JP6567232B2 - Image reproducing device, signal processing apparatus, and image reproducing method - Google Patents

Description

  The present invention relates to an image regenerator, a signal processing device, and an image reproduction method for reproducing a synthetic aperture radar image from observation data obtained by a synthetic aperture radar (SAR).

The SAR repeatedly radiates radio waves toward an observation target along an azimuth direction in which an antenna platform such as a satellite or an aircraft travels, and receives a reflected signal that is a radio wave reflected by the observation target.
The SAR is represented by a complex matrix of M rows and N columns as observation data regarding the range direction and the azimuth direction representing the distance to the observation object, where M is the number of reflected signal samplings and N is the number of reflected waves. Observation data can be obtained.

A signal processing device that reproduces a SAR image from observation data obtained by SAR executes, for example, the following image reproduction algorithm.
The image reproduction algorithm is, for example, an algorithm for accumulating the luminance value of each pixel in the SAR image data by combining signal processing as described below.
(1) A process of converting a time domain signal into a frequency domain signal by Fourier transforming the time domain signal as observation data. (2) A frequency domain signal by performing an inverse Fourier transform on the frequency domain signal. (3) Processing for reproducing a SAR image by performing signal processing on a time domain signal or a frequency domain signal

When the processing of (1) to (3) described above is included in the image reproduction algorithm, the signal in the time domain, which is observation data represented by a complex matrix of M rows and N columns, has a row or column. As a unit, the signal is converted into a frequency domain signal by Fourier transform.
Further, the frequency domain signal is converted into a time domain signal by inverse Fourier transform in units of rows or columns.
Matrix elements in the frequency domain signal after Fourier transform are calculated independently of each other. Further, matrix elements in the time domain signal after inverse Fourier transform are calculated independently of each other.
Therefore, the image reproduction algorithm includes M-point Fourier transform for each column, or signal processing after repeating M points inverse Fourier transform for each column for N columns, and N-point Fourier transform for each row. Or it becomes the structure which combined the procedure of performing signal processing, after repeating the inverse Fourier transform of N points | pieces for every line by M line.
In this image reproduction algorithm, the processing direction of the Fourier transform and the processing direction of the inverse Fourier transform are switched to the row direction or the column direction.

When the signal processing apparatus for reproducing the SAR image has the same configuration as that of a general computer, the input / output performance for the memory is lower than the computing performance of the computer. Cost.
Specifically, the signal processing apparatus stores signal data in a large capacity and high delay main memory, and transfers the signal data stored in the main memory to a small capacity and low delay local memory (1). I do.
Also, the signal processing device performs arithmetic processing on the signal data stored in the local memory, and transfers the processing result of the arithmetic processing or intermediate data in the middle of the arithmetic processing that does not fit in the local memory to the main memory. Process (2) is performed. The signal processing device repeatedly performs the processing (1) and the processing (2).

Therefore, in the signal processing apparatus, signal data waiting for processing is generated in the input / output processing for the memory, so that the computation performance possessed by the signal processing device may not be fully utilized.
For this reason, in the signal processing device, it is desirable to minimize the number of times signal data is transferred to the local memory, and to maximize the amount of calculation for each input / output process for the memory.
By maximizing the amount of calculation for each input / output process for the memory, the calculation performance possessed can be fully utilized, and the image reproduction process can be completed in a short time.

Further, since the main memory of the signal processing apparatus generally employs one-dimensional data management, two-dimensional data such as SAR observation data is arranged as one-dimensional data on the main memory. . For this reason, either one of the rows or columns of the observation data is arranged in a continuous local region in the main memory, while the other is arranged in a position separated by the row size or the column size. .
Since a general local memory performs input / output processing with the main memory in units of local areas that are continuous from the referenced data position, processing in the discontinuous direction of the main memory is slow.

