JP7186816B2 - RADAR DEVICE, SIGNAL PROCESSING DEVICE, AND SIGNAL PROCESSING METHOD - Google Patents

Description

TECHNICAL FIELD Embodiments of the present invention relate to a radar device, a signal processing device, and a signal processing method.

Radar devices are known, for example, which are mounted on aircraft and for detecting incoming targets. For this type of radar device, for example, it has been proposed to suppress clutter using a pattern classifier to which machine learning is applied.

JP 2021-015079 A

D Gusland, S Rolfsjord, B Torvik , "Deep temporal detection-A machine learning approach to multiple-dwell target detection," 2020 IEEE International Radar Conference (RADAR), April 2020.

If the target is far away or if the RCS is small, the S/N (Signal to Noise ratio) of the received signal will be small. Enlarging the aperture area of the radar antenna and increasing the transmission power are effective in increasing the S/N, but the scale of the hardware increases. Also, the S/N can be increased by increasing the number of pulses used for coherent integration or the like, but this may increase the update rate or limit the detection area. Furthermore, in the case of a fast-moving target, the detection performance is significantly degraded. Therefore, it is necessary to improve the detection performance of fast-moving targets without increasing the hardware size of the radar antenna by using a signal processing method using machine learning that is different from the conventional method. However, in order to deal with moving targets of various speeds, it is necessary to train pattern classifiers that can deal with various speeds, which increases the scale of machine learning.

Accordingly, it is an object of the present invention to provide a radar device, a signal processing device, and a signal processing method that reduce noise and improve the performance of detecting a moving target.

According to an embodiment, a radar device comprises an antenna section, a transceiver section, and a signal processing section. The transmitting/receiving unit transmits radar pulses from the antenna unit, receives radio waves arriving at the antenna unit, and generates a received signal. The signal processor processes the received signal. The signal processing section includes a pulse compression processing section, an integration processing section, a sensitivity enhancement processing section, and a detection processing section. The pulse compression processing unit pulse-compresses the received signal to generate a compressed pulse signal. The integration processing unit pulse-integrates the compressed pulse signal to generate an integrated pulse signal. The sensitivity enhancement processing section reduces noise in the integrated pulse signal to generate a sensitivity enhancement signal. A detection processing unit detects a target from the sensitized signal.

FIG. 1 is a functional block diagram showing an example of a radar device according to an embodiment. FIG. 2 is a functional block diagram showing an example of a signal processing unit according to the first embodiment; FIG. 3 is a functional block diagram illustrating an example of a sensitivity enhancement processing unit according to the embodiment; 4 is a flowchart illustrating an example of a processing procedure of a signal processing unit according to the embodiment; FIG. FIG. 5 is a diagram for explaining integration processing and sensitivity enhancement processing. FIG. 6 is a functional block diagram showing an example of a signal processing section according to the second embodiment. FIG. 7 is a functional block diagram showing an example of a signal processing section according to the third embodiment. FIG. 8 is a functional block diagram showing an example of a signal processing section according to the fourth embodiment. FIG. 9 is a functional block diagram showing an example of a signal processing section according to the fifth embodiment. FIG. 10 is a diagram for explaining the dimensionality reduction processing of received pulse data. FIG. 11 is a functional block diagram showing an example of a signal processing section according to the sixth embodiment. FIG. 12 is a functional block diagram showing an example of a signal processing unit according to the seventh embodiment; FIG. 13 is a diagram showing an example of a movement trajectory of a target for each azimuth. FIG. 14 is a diagram illustrating an example of a target detection result; FIG. 15 is a diagram for explaining the movement correction process. FIG. 16 is a diagram for explaining integration processing, movement amount correction processing, sensitivity enhancement processing, and target detection processing.

Embodiments will be described below with reference to the drawings.
FIG. 1 is a functional block diagram showing an example of a radar device according to an embodiment. A pulse radar is assumed in the embodiment. As shown in FIG. 1 , the radar device 1 includes an antenna section 10 , a transmission/reception section 11 and a signal processing section 12 . The radar device 1 may also include a display unit 13 for visually displaying the output of the signal processing unit 12 and presenting it to the observer (user).

