JP6906428B2 - Radar device and its radar signal processing method - Google Patents

Description

The present embodiment relates to a radar device and a radar signal processing method thereof.

In a conventional radar device, in general, after pulse compression (Non-Patent Document 1) or FFT (Fast Fourier Transform), CFAR (Constant False Alarm Rate, Non-Patent Document 2) processing is often used. However, in the CFAR process, when detecting a small target having a small RCS (Radar Cross Section) and a low reception SN, the reception amplitude does not exceed the threshold of the CFAR, and there is a case where the detection is not performed.

As a countermeasure for this, there is also a method of detecting a target from an amplitude histogram of a block unit of the range-Doppler axis (Patent Document 1), but since a feature amount representing the histogram shape etc. is further extracted from the amplitude histogram shape, an assumed feature amount is used. If the extraction method of was not suitable for the data, it could be undetected.

JP-A-2015-176665

Ouchi, "Basics of Synthetic Aperture Radar for Remote Sensing", Tokyo Denki University Press, pp.131-149 (2003) CFAR (Constant False Alarm Rate) processing, Yoshida,'Revised Radar Technology', Institute of Electronics, Information and Communication Engineers, pp.87-89 (1996) DBSCAN (Density-based Spatial Clustering of Applications with Noise), Sebastian Raschka, Python,'Machine Learning Programming', Impress, pp.319-323 (2016) Neural network, Okaya,'Deep learning', Kodansha, pp.7-26 (2014) Activation function, Okaya,'Deep learning', Kodansha, pp.10-12 (2014) Loss function, Okaya,'Deep learning', Kodansha, pp.14-16 (2014) Convolution NN, Okaya,'Deep Learning', Kodansha, pp.79-91 (2014) Convolution NN (Upsampling), Fujita,'Implementation Deep Learning', Ohmsha, pp.40-45 (2016)

As described above, in the conventional radar device that performs CFAR processing after performing pulse compression or FFT processing, the reception amplitude is CFAR when detecting a small target with a small RCS (Radar Cross Section) and a low reception SN. There was a problem that it was not detected without exceeding the threshold of.

The present embodiment has been made in view of the above problems, and an object of the present invention is to provide a radar device capable of enhancing the detection capability of a small target having a small RCS and a low receiving SN, and a radar signal processing method thereof.

In order to solve the above problems, according to the present embodiment, in a coherent radar device that transmits N (N ≧ 1) hit pulses, the PRI axis is subjected to FFT processing, and if necessary, pulses are applied. Using the compressed range-doppler data (RD data), the range-doppler axis is divided into blocks (Nh, Mh) (Nh ≥ 1, Mh ≥ 1) to create an amplitude intensity histogram, and a one-dimensional vector of the histogram. Is input to the multi-layer NN for block detection, the block of the target detection candidate is extracted, the RD data of the block range is converted into a one-dimensional vector and input to the multi-layer NN for cell identification, and the range-Doppler axis is detected. Identify the cell.

That is, in the radar device according to the present embodiment, the transmission / reception signal of the Doppler range axis is used to extract a feature amount such as an amplitude histogram from the block-divided signal, and a block with or without a target is extracted by a multi-layer NN. For the data of the extracted block, the cell of the range-Doppler axis in the block can be specified by the multi-layer NN, and the distance and the speed can be output.

The block diagram which shows the schematic structure of the radar apparatus which concerns on 1st Embodiment. The figure for demonstrating block division of range-doppler data, vector creation of an amplitude histogram sequence, and utilization of a multi-layer neural network in the first embodiment. FIG. 5 shows a configuration in which each unit is connected using weighting and a bias coefficient in each layer of a multi-layer neural network in the first embodiment. The figure for demonstrating the technique of specifying the detection cell of the range-doppler axis in the detected target block in 1st Embodiment. The block diagram which shows the schematic structure of the radar apparatus which concerns on 2nd Embodiment. In the second embodiment, it is a figure which shows how the range-doppler data with pseudo-high resolution can be obtained by the range-doppler axis. The block diagram which shows the schematic structure of the radar apparatus which concerns on 3rd Embodiment. In the third embodiment, the block diagram which shows the processing flow of the convolutional neural network. The figure which shows the mode of acquiring the target distance and velocity by the convolutional neural network in the third embodiment. The block diagram which shows the schematic structure of the radar apparatus which concerns on 4th Embodiment. The conceptual diagram for demonstrating the processing of the DBSCAN method applied to 4th Embodiment. In the fourth embodiment, a conceptual diagram showing how to discriminate between a target and a false positive by cluster analysis.