For this reason, in the case of M rows and N columns of signal data, when M points are stored in the main memory in the order in which N consecutive M points are arranged (Column-Major), the processing in the M direction becomes faster, but the N direction Processing is slow.
Conversely, when N consecutive points are stored in the main memory in the order in which M sets are arranged (Row-Major), processing in the N direction is fast, but processing in the M direction is slow.
Therefore, when performing a process that requires multiple input / output processes for the memory, as a pre-process for shifting to the process, by performing a replacement process that replaces the memory layout of the signal data by matrix transposition, the signal data is always a continuous area. So that high-speed computation is possible.

In the SAR image reproduction process, since the process for each direction in units of columns or rows is repeated, it is necessary to frequently use a replacement process for replacing the memory layout of the observation data.
Also, since there are many types of signal processing and huge observation data with a data amount of several gigabytes (GB) to several tens of GB is handled, it takes a long time to complete image reproduction.
For this reason, a signal processing apparatus that divides observation data into a plurality of small regions and in which a plurality of processors perform a process of reproducing a SAR image from the divided observation data of the small regions is disclosed in Patent Document 1 below. Yes.
Patent Document 1 below discloses a method for improving a communication bottleneck when providing observation data of a small area to a plurality of processors connected by communication.

JP 2014-202670 A

  Since the conventional signal processing apparatus is configured as described above, it is possible to improve a communication bottleneck when providing observation data of a small area to a plurality of processors connected by communication. However, in order to improve the efficiency of the Fourier transform process and the inverse Fourier transform process by a plurality of processors, the signal in the time domain and the frequency domain can be implemented so that the multidimensional Fourier transform process and the multidimensional inverse Fourier transform process can be performed. It does not set the memory layout of the signals. For this reason, the multidimensional Fourier transform process and the multidimensional inverse Fourier transform process cannot be performed, and there is a problem that it may take a long time to complete the reproduction of the SAR image from the observation data. .

  The present invention has been made to solve the above-described problems, and provides an image regenerator, a signal processing device, and an image reproduction method capable of reducing the time required to complete the reproduction of the SAR image from the observation data. The purpose is to obtain.

  An image regenerator according to the present invention includes a data dividing unit that divides observation data obtained by a synthetic aperture radar into a plurality of local region signals, and data division according to a processing direction of Fourier transform processing or a processing direction of inverse Fourier transform processing. A layout setting unit for setting a memory layout of signals of a plurality of local regions divided by the unit, and a multi-dimensional Fourier transform of the signals of the plurality of local regions in which the memory layout is matched to the processing direction of the Fourier transform processing by the layout setting unit The Fourier transform unit, the inverse Fourier transform unit that performs multi-dimensional inverse Fourier transform on the signals of a plurality of local regions whose memory layout is matched to the processing direction of the inverse Fourier transform process by the layout setting unit, and multi-dimensional Fourier transform by the Fourier transform unit Multi-dimensional inverse Fourier transform by local region signal or inverse Fourier transform unit It is obtained as a signal processing unit for reproducing the SAR image from the signals of the local region.

  According to this invention, the layout setting unit that sets the memory layout of the signals of the plurality of local regions divided by the data dividing unit in accordance with the processing direction of the Fourier transform process or the process direction of the inverse Fourier transform process is provided. The transform unit performs multi-dimensional Fourier transform on the signals of a plurality of local regions whose memory layout is aligned with the processing direction of the Fourier transform process by the layout setting unit, and the inverse Fourier transform unit performs the inverse Fourier transform process on the memory layout by the layout setting unit. Since the multi-dimensional inverse Fourier transform is performed on the signals of a plurality of local regions matched to the processing direction, it is possible to reduce the time required to complete the reproduction of the synthetic aperture radar image from the observation data. .

BRIEF DESCRIPTION OF THE DRAWINGS It is a block diagram which shows the signal processing apparatus by Embodiment 1 of this invention. It is a hardware block diagram which shows the signal processing apparatus by Embodiment 1 of this invention. It is a hardware block diagram of a computer in case a signal processing apparatus is implement | achieved by software or firmware. It is a flowchart which shows the image reproduction method which is a process sequence in case a signal processing apparatus is implement | achieved by software or firmware. It is explanatory drawing which shows an example of the observation data by SAR. It is explanatory drawing which shows the example which is dividing P into the signal data 101, when implementing N point 1-dimensional Fourier-transform of Az direction. It is explanatory drawing which shows the example which is dividing Q into the signal data 101, when implementing M point 1-dimensional Fourier transformation of the Rg direction.