The antenna unit 10 is, for example, an array antenna having a plurality of regularly arranged antenna elements. The antenna unit 10 transmits radio waves (radar waves) into space and receives radio waves including reflected waves (received signals) from targets.

The transmitting/receiving section 11 generates a radar pulse and sends it to the antenna section 10 . Further, the transmitting/receiving section 11 captures radio waves arriving at the antenna section 10, performs processing such as amplification and analog/digital (A/D) conversion, and generates a received signal. The generated received signal (received data) is sent to the signal processing section 12 .

The signal processing unit 12 processes the received signal to perform signal processing such as clutter suppression, radar video generation, and target detection. The detected targets are sent to the display unit 13 and visually displayed on a display or the like. The signal processing unit 12 is, for example, a computer having a memory and a processor, and processes the reception signal (radar reception signal: reception data) output from the transmission/reception unit 11 .
Next, a plurality of embodiments will be described based on the above configuration.

[First embodiment]
FIG. 2 is a functional block diagram showing an example of a signal processing unit according to the first embodiment; The signal processing section 12 includes a pulse compression processing section 121 , an integration processing section 122 , a sensitivity enhancement processing section 123 and a detection processing section 124 .
The pulse compression processing unit 121 pulse-compresses the received signal from the transmission/reception unit 11 to generate a compressed pulse signal.
The integration processing unit 122 pulse-integrates the compressed pulse signal from the pulse compression processing unit 121 to generate an integrated pulse signal. Coherent integration is typical of the integration processing method, but non-coherent integration can also be applied.

The sensitivity enhancement processing unit 123 reduces (denoises) noise in the integrated pulse signal from the integration processing unit 122 to generate a sensitivity enhancement signal.

The detection processing unit 124 detects a target from the sensitivity enhancement signal from the sensitivity enhancement processing unit 123 .

FIG. 3 is a functional block diagram showing an example of the sensitivity enhancement processing section 123. As shown in FIG. The sensitivity enhancement processing unit 123 is, for example, a computer having an entity as hardware, and includes a processor 21 and a storage unit 24 . Furthermore, the sensitivity enhancement processing unit 123 includes a ROM 22 , a RAM 23 and a communication unit 25 .

The ROM 22 is a non-volatile memory and functions as a program memory. The ROM 22 stores programs executed by the processor 21, control data, and the like.
The RAM 23 functions as a working memory that temporarily holds data. The RAM 23 is loaded with programs and holds data being processed by the processor 21 . The RAM 23 also functions as a buffer memory that temporarily holds data and the like.
The storage unit 24 is a non-volatile memory such as an SSD (Solid State Drive). The storage unit 24 stores a program 24a executed by the processor 21, data 24b used by the processor 21, and a learned model 24c.

The processor 21 is, for example, a computing device such as a CPU (Central Processing Unit) or an MPU (Micro Processing Unit). The processor 21 implements processing functions according to the embodiment by reading programs and data stored in the ROM 22 or the storage unit 24 into the RAM 23 and executing them.

The communication unit 25 is connected to the integration processing unit 122 and acquires the integrated pulse signal converted into data. In addition, the communication unit 25 is connected to the detection processing unit 124 and outputs the digitized sensitization signal to the detection processing unit 124 .

By the way, the processor 21 has a learning unit 21a and an identification unit 21b as processing functions according to the embodiment. These are processing functions realized by the processor 21 executing instructions written in the program 24a.

The learning unit 21a repeats machine learning using the integrated pulse signal from the integration processing unit 122 to generate a trained model. The generated learned model is stored in the storage unit as a learned model 24c.

For example, training data can be repeatedly fed to a convolutional neural network (CNN) to generate a trained model. This kind of machine learning is known as deep learning. That is, in the first embodiment, the learning unit 21a repeats supervised machine learning using the integrated pulse signal as learning data to generate the trained model 24c.
The identification unit 21b reduces noise in the integrated pulse signal from the integration processing unit 122 by performing identification processing using the trained model 24c to generate a high sensitivity signal.

Note that the sensitivity enhancement processing unit 123 can also be implemented as software. In this case, the sensitivity enhancement processing unit 123 can be understood as a program (process) developed in a memory and executed by a processor. Next, the operation of the above configuration will be described.