Hereinafter, embodiments will be described with reference to the drawings. In the description of each embodiment, the same parts are designated by the same reference numerals, and duplicate description will be omitted.

(First Embodiment)
The radar device according to the first embodiment will be described with reference to FIGS. 1 to 4.

FIG. 1 is a block diagram showing a schematic configuration of a radar device according to the first embodiment. This radar device is roughly classified into a transmission system, a reception system, and a signal processing system.

In the transmission system, a transmission signal such as a chirp signal generated by the transmission signal generation unit 1 is converted into a high frequency signal by the frequency conversion unit 2, pulse-modulated by the pulse modulation unit 3, and a transmission / reception separation unit 4 such as a circulator is used. It is transmitted as a radar wave from the antenna unit 5 via the antenna unit 5.

In the receiving system, the reflected signal of the radar wave received by the antenna unit 5 is converted into an intermediate frequency by the frequency conversion unit 11 via the transmission / reception separation unit 4, and is converted into a digital signal by the AD conversion unit 12.

In the signal processing system, the received signal converted into a digital signal is subjected to pulse compression (Non-Patent Document 1) as necessary by the pulse compression unit 13, and the FFT (Fast Fourier Transform) processing unit 14 performs PRI (Pulse Repetition Interval). FFT processing of the Doppler axis (slow-time axis) between them is performed to obtain range-Doppler (RD) data. Then, the RD data is divided into predetermined blocks by the block division unit 16, and an amplitude histogram (specifically, a vector of the amplitude histogram sequence) is generated for each division unit by the histogram generation unit 17, and the multilayer NN (Neural Network, Non-Patent Document 4) The block extraction unit 18 identifies the presence or absence of a target for each divided block by a multilayer NN and extracts a block of a target detection candidate. Subsequently, the RD data of the block range of the extracted target detection candidate is converted into a one-dimensional vector by the block vectorization processing unit 19, and the multi-layer NN cell identification unit 20 multi-layers the detection cells of the range-Doppler axis from the one-dimensional vector. It is specified by NN, and the range / speed detection unit 22 detects the range and speed from the detection cell.

In the above configuration, the radar signal processing of the present embodiment will be described with reference to FIGS. 2 to 4. FIG. 2 is a diagram for explaining block division of RD data, histogram creation, and utilization of a multilayer NN, and FIG. 3 is a diagram showing a configuration in which each unit is combined using weighting and a bias coefficient in each layer of the multilayer NN. FIG. 4 is a diagram for explaining a method of specifying a detection cell of the range-Doppler axis in the target block detected by the multilayer NN.

When the digital reception signal obtained by the reception system is pulse-compressed by the pulse compression unit 13 as necessary, and the FFT 14 is subjected to FFT processing of the slow-time axis between PRIs to acquire RD data. , This RD data is input to the RD data block dividing unit 16.

As shown in FIG. 2A, the RD data block partitioning unit 16 covers the RD data RD (r, fd) (r: range, fd: Doppler frequency) as a region (1 to Nh, 1 to Mh) (Nh). ≧ 1, Mh ≧ 1 integer) Divide into blocks. Although FIG. 2A shows a case where the blocks do not overlap, the blocks may be slid and overlapped as a countermeasure when there is a target near the boundary between the blocks. .. The RD data of each divided block is input to the histogram generation unit 17. As shown in FIG. 2B, the histogram generation unit 17 generates an amplitude histogram in divided block units. The amplitude histogram created here is input to the multilayer NN block extraction unit 18 arranged in multiple layers as shown in FIG. 2 (c).