  Hereinafter, in order to explain the present invention in more detail, modes for carrying out the present invention will be described with reference to the accompanying drawings.

Embodiment 1 FIG.
1 is a block diagram showing a signal processing apparatus according to Embodiment 1 of the present invention. FIG. 2 is a hardware configuration diagram showing the signal processing apparatus according to Embodiment 1 of the present invention.
1 and 2, the data management unit 1 is realized by, for example, the data management circuit 21 shown in FIG.
The data management unit 1 stores a time-domain signal, which is observation data obtained by a synthetic aperture radar (SAR), in the main memory 2 and manages the time-domain signal.
The main memory 2 is realized by a main recording circuit 28 shown in FIG. 2, for example.
The main memory 2 is a recording medium for storing a time domain signal which is observation data by SAR.
The main memory 2 stores a signal being reproduced by the image reproducers 3a and 3b.

The image reproducers 3a and 3b perform processing in parallel with each other.
The image regenerators 3 a and 3 b perform a process of reproducing the SAR image from the time domain signal managed by the data management unit 1.
In the first embodiment, an example in which the signal processing device is mounted with two image regenerators 3a and 3b will be described, but it is sufficient that one or more image regenerators are mounted. It is not restricted to the example which mounts the containers 3a and 3b.
The data dividing units 11a and 11b are realized by, for example, the data dividing circuit 22 shown in FIG.
The data division units 11a and 11b divide the time domain signal stored in the main memory 2 into a plurality of local area signals, and store the divided local area signals in the local memories 12a and 12b. .
The local memories 12a and 12b are realized by, for example, a local recording circuit 29 shown in FIG.
The local memories 12a and 12b are recording media that store the local areas divided by the data dividing units 11a and 11b.

The layout setting units 13a and 13b are realized by, for example, the layout setting circuit 23 shown in FIG.
The layout setting units 13a and 13b are divided by the data dividing units 11a and 11b according to the processing direction of the Fourier transform processing by the Fourier transform units 14a and 14b or the processing direction of the inverse Fourier transform processing by the inverse Fourier transform units 15a and 15b. In addition, a process for setting a memory layout of signals in a plurality of local areas is performed.
In addition, when the processing direction of the Fourier transform process or the processing direction of the inverse Fourier transform process is changed, the layout setting units 13a and 13b can set the local memories 12a and 12b before setting the memory layout of the signals in the plurality of local regions. The process of transferring the signal being reproduced stored in step 1 to the main memory 2 is performed.
The memory layout setting process by the layout setting units 13a and 13b is performed on the local memories 12a and 12b.

The Fourier transform units 14a and 14b are realized by, for example, a Fourier transform circuit 24 shown in FIG.
The Fourier transform units 14a and 14b perform a multi-dimensional Fourier transform process on the signals of a plurality of local regions in which the memory layout is matched with the processing direction of the Fourier transform process by the layout setting units 13a and 13b. The Fourier transform processing by the Fourier transform units 14a and 14b is performed on the local memories 12a and 12b.
The inverse Fourier transform units 15a and 15b are realized by, for example, an inverse Fourier transform circuit 25 shown in FIG.
The inverse Fourier transform units 15a and 15b perform a multidimensional inverse Fourier transform on the signals of a plurality of local regions whose memory layout is matched with the processing direction of the inverse Fourier transform process by the layout setting units 13a and 13b. The inverse Fourier transform process by the inverse Fourier transform units 15a and 15b is performed on the local memories 12a and 12b.

The signal processing units 16a and 16b are realized by a signal processing circuit 26 shown in FIG. 2, for example.
The signal processing units 16a and 16b perform signal processing on the local region signals subjected to multidimensional Fourier transform by the Fourier transform units 14a and 14b or the local region signals subjected to multidimensional inverse Fourier transform by the inverse Fourier transform units 15a and 15b. Then, the process of reproducing the synthetic aperture radar image is performed. Signal processing by the signal processing units 16a and 16b is performed on the local memories 12a and 12b.