FIG. 4 is a flowchart showing an example of the processing procedure of the signal processing section 12. As shown in FIG. In FIG. 4, the signal processing unit 12 pulse-compresses the radar reception signal from the transmission/reception unit 11 to generate a compressed pulse signal (step S1). Next, the signal processing unit 12 pulse-integrates the compressed pulse signal to generate an integral pulse signal (step S2).

Next, the signal processing unit 12 reduces noise in the integral pulse signal to generate a sensitivity enhancement signal (step S3). Further, the signal processing unit 12 detects a target from the sensitized signal (step S4).

FIG. 5 is a diagram for explaining integration processing and sensitivity enhancement processing in the procedure of FIG. In FIG. 5, received pulse data received in time series are grouped into a plurality of units, and integration processing I(·) is performed for each unit. For example, the left side of Equation (1) indicates the integration processing result X _I (T) at time T. The brackets of I(·) on the left side of Equation (1) indicate a set of pulse data to be integrated. A set of pulse data is M pieces of pulse data including pulse data at time T. FIG.

Similarly, integration processing is also performed on sets of past received pulse data. Equation (2) shows the result of past N times integration processing.

The result of the past N times of integration processing is passed to the sensitivity enhancement processing unit 123 and subjected to noise reduction processing D(·). This yields the noise-reduced result Y(T) of XI(T), as shown in equation (3).

However, in Equation (3), T _Sk (k=1, 2, . . . , N) represents the integration result before T _Sk .

As described above, according to the first embodiment, the sensitivity enhancement process is performed on the received pulse data whose S/N is improved by the integration process. As a result, it is possible to further improve the S/N with respect to the result after the integration process. By capturing the characteristics of the target using a highly sensitive signal, it becomes possible to detect the target with a smaller number of pulse hits than with existing techniques.

In the existing technology, when the S/N of the received signal is low, it is not possible to obtain a sufficiently strong signal only by the integration process, and the signal becomes lower than the threshold value in the detection process, making detection difficult. Lowering the threshold results in false positives, where noise is detected as a target. If the number of pulse hits in the integration process is increased as a countermeasure against this, the time required to acquire data per azimuth increases, leading to a decrease in the update rate of the entire search range. This is a problem when detecting and tracking fast moving targets. Further, increasing the aperture area and transmission power of the antenna may be considered as an improvement of the S/N, but the ease of installation is impaired due to the increase in the size of the antenna.

On the other hand, according to the first embodiment, when detecting a target with the same radar reflection cross section (RCS), the same detection accuracy is maintained even if the hardware scale is smaller than that of the existing technology. be able to.

Furthermore, in the first embodiment, the sensitivity enhancement process is performed by machine learning using the received signal after integration (integrated pulse signal). That is, data including a low S/N target after integration processing is used as learning data, and learning is performed using only the same target data as corresponding correct data. As a result, the trained model 24c of the sensitivity enhancement processing unit 123 is trained so as to bring the low S/N target data closer to noise-free data. By using this trained model 24c, noise in the low S/N target data after integration processing is reduced. In this way, by performing the sensitivity enhancement process by machine learning on the data after the integration process, it becomes possible to detect a target that was difficult to detect only by the integration process. Also, the learning may be performed so that the target-likeness increases at the target position and decreases in the noise region.

Normally, the number of pulses used for coherent integration is set according to the smallest RCS target among assumed targets. The number of integrations would then be excessive for the other RCS targets. Therefore, for example, it is conceivable to suppress the number of times of integration by setting a medium RCS as a target to be detected by coherent integration processing, and to detect a low RCS target by the proposed method. As a result, the effect of reducing resource shortages in beam scheduling due to an increase in the number of integrations can be expected.

For these reasons, according to the first embodiment, it is possible to provide a radar apparatus, a signal processing apparatus, and a signal processing method with reduced noise and improved target detection performance. As a result, the hardware scale can be reduced, and a radar device that is advantageous for installation on an aircraft or the like can be realized.