In the multilayer NN block extraction unit 18, each layer is connected by using weighting and a bias coefficient as shown in FIG. Expressed in a mathematical formula, it becomes the following formula.

An activation function (Non-Patent Document 5), which is a non-linear function, is applied to this output Zm (m = 1 to M) to obtain an output ym (m = 1 to M).

As described above, the RD data RD (r, fd) is divided into blocks of regions (1 to Nh, 1 to Mh), and the amplitude histogram generated for each block shows the amplitude intensity on the horizontal axis with a predetermined fixed value. When standardized, if a signal having a different amplitude with a larger intensity than that of other blocks exists in the block, its shape changes. Therefore, in the multi-layer NN block extraction unit 18, the presence or absence of the target is identified by the multi-layer NN by utilizing the change in the amplitude histogram shape. Since the multi-layer NN has a function of extracting a feature amount that makes it easier to determine the presence or absence of a target from the amplitude histogram shape, the presence or absence of a target is compared to the method of extracting a fixed feature amount from the amplitude histogram (Patent Document 1). It can be expected that the ability to discriminate between the two will improve.

In order to acquire the optimum parameters (weight and bias) at this time, the presence or absence of the target is learned as a teacher signal using the RD data with and without the target. That is, the amplitude histogram generated from the RD data for training is used as a column vector, and the loss function (Non-Patent Document 6) due to the difference between the final output and the identification output without the target and the identification output with the target is minimized. Determine the parameters. Then, using this parameter, inference processing (inference processing by multi-layer NN with learned parameters) is executed for each block on unknown input data, and a block having a target is extracted. As a result, the range of the RD data in which the target exists can be limited, so that the accuracy of cell identification described later is improved.

A method of identifying the detection cell of the range-Doppler axis within the detected target block will be described with reference to FIG.

First, the block vectorization processing unit 19 converts the amplitude data of the target detection block extracted by the multi-layer NN block extraction unit 18 into a column vector. The number of dimensions of the column vector is all Dopplas cells × all range cells. Further, as for the output, in order to specify the cell among all the cells, it is necessary to output the same number of units as the number of input units (total number of cells). In order to learn the parameters (weight and bias) in this case, a teacher signal is created by using the RD data with and without the target, with the cell where the target exists as 1 and the cell where the target does not exist as 0. Learn based on teacher signals. Therefore, in the multi-layer NN cell identification unit 20, all cells of the RD data for learning are input as a one-dimensional column vector, and a loss function due to the difference between the final output with or without the target of all cells and the learning output (Non-Patent Document 6). ) Is minimized, and the parameters of each layer are determined. Using this parameter, inference processing (inference processing by multi-layer NN with learned parameters) is executed on the RD data of the block in which the target exists, and the target cell is specified. The range / speed detection unit 22 calculates and outputs a target range and speed based on the range (distance) and Doppler frequency of the specified cell. The speed can be calculated from the Doppler frequency by the following formula.

As described above, in the radar device according to the first embodiment, in the coherent method of transmitting N (N ≧ 1) hit pulses, the PRI axis is subjected to FFT processing and pulse-compressed RD as necessary. Using the data, divide the blocks into (Nh, Mh) (Nh ≥ 1, Mh ≥ 1) on the range-Doppler axis, generate an amplitude intensity histogram, and convert the one-dimensional vector of the amplitude histogram into a multi-layer NN for block detection. The block of the target detection candidate is extracted by inputting, the RD data of the block range is converted into a one-dimensional vector and input to the multi-layer NN for cell identification, and the detection cell of the range-Doppler axis is specified. That is, in the radar device, the feature amount of the amplitude histogram is extracted from the block-divided signal by using the transmission / reception signal of the range-Doppler axis, and the block with or without the target is extracted by the multilayer NN. Further, with respect to the RD data of the extracted block, the distance and the velocity are calculated and output by specifying the cell of the range-Doppler axis in the block by the multi-layer NN.