The image management unit 4 is realized by, for example, the image management circuit 27 shown in FIG.
The image management unit 4 stores the synthetic aperture radar image reproduced by the image regenerators 3a and 3b in the main memory 5, and manages the synthetic aperture radar image.
The main memory 5 is realized by a main recording circuit 28 shown in FIG. 2, for example.
The main memory 5 is a recording medium that stores a synthetic aperture radar image.
FIG. 1 shows an example in which the main memory 2 and the main memory 5 are separate memories, the main memory 2 is built in the data management unit 1, and the main memory 5 is built in the image management unit 4. ing. However, this is only an example. For example, the main memory 2 and the main memory 5 are the same memory, inside the data management unit 1 or the image management unit 4, or outside the data management unit 1 and the image management unit 4. It may be provided.

  In FIG. 1, the data management unit 1, main memories 2 and 5, data division units 11a and 11b, local memories 12a and 12b, layout setting units 13a and 13b, Fourier transform units 14a and 14b, which are components of the signal processing device, Each of the inverse Fourier transform units 15a and 15b, the signal processing units 16a and 16b, and the image management unit 4 is assumed to be realized by dedicated hardware as shown in FIG. That is, the data management circuit 21, the main recording circuit 28, the data division circuit 22, the local recording circuit 29, the layout setting circuit 23, the Fourier transform circuit 24, the inverse Fourier transform circuit 25, the signal processing circuit 26, and the image management circuit 27 are realized. Is assumed.

Here, the main recording circuit 28 and the local recording circuit 29 are, for example, a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable Read Only Memory), an EEPROM (Electrically Memory). Non-volatile or volatile semiconductor memory such as), magnetic disks, flexible disks, optical disks, compact disks, mini disks, DVDs (Digital Versatile Discs), and the like.
The data management circuit 21, the data division circuit 22, the layout setting circuit 23, the Fourier transform circuit 24, the inverse Fourier transform circuit 25, the signal processing circuit 26, and the image management circuit 27 are, for example, a single circuit, a composite circuit, or a program And a processor programmed in parallel, an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination thereof.

The components of the signal processing device are not limited to those realized by dedicated hardware, and the signal processing device may be realized by software, firmware, or a combination of software and firmware.
Software or firmware is stored as a program in the memory of a computer. The computer means hardware that executes a program, and includes, for example, a CPU (Central Processing Unit), a central processing unit, a processing unit, an arithmetic unit, a microprocessor, a microcomputer, a processor, a DSP (Digital Signal Processor), and the like. .
FIG. 3 is a hardware configuration diagram of a computer when the signal processing apparatus is realized by software or firmware.

  When the signal processing device is realized by software or firmware, the main memories 2 and 5 and the local memories 12a and 12b are configured on the memory 31 of the computer, the data management unit 1, the data division units 11a and 11b, and the layout setting. A program for causing the computer to execute the processing procedures of the units 13a and 13b, the Fourier transform units 14a and 14b, the inverse Fourier transform units 15a and 15b, the signal processing units 16a and 16b, and the image management unit 4 is stored in the memory 31. The processor 32 may execute the program stored in the memory 31.

FIG. 4 is a flowchart showing an image reproduction method which is a processing procedure when the signal processing apparatus is realized by software or firmware.
2 shows an example in which each component of the signal processing device is realized by dedicated hardware, and FIG. 3 shows an example in which the signal processing device is realized by software, firmware, etc. Some components in the signal processing device may be realized by dedicated hardware, and the remaining components may be realized by software, firmware, or the like.

Observation data by SAR is input to the data management unit 1 and stored in the main memory 2.
FIG. 5 is an explanatory diagram showing an example of observation data by SAR.
In FIG. 5, signal data 101 which is observation data by SAR is represented by a complex matrix of M rows and N columns, a column is a signal in the range (Rg: Range) direction, and a row is a signal in the azimuth (Az: Azimuth) direction. is there.
Here, for convenience of explanation, the main memory 2 has a layout in which the Rg direction of the signal data 101 is continuous (Column-Major), but even in a layout in which the Az direction is continuous (Row-Major). I do not care.

