Integration time input unit, 5 Learning unit, 6 Target classifier, 7 New ISAR image input unit,
8 New integration time input unit, 9 Target classification unit, 10 Radar device,
11 tracking radar, 12 inverse synthetic aperture radar (ISAR), 13 ISAR image generator,
14. Integral processing unit.

Claims (20)

An ISAR image data learning device that generates learning results for classifying targets from ISAR image data obtained by an inverse synthetic aperture radar installed at an observation point, comprising:
an ISAR image input unit to which the ISAR image data is input;
an integral time input unit to which an integral time for deriving the ISAR image data is input;
a learning unit configured to learn to classify the target from the ISAR image data corresponding to the integral time input to the integral time input unit. The ISAR image data learning device according to claim 1 , wherein the learning unit associates the ISAR image data corresponding to the integral time input to the integral time input unit with a type of the target. The ISAR image data learning device of claim 2 , wherein the type of the target is information including at least one of the name of the target, the dimensions of the target, the performance of the target, the country to which the target belongs, and the organization to which the target belongs. The ISAR image data learning device according to any one of claims 1 to 3 , wherein the integral time input unit inputs the integral time in which an integral start time and an integral end time are set between an observation start time and an observation end time of the target. 5. The ISAR image data learning device according to claim 4 , wherein the integral time input unit inputs the integral times, the integral start times being the same and the integral times being different . The ISAR image data learning device according to claim 5 , wherein the integral time input unit inputs the integral time in which the integral start time is the same and the integral end time is set within the time it takes for the period of the target's vibration to go round trip. 5. The ISAR image data learning device according to claim 4 , wherein the integral time input unit inputs the integral time from the integral start time to the integral end time , the integral time corresponding to one round trip of a period of the target's motion. The ISAR image data learning device further includes an ISAR image generation unit,
The ISAR image data learning device according to claim 1 , wherein the ISAR image input unit and the integral time input unit respectively receive the ISAR image data and the integral time output from the ISAR image generation unit. 9. The ISAR image data learning device according to claim 1 , further comprising: an ISAR image input unit to which the ISAR image data associated with information indicating the distance from the observation point to the target is input; or an integral time input unit to which the integral time associated with information indicating the distance from the observation point to the target is input. The ISAR image data learning device according to claim 9 , wherein the learning unit performs learning using the ISAR image data obtained by converting the ISAR image data into a predetermined scale based on the information indicating the distance. an ISAR image input unit to which ISAR image data is input;
an integral time input unit to which an integral time for deriving the ISAR image data is input;
a learning unit that classifies and learns targets from the ISAR image data corresponding to the integral time input to the integral time input unit ;
a new ISAR image input unit to which newly obtained new ISAR image data is input;
a new integration time input unit to which an integration time for deriving the new ISAR image data is input;
A target classification device comprising: a target classification unit that classifies the target from the new ISAR image data based on a learning result learned by the learning unit. The target classification device according to claim 11 , wherein the target classifier determines a type of the target corresponding to the target. The target classification device according to claim 11 or 12 , wherein the target classification unit limits a range of the learning results used for classification from the integration time for each of the new ISAR image data . The target classification device according to claim 11 , wherein the target classification unit classifies the target based on the learning result generated by the learning unit during learning for each integral time, and determines the type of the target corresponding to the target . 15. The target classification device according to claim 11 , wherein when the integral time for each of the new ISAR image data does not match the integral time input to the integral time input unit, or when an error from the integral time input to the integral time input unit is not within a predetermined range, the target classification unit classifies the target based on the learning result of the closest time, and determines the type of the target corresponding to the target . The new ISAR image input unit receives a plurality of new ISAR image data sets having the same target and different integration times,
the new integration time input unit receives a plurality of different integration times corresponding to the plurality of new ISAR image data,
The target classification unit classifies the targets corresponding to each of the plurality of new ISAR image data based on the learning result learned by the learning unit, and determines the type of the target corresponding to the target . The target classification device according to any one of claims 11 to 15 . The new ISAR image input unit receives a plurality of new ISAR image data sets each having the same target and integral time, and each having a different integral start time and an integral end time,
The target classification unit classifies the targets corresponding to each of the multiple new ISAR image data based on the learning results learned by the learning unit, and determines the type of the target corresponding to the target . The target classification device described in any one of claims 11 to 15 . The target classification device according to claim 16 or 17 , wherein the target classification unit determines the type of the target corresponding to the target when all results of classifying the corresponding targets for each of the multiple new ISAR image data match. The target classification unit determines the type of the target corresponding to the target as a result of classifying the matching targets when the results of classifying the corresponding targets for each of the multiple new ISAR image data include mismatches and the number of mismatches is small compared to the number of matches . The target classification device according to claim 16 or 17 . an ISAR image input unit to which ISAR image data is input;
an integral time input unit to which an integral time for deriving the ISAR image data is input;
a learning unit that classifies and learns targets from the ISAR image data corresponding to the integral time input to the integral time input unit;
a new ISAR image input unit to which newly obtained new ISAR image data is input;
a new integration time input unit to which an integration time for deriving the new ISAR image data is input;
a target classification unit that classifies the target from the new ISAR image data based on a learning result learned by the learning unit;
a tracking radar for tracking the target;
an inverse synthetic aperture radar that transmits a transmission beam to the target tracked by the tracking radar and receives radio waves reflected from the target ;
and an integration processing unit that integrates a video signal obtained from radio waves reflected from the target to generate the new ISAR image data and send it to the new ISAR image input unit.

Images