Here, in the case of a radar device that detects a target using CFAR by conventional pulse compression or a Doppler filter, it is common to use the amplitude intensity of the target to increase the detection when the CFAR threshold is exceeded. rice field. In this case, even if there are multiple reflection points for each target, it cannot be detected when the amplitude intensity of the maximum point is low because it is a method of comparing and detecting the point with the maximum amplitude intensity and the threshold. There was a problem. On the other hand, in the present embodiment, since there are a plurality of reflection points for each target, it is possible to improve the detection probability when the amplitude intensity of the maximum point is low.

(Second embodiment)
The first embodiment is a method using the RD data as it is. In this case, if the block is subdivided, the number of cells is reduced, and it is assumed that the characteristics of the amplitude histogram cannot be extracted correctly. Therefore, in this embodiment, this countermeasure method will be described.

FIG. 5 is a block diagram showing a configuration of the radar device according to the second embodiment. The radar device of the present embodiment has almost the same configuration as that of the first embodiment, but the RD data subjected to signal processing such as pulse compression (13) and FFT (14) is sent to the pseudo high resolution processing unit 15. Input and perform pseudo-high resolution for at least one of the range axis (fast-time axis) and the Doppler axis (slow-time axis). For this purpose, the pseudo high resolution processing unit 15 performs the inverse Fourier transform (IFFT) shown in the following equation on the RD data.

Here, both the range axis and the Doppler axis are described as a method for increasing the pseudo high resolution, but in the case of increasing the pseudo high resolution of either axis, processing of only that axis may be sufficient.

For the RD data RDf that has been two-dimensionally transformed as described above, the range axis and Doppler axis are zero-padded, and the number of dimensions is Nz × Mz (Nz> N, Mz> M), which is defined as RDfz. .. Two-dimensional FFT is performed on this RD data RDfz.

As a result, as shown in FIG. 6A, the RD data including false positives is pseudo-high resolution on the range-Doppler axis, and the reflection point at the foot of the mountain of amplitude due to the target is also observed near the target. It can be expected that the number of reflection points will increase as shown in FIG. 6 (b). By detecting the target by the same processing as in the first embodiment using this RD data, it becomes easier to detect the feature amount of the amplitude histogram, and the target cell is also specified in the cell identification in the block in which the target exists. It can be expected that it will be easier to do.

As described above, in the radar device according to the second embodiment, at least one of the range axis (fast-time axis) and the Doppler axis (slow-time axis) is subjected to Fourier transform processing on the RD data, and each axis is processed. By applying the processing of the first embodiment to the RD data which has been subjected to FFT processing to increase the pseudo-high resolution after filling with zeros, the RD data is pseudo-high resolution and the number of cells is increased, and the amplitude histogram is used. Since it is easy to extract the feature amount such as, the block of the target candidate can be extracted, and the target cell can be specified.

(Third Embodiment)
In the second embodiment, a method of increasing the resolution of RD data by increasing the pseudo-high resolution and inputting the data into the multi-layer NN to perform block extraction and cell identification has been described. In this case, since the input signal is column-vectorized when the cell is specified, the information on the cell position on the range-Doppler axis is lost, and the feature of the position on the range-Doppler axis is not included. In this embodiment, a method of identifying a cell will be described by using the position information of the range-Doppler axis.

FIG. 7 is a block diagram showing a schematic configuration of a radar device according to a third embodiment. The radar device of the present embodiment includes a CNN (Convolutional Neural Network, Non-Patent Documents 7 and 8) cell identification unit 21 in place of the block vectorization processing unit 19 and the multilayer NN cell identification unit 20 shown in FIG. That is, the same processing as in the second embodiment is performed until the block with the target is extracted by the multi-layer NN, but after the block is extracted, the CNN cell identification unit 20 performs the convolution NN at the time of cell identification. Use.