また、係数データ１０３におけるｎ列目の値ｆ_ｎは、信号データ１０１におけるｍ行ｎ列目の信号成分Ｓ_ｍ，ｎに対して、以下の式（２）のように乗算される。ｎ列目の値ｆ_ｎを乗算する信号処理は、信号データ１０１の全ての行に対して繰り返される。

Each signal processing performed in the time domain or the frequency domain includes one coefficient data 102 which is a real or complex matrix of M rows and 1 column and one coefficient data 103 which is a real or complex matrix of 1 row and N columns. Used above.
The value f _m in the m-th row in the coefficient data 102 is multiplied by the signal component S _{m, n} in the m-th row and the n-th column in the signal data 101 as shown in the following equation (1). signal processing for multiplying the m-th row of values f _m is repeated for all columns of signal data 101.

Further, the value f _n of the n-th column in the coefficient data 103 is multiplied by the signal component S _{m, n} in the m-th row and the n-th column in the signal data 101 as shown in the following equation (2). The signal processing for multiplying the value f _n in the n-th column is repeated for all the rows of the signal data 101.

For this reason, since the signal processing for the signal data 101 is an independent calculation for each signal component, the calculation can be performed by dividing the signal data 101 into local regions of arbitrary shapes.
When converting a time-domain signal into a frequency-domain signal in at least one of the Az direction and the Rg direction, the signal component is referenced with reference to all signal components in one column or one row for each direction. It is necessary to carry out a Fourier transform process for updating.
In addition, when converting a frequency domain signal into a time domain signal in at least one of the Az direction and the Rg direction, refer to the signal components in one column or one row for each direction. Therefore, it is necessary to perform an inverse Fourier transform process for updating the signal component.

When performing Fourier transform processing or inverse Fourier transform processing in the Rg direction, the signal data 101 is divided in the Az direction that can be processed independently of each other, and the signal data of the divided region is Fourier transformed in the Rg direction for each divided region. Processing or inverse Fourier transform processing. As an example of the division of the signal data 101 in the Az direction, an example in which the signal data 101 is divided into four by dividing lines 104 to 106 indicating the divided portions can be considered.
Further, when performing Fourier transform processing or inverse Fourier transform processing in the Az direction, the signal data 101 is divided in Rg directions that can be processed independently of each other, and the signal data of the divided regions is divided in the Az direction for each divided region. Fourier transform processing or inverse Fourier transform processing is performed. As an example of the division of the signal data 101 in the Rg direction, an example in which the signal data 101 is divided into three by the dividing lines 107 to 108 indicating the divided portions can be considered.
When the processing direction of the Fourier transform process or the process direction of the inverse Fourier transform process is changed, if the processing result of the Fourier transform process or the process result of the inverse Fourier transform process in the previous Az direction or Rg direction is not complete, the Fourier transform Processing or inverse Fourier transform processing cannot be started.
For this reason, it is necessary to store the intermediate calculation result that is the processing result of the Fourier transform process or the inverse Fourier transform process in the preceding Az direction or Rg direction, and perform the calculation after re-dividing the signal data 101.

For example, if the one-dimensional Fourier transform in one direction is the one-dimensional Fourier transform of the W × H point, the two-dimensional Fourier transform that produces the same computation result with the same amount of computation as the one-dimensional Fourier transform of the W × H point. Shall be handled.
As a two-dimensional Fourier transform having the same calculation amount and the same calculation result as the one-dimensional Fourier transform of the W × H point, a two-dimensional Fourier transform composed of the one-dimensional Fourier transform of the W point and the one-dimensional Fourier transform of the H point. Conceivable.
Here, an example in which the one-dimensional Fourier transform is handled as a two-dimensional Fourier transform will be described, but it may be handled as a three-dimensional or more multidimensional Fourier transform.