[Second embodiment]
FIG. 6 is a functional block diagram showing an example of a signal processing section according to the second embodiment. The signal processing unit 12 shown in FIG. 6 includes a CFAR processing unit 125 between the integration processing unit 122 and the sensitivity enhancement processing unit 123 in addition to the configuration of FIG.

The CFAR processing unit 125 performs CFAR (Constant False Alarm Rate) processing on the integrated pulse signal from the integration processing unit 122 to generate a smoothed signal in which variations in signal strength are smoothed.

For example, CFAR processing is known for suppressing clutter. Signals due to clutter have the characteristic of being received over a wide range. Therefore, in order to reduce fluctuations in signal intensity due to clutter, CFAR processing is performed to obtain a smoothed signal, which is input to the sensitivity enhancement processing section 123 . Here, for CFAR, for example, CA-CFAR or GO-CFAR may be used.

That is, in the second embodiment, the learning unit 21a (FIG. 3) repeats machine learning using the smoothed signal as learning data to generate the trained model 24c.
The identification unit 21b (FIG. 3) reduces noise in the smoothed signal from the CFAR processing unit 125 by performing identification processing using the trained model 24c to generate a high-sensitivity signal.

In machine learning processing, it is difficult to predict what kind of results will be output when data with features not assumed during learning is given. In this case, there is a possibility that a processing result that adversely affects subsequent processing may be output. For this reason, it is desirable to limit the characteristics of the input data to a range that can be limited in advance as much as possible.

Therefore, in the second embodiment, CFAR processing is used to keep the dynamic range of the received signal within a certain range, thereby promoting stable operation of the learning processing and the identification processing. As a result, a reduction in the number of signal intensity patterns for creating learning data during learning can be expected.

For these reasons, the second embodiment can also obtain the same effect as the first embodiment.

[Third Embodiment]
FIG. 7 is a functional block diagram showing an example of a signal processing section according to the third embodiment. The signal processing unit 12 shown in FIG. 7 includes a detection processing unit 126 between the integration processing unit 122 and the sensitivity enhancement processing unit 123 in addition to the configuration of FIG.

The detection processing unit 126 is a pre-detection processing unit that detects a high S/N target from the integrated pulse signal from the integration processing unit 122, replaces the high S/N target with a noise component, and generates a replacement signal. .

That is, in the third embodiment, the learning unit 21a (FIG. 3) repeats machine learning using the replacement signal as learning data to generate the trained model 24c.
The identification unit 21b (FIG. 3) reduces noise in the replacement signal from the detection processing unit 126 by performing identification processing using the trained model 24c to generate a high-sensitivity signal.

In the third embodiment, a high S/N target is detected in advance by threshold determination, for example, and a process of replacing it with a noise component is performed before the sensitivity enhancement process. Here, the noise component to be replaced may be, for example, a pseudo-random number that matches the noise level of the received signal. Also, the range to be replaced with noise may be set to a range with a margin with respect to the detection range.

Radar reception data may have a high signal strength. When trying to detect small signals, learning large signals can lead to overall performance degradation. Therefore, in the third embodiment, a signal with a large reception intensity is detected by, for example, threshold determination after integration processing. Then, the detected portion of high signal strength is replaced with, for example, pseudo-noise based on random numbers. This makes it possible to determine the upper limit of the signal input to the machine learning process, and stable operation of the process can be expected. In addition, reduction in the number of signal intensity patterns for creating learning data at the time of learning can be expected.

When using machine learning, if a signal with an unexpectedly high S/N is input, an unexpected result may be output, which may adversely affect subsequent processing. Therefore, by detecting a high S/N target in advance and replacing it with noise, the influence can be reduced.

For these reasons, the third embodiment can also obtain the same effects as those of the first and second embodiments. Furthermore, the effect of improving the efficiency of machine learning can also be obtained.

[Fourth embodiment]
FIG. 8 is a functional block diagram showing an example of a signal processing section according to the fourth embodiment. The signal processing unit 12 shown in FIG. 8 includes, in addition to the configuration of FIG.

The region division processing unit 127 limits the region targeted for target detection processing, extracts a signal related to the region from the integrated pulse signal from the integration processing unit 122, and generates an extraction signal.
That is, in the fourth embodiment, the learning unit 21a (FIG. 3) repeats machine learning using the extracted signal as learning data to generate the trained model 24c.
The identification unit 21b (FIG. 3) reduces noise in the extracted signal from the segmentation processing unit 127 by performing identification processing using the trained model 24c to generate a high-sensitivity signal.