As shown in FIG. 8, the CNN includes an input layer 31, a convolutional layer 32, a pooling layer 33, a convolutional layer 34, a pooling layer 35, a multi-stage upsampling layer 36, and a fully connected layer 37. A filter that retains the position information is applied, and the pooling layer 33 outputs the maximum value and the average value of cells in a predetermined range after the convolutional filter. As a result, it is possible to absorb errors such as slight misalignment of the input data. This is repeated as necessary to make it easier to extract the features. Since the number of cells of the input two-dimensional data is reduced by CNN, the output is the upsampling layer 36 (non-patented) after the number of convolutions and the pooling layer so that the number of cells is greater than the number of cells before the pseudo-high resolution is achieved. A multi-layer NN to which Document 8) and the like are added is constructed. The learning method is the same as that of the second embodiment.

As described above, in the radar device according to the present embodiment, as shown in FIG. 9, the target cell is used by the CNN using the feature amount including the positional relationship of the multiple reflection points of the individual targets on the range-Doppler axis. Therefore, the target can be extracted more efficiently than in the case of inputting the RD data to the multi-layer NN by column vectorization in the first embodiment. Thereby, the target range and speed can be output as in the second embodiment. Further, for the RD data, the range axis (fast-time axis) and the Doppler axis (slow-time axis) are IFFT-processed, each axis is filled with zeros, and then FFT processing is performed to increase the pseudo-high resolution of the RD. The data is divided into blocks (Nh, Mh) (Nh ≧ 1, Mh ≧ 1) on the range-Doppler axis, a histogram of amplitude intensity is generated, and the one-dimensional vector of the histogram is converted into a multi-layer NN for block detection. Input, extract the block of the target detection candidate, input the RD data of the block range to CNN, and specify the detection cell of the range-Doppler axis. That is, by increasing the number of cells by increasing the pseudo-high resolution of the RD data, it becomes easy to extract the feature amount such as the amplitude histogram, and thereby the block of the target candidate can be easily extracted. Further, the convolutional NN (CNN) can be used to use the feature amount considering the arrangement of the reflection points on the range-Doppler axis, and the target cell can be specified accurately.

(Fourth Embodiment)
In the first to third embodiments, a method of identifying a target detection cell in the extracted block after extracting the block by the multi-layer NN has been described. In this case, erroneous detection may occur depending on the degree of learning of the multilayer NN. In this embodiment, an example of the countermeasure will be described.

FIG. 10 is a block diagram showing a schematic configuration of a radar device according to a fourth embodiment. In the present embodiment, block detection by the multilayer NN and cell identification by the multilayer NN (first and second embodiments) and CNN (third embodiment) are the same as those of the first to third embodiments. However, after the cell is identified by the multi-layer NN cell identification unit 20, the cluster analysis unit 23 discriminates the cell by using the density difference between the target and the false detection on the range-Doppler axis.

As an example of the cluster analysis, there is a DBSCAN (Density-based Spatial Clustering of Applications with Noise, Non-Patent Document 3) method shown in FIG. In this method, the gate radius (distance from the core point to the outer circumference of the gate) ε and the number of border point observation values in the gate are specified to form a cluster that meets the conditions. In the vicinity of the range-doppler where the target exists, there are multiple reflection points depending on the target intensity, but the number of points is small in the false detection that appears randomly. Utilizing this fact, as shown in FIG. 12, it is possible to discriminate between the target and the false positive by analyzing the cluster with respect to the data of the range-Doppler axis after the cell is identified by the multi-layer NN. ..

As the input of this cluster analysis, the cells detected by the multi-layer NN are set to 1, and the other cells are set to 0, and all the two-dimensional range-doppler cell data are input. As the cluster analysis, a method other than the DBSAN method may be used. Further, in the present embodiment, the case of the radar that performs pulse compression has been described, but it can also be applied to the case of not performing pulse compression.