For example, since the S/N ratio of a signal received from a distant target decreases, the segmentation processing unit 127 performs processing for extracting only the far-distance part. If the azimuth and elevation angle for performing the processing are determined in advance, the area may be extracted by the azimuth and elevation angle. For the limitation of the processing range, the position information of the processing target may be used in addition to the processing of extracting the processing range.

As described above, according to the fourth embodiment, it can be expected that the amount of processing can be reduced by limiting the processing range. As a result, reduction in memory capacity and shortening of processing time can be expected.

For these reasons, the fourth embodiment can also obtain the same effects as those of the first and second embodiments.

[Fifth Embodiment]
FIG. 9 is a functional block diagram showing an example of a signal processing section according to the fifth embodiment. The signal processing unit 12 shown in FIG. 9 includes a dimensionality reduction processing unit 128 between the integration processing unit 122 and the sensitivity enhancement processing unit 123 in addition to the configuration of FIG.

The dimensionality reduction processing unit 128 reduces the dimensionality of the integrated pulse signal from the integration processing unit 122 to generate a dimensionality reduction signal.

That is, in the fifth embodiment, the learning unit 21a (FIG. 3) repeats machine learning using the dimensionality reduction signal as learning data to generate the trained model 24c.
The identification unit 21b (FIG. 3) reduces noise in the dimensionality reduction signal from the dimensionality reduction processing unit 128 by identification processing using the trained model 24c to generate a high sensitivity signal.

FIG. 10 is a diagram for explaining the dimensionality reduction processing of received pulse data. Assume that five types of integrated pulse data are obtained in different directions (cells), as shown in FIG. 10(a). For example, the number-of-dimensionality reduction processing unit 128 reduces the number of dimensions of data by performing maximum value selection or averaging processing on cells in an arbitrary range. For example, selecting the maximum value in range 3 gives the result shown in FIG. 10(b). The results of processing by sliding the target range may be combined.

Note that the number of axes on which each integral pulse data is plotted may be reduced as the dimensionality reduction process. For example, when the integrated pulse data further includes elevation angle information, this elevation angle information may be reduced. Further, for example, the data of the elevation angle may be reduced from the data of the azimuth, the elevation angle, and the like.

By reducing the number of dimensions of data, it is expected that the amount of processing will be reduced. As a result, reduction in memory capacity and shortening of processing time can be expected.
For these reasons, the fifth embodiment can also obtain the same effects as those of the first and second embodiments. In addition, it can help reduce the required resources.

[Sixth embodiment]
FIG. 11 is a functional block diagram showing an example of a signal processing section according to the sixth embodiment. The signal processing unit 12 shown in FIG. 11 includes a movement correction processing unit 129 between the pulse compression processing unit 121 and the integration processing unit 122 in addition to the configuration shown in FIG.

A movement correction processing unit 129 compensates for the movement of the target in the compressed pulse signal from the pulse compression processing unit 121 and generates a compensation signal. Then, the integration processing unit 122 performs pulse integration on the compensation signal to generate an integration pulse signal, and inputs it to the sensitivity enhancement processing unit 123 .

That is, in the sixth embodiment, the learning unit 21a (FIG. 3) repeats machine learning using the movement-corrected integrated pulse signal as learning data to generate the learned model 24c.
The identification unit 21b (FIG. 3) reduces noise in the movement-corrected integration pulse signal from the integration processing unit 122 by identification processing using the trained model 24c to generate a high-sensitivity signal.

If the target is moving, the signal may not build up in the integration process. Therefore, in the sixth embodiment, by compensating for the movement of the target, it is possible to efficiently improve the S/N by integration processing. This will be explained using mathematical formulas.

As shown in equation (4), the motion compensation processor performs motion compensation processing R(·) on the received signal X(T). X _R (T) on the left side of Equation (4) indicates the compensation signal input to integration processing section 122 .

The motion compensation process may shift the range cells by the required amount. Further, for example, the pulse compression processing unit 121 may limit the band and perform processing with reduced distance resolution. In this case, the pulse compression processing unit 121 performs motion compensation processing.

The integration processing unit 122 performs the integration processing I(·) on past M pieces of motion-compensated data shown in Equation (5).

If the target is moving, the position of the target is different in each piece of data after integration processing. Therefore, even if integration processing is performed between each piece of data after integration processing, there is a concern that target signals will not accumulate. Therefore, in the sixth embodiment, in consideration of the movement of the target between integration processes, a compensation signal is generated in which the amount of movement of the target is corrected so as to cancel the movement between integration processes. As a result, the S/N can be effectively improved by integration processing.

Also, by giving the compensation signal to the model as learning data, it is possible to create a learned model that considers the movement of the target during the integration process. As a result, even if the target exists at different positions between integration processes, high sensitivity processing is possible.

As described above, according to the sixth embodiment, it is possible to perform machine learning that considers the movement of the target, thereby eliminating the adverse effects caused by changes in the target position after integration processing. In other words, normal video integration requires movement correction processing, which may increase the amount of processing. On the other hand, in the embodiment, effective processing can be expected even when the target is moving by using data that takes into consideration the deviation of the target position between data during learning.

For these reasons, according to the sixth embodiment, it is possible to provide a radar device, a signal processing device, and a signal processing method that reduce noise and improve detection performance of a moving target.

[Seventh Embodiment]
FIG. 12 is a functional block diagram showing an example of a signal processing unit according to the seventh embodiment; The signal processing unit 12 shown in FIG. 12 includes a movement correction processing unit 129 between the integration processing unit 122 and the sensitivity enhancement processing unit 123 in addition to the configuration shown in FIG.

A movement correction processing unit 129 compensates for the movement of the target in the integrated pulse signal from the integration processing unit 122 and generates a compensation signal. Then, the sensitivity enhancement processing unit 123 reduces noise in the compensation signal from the movement correction processing unit 129 to generate a sensitivity enhancement signal.

That is, in the seventh embodiment, the learning unit 21a (FIG. 3) repeats machine learning using the compensation signal as learning data to generate the trained model 24c.
The identification unit 21b (FIG. 3) reduces noise in the compensation signal from the movement correction processing unit 129 by performing identification processing using the learned model 24c to generate a high sensitivity signal. Movement correction processing in the seventh embodiment will be described with reference to FIGS. 13 to 15. FIG.

FIG. 13 is a diagram showing an example of a movement trajectory of a target for each azimuth. In FIG. 13, received data 30 before pulse compression includes data for each of multiple azimuths. Reference numeral 32 indicates the movement trajectory of the target plotted with distance and time in each direction. The moving speed of the target is represented as the slope of the moving trajectory 32 . For example, when monitoring the sea, sea clutter 31 may be present.

FIG. 14 is a diagram illustrating an example of a target detection result; As shown in FIG. 14, the target is observed together with sea clutter 31 . In the time-distance data, the sea clutter 31 is a short line with small amplitude and width in the spatial direction of the signal. On the other hand, since the target has a greater amplitude and width in the spatial direction than the sea clutter 31, it draws a highly continuous trajectory 32. FIG.

FIG. 15 is a diagram for explaining the movement correction process. The slope of the target trajectory 32 is intermediate in the intermediate speed range, increases in the high speed range, and decreases in the low speed range. Therefore, in the seventh embodiment, the integral pulse signal after pulse integration is converted spatially and temporally, and compensated to a common reference speed to generate a compensation signal. Then, this compensation signal is given to the sensitivity enhancement processing unit 123 to learn it, thereby performing the sensitivity enhancement processing.

FIG. 16 is a diagram for explaining integration processing, movement amount correction processing, sensitivity enhancement processing, and target detection processing. In FIG. 16, received pulse data (IQ data) received in time series are grouped into units for each CPI (Coherent Processing Interval), and coherent integration processing is performed within each CPI. The integrated pulse signal (data after coherent integration) thus obtained is corrected at speed A, speed B, . Then, noise reduction processing (denoise) is performed on each compensation signal by the sensitivity enhancement processing unit 123 to generate a sensitivity enhancement signal (noise-reduced data). This sensitization signal is given to the detection processing unit 124, and the target is detected. Here, the target detection process may be performed in combination with the CFAR process.