As described above, the radar device according to the present embodiment plots on two-dimensional coordinates (range-Doppler axis) using the detected cell data (observed value coordinates), and uses the two-dimensional plot coordinates as input to form a cluster. By performing the analysis, the target detection can be confirmed by the density difference. That is, after extracting the target candidate block, when the cells in the candidate block are identified, cluster analysis is performed to extract the cells, so that even if there is a false detection, the target and the false detection can be discriminated by the density difference. And the target cell can be identified.

The present invention is not limited to the above embodiment as it is, and at the implementation stage, the components can be modified and embodied within a range that does not deviate from the gist thereof. In addition, various inventions can be formed by an appropriate combination of the plurality of components disclosed in the above-described embodiment. For example, some components may be removed from all the components shown in the embodiments. In addition, components across different embodiments may be combined as appropriate.

1 ... Transmission signal generation unit, 2 ... Frequency conversion unit, 3 ... Pulse modulation unit, 4 ... Transmission / reception separation unit, 5 ... Antenna unit, 11 ... Frequency conversion unit, 12 ... AD conversion unit, 13 ... Pulse compression unit, 14 ... FFT (Fast Fourier Transform) processing unit, 15 ... pseudo high resolution processing unit, 16 ... RD data block division unit, 17 ... histogram generation unit, 18 ... multi-layer NN block extraction unit, 19 ... block vectorization processing unit, 20 ... Multi-layer NN cell identification unit, 21 ... CNN cell identification unit, 22 ... Range / speed detection unit, 23 ... Cluster analysis unit, 31 ... Input layer, 32 ... Convolution layer, 33 ... Pooling layer, 34 ... Convolution layer, 35 ... Pooling Layers, 36 ... upsampling layers, 37 ... fully connected layers.

Claims (7)