Note that the target speed is not known when the high-sensitivity processing unit 123 is actually used. In this case, the movement correction processing unit 129 sequentially generates compensation signals while changing the speed correction value. Then, the movement compensation processing section 129 may determine the target velocity based on the velocity correction value when the compensation signal with the highest S/N is obtained. After the target speed is determined, the movement compensation processing unit 129 performs correction according to the speed.

As described above, according to the seventh embodiment, with respect to data after coherent integration processing for each scan, the moving speed of the target is considered in advance so that the target moves within a certain constant speed. correct the amount of movement. Further, using the corrected multiple scan data, noise reduction processing is performed by a deep-learned noise removal convolutional neural network to detect the target when the S/N is low.

By using multiple scan data, differences in target and noise characteristics between scans can be learned. Therefore, noise reduction can be expected more than video integration processing, and the target can be detected when the S/N is low.

In the existing technology, regardless of whether the target speed range of the object is small, the scale of the learning data becomes large when the range of the target movement speed becomes large. In addition, the scale of the denoising convolutional neural network becomes large, and the amount of calculation becomes bloated. In addition, since machine learning does not have guarantees based on data considering the movement of the target, it is expected that the performance will decrease as the moving speed of the target increases.

On the other hand, in the seventh embodiment, by correcting the amount of movement of the target, the scale of the convolutional neural network based on deep learning can be reduced, and the processing amount of the computer can be reduced. Therefore, according to the seventh embodiment, it is possible to provide a radar device, a signal processing device, and a signal processing method that reduce noise and improve detection performance of a moving target. Furthermore, it is possible to prevent the trained model from becoming bloated, reduce hardware resources, and realize a radar device that is particularly advantageous for aircraft-mounted applications.

It should be noted that the present invention is not limited to the above embodiments. For example, in the movement correction processing of the sixth and seventh embodiments, it is possible to perform correction considering not only movement in the distance direction, but also movement in the angular direction, or both.

While several embodiments of the invention have been described, these embodiments have been presented by way of example and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, replacements, and modifications can be made without departing from the scope of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are included in the scope of the invention described in the claims and equivalents thereof.

DESCRIPTION OF SYMBOLS 1... Radar apparatus 10... Antenna part 11... Transmission/reception part 12... Signal processing part 13... Display part 21... Processor 21a... Learning part 21b... Identification part 22... ROM 23... RAM 24... Memory unit 24a Program 24b Data 24c Trained model 25 Communication unit 30 Received data 31 Sea clutter 32 Target movement trajectory 121 Pulse compression processing unit 122 Integration processing unit , 123... Sensitivity enhancement processing unit, 124... Detection processing unit, 125... CFAR processing unit, 126... Detection processing unit, 127... Area division processing unit, 128... Number of dimensions reduction processing unit, 129... Movement correction processing unit.

Claims (11)