A generation means for generating range-Doppler data by converting the received signal in the pulse repetition period of the N (N ≧ 1) hit transmission pulse into the frequency domain of the Doppler axis.
A partitioning means for dividing the range-doppler data generated by the generation means into a plurality of blocks along the range-doppler axis, and
A generation means for generating a histogram of amplitude intensity for each of a plurality of blocks divided by the division means, and a generation means.
An extraction means that identifies the presence or absence of a target for each divided block by a multi-layer neural network from the histogram of each of the plurality of blocks generated by the generation means, and extracts the target detection candidate block from the identification result.
A vectorizing means for converting the range-Doppler data of the block range of the target detection candidate block extracted by the extraction means into a one-dimensional vector, and
A specific means for identifying the target detection cell of the range-Doppler axis by a multi-layer neural network from the one-dimensional vector obtained by the vectorization means, and
A radar device including a calculation means for calculating at least one of a range and a speed from a target detection cell specified by the specific means. A generation means for generating range-Doppler data by converting the received signal in the pulse repetition period of the N (N ≧ 1) hit transmission pulse into the frequency domain of the Doppler axis.
Pseudo-high resolution means for inverse Fourier transforming the range axis and Doppler axis on the range-Doppler data generated by the generation means, filling each axis with zeros, and then performing Fourier transform to increase the pseudo-high resolution. When,
A partitioning means for dividing the range-doppler data whose pseudo-high resolution has been increased by the pseudo-high resolution means into a plurality of blocks along the range-doppler axis.
A generation means for generating a histogram of amplitude intensity for each of a plurality of blocks divided by the division means, and a generation means.
An extraction means that identifies the presence or absence of a target for each divided block by a multi-layer neural network from the histogram of each of the plurality of blocks generated by the generation means, and extracts the target detection candidate block from the identification result.
A vectorizing means for converting the range-Doppler data of the block range of the target detection candidate block extracted by the extraction means into a one-dimensional vector, and
A specific means for identifying the target detection cell of the range-Doppler axis by a multi-layer neural network from the one-dimensional vector obtained by the vectorization means, and
A radar device including a calculation means for calculating at least one of a range and a speed from a target detection cell specified by the specific means. A generation means for generating range-Doppler data by converting the received signal in the pulse repetition period of the N (N ≧ 1) hit transmission pulse into the frequency domain of the Doppler axis.
Pseudo-high resolution means for inverse Fourier transforming the range axis and Doppler axis on the range-Doppler data generated by the generation means, filling each axis with zeros, and then performing Fourier transform to increase the pseudo-high resolution. When,
A partitioning means for dividing the range-doppler data whose pseudo-high resolution has been increased by the pseudo-high resolution means into a plurality of blocks along the range-doppler axis.
A generation means for generating a histogram of amplitude intensity for each of a plurality of blocks divided by the division means, and a generation means.
An extraction means that identifies the presence or absence of a target for each divided block by a multi-layer neural network from the histogram of each of the plurality of blocks generated by the generation means, and extracts the target detection candidate block from the identification result.
A specific means for identifying the target detection cell of the range-Doppler axis by convolving the range-Doppler data of the block range of the target detection candidate block extracted by the extraction means and using a neural network.
A radar device including a calculation means for calculating at least one of a range and a speed from a target detection cell specified by the specific means. Further, the observed value of the target detection cell specified by the specific means is plotted on the two-dimensional coordinates of the range-Doppler axis, and the coordinates of the observed value plotted on the two-dimensional coordinates are cluster-analyzed to analyze the density difference of the plot points. The radar device according to any one of claims 1 to 3, further comprising a cluster analysis means for determining the target detection based on the density difference and outputting the target detection to the calculation means. The range-Doppler data is generated by converting the received signal in the pulse repetition period of the N (N ≧ 1) hit transmission pulse into the frequency domain of the Doppler axis.
The range-doppler data is divided into a plurality of blocks along the range-doppler axis, and the data is divided into a plurality of blocks.
An amplitude intensity histogram was generated for each of the plurality of blocks.
From the histogram of each of the plurality of blocks, the presence or absence of a target is identified for each divided block by a multi-layer neural network, and the target detection candidate block is extracted from the identification result.
The range-Doppler data of the block range of the target detection candidate block is converted into a one-dimensional vector.
From the one-dimensional vector, the target detection cell of the range-Doppler axis is identified by the multi-layer neural network, and the target detection cell is identified.
A radar signal processing method for a radar device that calculates at least one of a range and a speed from the target detection cell. The range-Doppler data is generated by converting the received signal in the pulse repetition period of the N (N ≧ 1) hit transmission pulse into the frequency domain of the Doppler axis.
For the range-Doppler data, the range axis and the Doppler axis are inverse-Fourier-transformed, each axis is filled with zeros, and then Fourier-transformed to increase the pseudo-high resolution.
The pseudo-high resolution range-doppler data is divided into a plurality of blocks along the range-doppler axis.
A generation means for generating an amplitude intensity histogram for each of the plurality of blocks,
From the histogram of each of the plurality of blocks, the presence or absence of a target is identified for each divided block by a multi-layer neural network, and the target detection candidate block is extracted from the identification result.
The range-Doppler data of the block range of the target detection candidate block is converted into a one-dimensional vector.
From the one-dimensional vector, the target detection cell of the range-Doppler axis is identified by the multi-layer neural network, and the target detection cell is identified.
A radar signal processing method for a radar device that calculates at least one of a range and a speed from the target detection cell. The range-Doppler data is generated by converting the received signal in the pulse repetition period of the N (N ≧ 1) hit transmission pulse into the frequency domain of the Doppler axis.
For the range-Doppler data, the range axis and the Doppler axis are inverse-Fourier-transformed, each axis is filled with zeros, and then Fourier-transformed to increase the pseudo-high resolution.
The pseudo-high resolution range-doppler data is divided into a plurality of blocks along the range-doppler axis.
An amplitude intensity histogram was generated for each of the plurality of blocks.
From the histogram of each of the plurality of blocks, the presence or absence of a target is identified for each divided block by a multi-layer neural network, and the target detection candidate block is extracted from the identification result.
The target detection cell of the range-Doppler axis is identified by convolving the range-Doppler data of the block range of the target detection candidate block by a neural network.
A radar signal processing method for a radar device that calculates at least one of a range and a speed from the target detection cell.

Images