an antenna section;
a transmitting/receiving unit that transmits a radar pulse from the antenna unit, receives radio waves including reflected waves from a target arriving at the antenna unit, and generates a received signal;
A signal processing unit that processes the received signal,
The signal processing unit is
a pulse compression processing unit that pulse-compresses the received signal to generate a compressed pulse signal;
an integration processing unit that pulse-integrates the compressed pulse signal to generate an integrated pulse signal;
a sensitivity enhancement processing unit that reduces noise in the integrated pulse signal to generate a sensitivity enhancement signal;
a detection processing unit that detects the target from the sensitized signal;
a pre-detection processing unit that detects a high S/N target from the integrated pulse signal and replaces the high S/N target with a noise component to generate a replacement signal;
The radar apparatus, wherein the sensitivity enhancement processing unit reduces noise in the replacement signal to generate the sensitivity enhancement signal. The sensitivity enhancement processing unit is
a learning unit that repeats machine learning using the replacement signal to generate a trained model;
2. The radar apparatus according to claim 1, further comprising a discrimination section that reduces noise in said replacement signal by discrimination processing using said trained model to generate said sensitized signal. an antenna section;
a transmitting/receiving unit that transmits a radar pulse from the antenna unit, receives radio waves including reflected waves from a target arriving at the antenna unit, and generates a received signal;
A signal processing unit that processes the received signal,
The signal processing unit is
a pulse compression processing unit that pulse-compresses the received signal to generate a compressed pulse signal;
an integration processing unit that pulse-integrates the compressed pulse signal to generate an integrated pulse signal;
a sensitivity enhancement processing unit that reduces noise in the integrated pulse signal to generate a sensitivity enhancement signal;
a detection processing unit that detects the target from the sensitized signal;
a reduction processing unit that reduces information of the integrated pulse signal to generate a reduction signal ,
The radar apparatus, wherein the sensitivity enhancement processing unit reduces noise in the reduced signal to generate the sensitivity enhancement signal. The sensitivity enhancement processing unit is
a learning unit that repeats machine learning using the reduced signal to generate a trained model;
4. The radar apparatus according to claim 3, further comprising a discrimination section that reduces noise in said reduced signal by discrimination processing using said trained model to generate said sensitized signal. an antenna section;
a transmitting/receiving unit that transmits a radar pulse from the antenna unit, receives radio waves including reflected waves from a target arriving at the antenna unit, and generates a received signal;
A signal processing unit that processes the received signal,
The signal processing unit is
a pulse compression processing unit that pulse-compresses the received signal to generate a compressed pulse signal;
an integration processing unit that pulse-integrates the compressed pulse signal to generate an integrated pulse signal;
a sensitivity enhancement processing unit that reduces noise in the integrated pulse signal to generate a sensitivity enhancement signal;
a detection processing unit that detects the target from the sensitized signal;
a movement correction processing unit that compensates for the movement of the target in the integrated pulse signal and generates a compensation signal;
The radar apparatus, wherein the sensitivity enhancement processing unit reduces noise in the compensation signal to generate the sensitivity enhancement signal. The sensitivity enhancement processing unit is
a learning unit that repeats machine learning using the compensation signal to generate a trained model;
6. The radar apparatus according to claim 5, further comprising a discrimination section that reduces noise in said compensation signal by discrimination processing using said trained model to generate said sensitized signal. The radar device according to any one of claims 2, 4, and 6, wherein said machine learning is deep learning in which learning data is repeatedly given to a convolutional neural network to generate said trained model. In a signal processing device that processes received signals of radio waves including reflected waves of transmitted radar pulses from targets,
a pulse compression processing unit that pulse-compresses the received signal to generate a compressed pulse signal;
an integration processing unit that pulse-integrates the compressed pulse signal to generate an integrated pulse signal;
a sensitivity enhancement processing unit that reduces noise in the integrated pulse signal to generate a sensitivity enhancement signal;
a detection processing unit that detects the target from the sensitized signal;
a pre-detection processing unit that detects a high S/N target from the integrated pulse signal and replaces the high S/N target with a noise component to generate a replacement signal;
The signal processing device, wherein the sensitivity enhancement processing unit reduces noise in the replacement signal to generate the sensitivity enhancement signal. The sensitivity enhancement processing unit is
a learning unit that repeats machine learning using the replacement signal to generate a trained model;
9. The signal processing apparatus according to claim 8, further comprising a discrimination section that reduces noise in said replacement signal by discrimination processing using said trained model to generate said sensitized signal. In a signal processing method in which a computer processes received signals of radio waves including reflected waves of transmitted radar pulses from targets,
The computer pulse-compresses the received signal to generate a compressed pulse signal;
the computer pulse-integrates the compressed pulse signal to generate an integrated pulse signal;
the computer reduces noise in the integrated pulse signal to generate a sensitized signal;
the computer detects the target from the sensitized signal;
the computer detects a high S/N target from the integrated pulse signal and replaces the high S/N target with a noise component to generate a replacement signal;
A signal processing method, wherein the computer reduces noise in the replacement signal to generate the sensitized signal. The computer repeats machine learning using the replacement signal to generate a learned model,
11. The signal processing method according to claim 10, wherein said computer reduces noise in said replacement signal by discrimination processing using said trained model to generate said sensitized signal.

Images