FIG. 6 is an explanatory diagram illustrating an example in which the signal data 101 is P-divided when performing the N-point one-dimensional Fourier transform in the Az direction.
By dividing the signal data 101 into P, the one-dimensional Fourier transform of the N point in the Az direction is the one-dimensional Fourier transform of the P point in the Az (2) direction and 1 of the N / P point in the Az (1) direction. It becomes a two-dimensional Fourier transform consisting of a dimensional Fourier transform.
Each of the P point and the N / P point is set so as to be within a local region size that is a size capable of completing the processing by one two-dimensional Fourier transform.
In FIG. 6, signal data 201 to 204 are signal data obtained by dividing the signal data 101 into four in the Az direction.
The signal data 205 is signal data in a local region in which the four signal data 201 to 204 are superimposed, and is signal data having a size that can be processed by one two-dimensional Fourier transform.

FIG. 7 is an explanatory diagram showing an example in which the signal data 101 is Q-divided when the M-point one-dimensional Fourier transform in the Rg direction is performed.
By dividing the signal data 101 into Q, the one-dimensional Fourier transform of the M point in the Rg direction is performed by one-dimensional Fourier transform of the Q point in the Rg (2) direction and one of the M / Q points in the Rg (1) direction. It becomes a two-dimensional Fourier transform consisting of a dimensional Fourier transform.
Each of the Q point and the M / Q point is set so as to be within a local region size that is a size capable of completing the processing by one two-dimensional Fourier transform.
In FIG. 7, signal data 301 to 303 are signal data obtained by dividing the signal data 101 into three in the Rg direction.
The signal data 304 is signal data in a local region in which the three signal data 301 to 303 are superimposed, and is signal data having a size that can be processed by a single two-dimensional Fourier transform.

In the first embodiment, out of the two-dimensional Fourier transforms for each direction, the two-dimensional Fourier transform in the Az (2) direction, which is the Fourier transform related to the division number, and the Fourier in the Rg (2) direction, which is the Fourier transform related to the division number. Perform the conversion first.
Next, the remaining M / Q points in the Rg (1) direction with respect to the local region of (M / Q) × (N / P) points obtained by dividing the Az direction by P and dividing the Rg direction by Q. And the one-dimensional Fourier transform of the remaining N / P points in the Rg (1) direction.
Thereafter, signal processing of (M / Q) × (N / P) points, one-dimensional inverse Fourier transform of M / Q points in the Rg (1) direction, and one-dimensional N / P points in the Az (1) direction Inverse Fourier transform is performed on the local memory.
The memory layout is set on the local memory so that the memory layout is suitable for the next processing direction, and then the signal data is stored in the main memory.
Here, for convenience of explanation, signal processing is performed after performing Fourier transform in both the Ra direction and the Az direction. However, either or both of the Ra direction and the Az direction are processed in the time domain. In the case of signal processing, only signal data division processing may be performed and the corresponding Fourier transform may not be performed.

Next, the operation will be described.
The data management unit 1 stores signal data 101 that is observation data by SAR in the main memory 2.
The image reproducers 3a and 3b reproduce SAR images from the signal data 101 managed by the data management unit 1.
SAR image reproduction processing by the image reproducer 3a and the image reproducer 3b is performed in parallel.
The image management unit 4 stores the synthetic aperture radar image reproduced by the image regenerators 3a and 3b in the main memory 5, and manages the synthetic aperture radar image.
Hereinafter, the reproduction process of the SAR image by the image reproducers 3a and 3b will be specifically described.
However, since the SAR image reproduction process by the image reproducer 3a is the same as the SAR image reproduction process by the image reproducer 3b, the SAR image reproduction process by the image reproducer 3a will be described below with reference to FIG. An example will be described.

As shown in FIG. 6, the data dividing unit 11a of the image regenerator 3a performs P division on the signal data 101 stored in the main memory 2 in the Az direction.
The layout setting unit 13a performs, as a memory layout replacement process for continuing the signal data divided by the data dividing unit 11a in the Az direction, Az (2) for each signal data divided by the data dividing unit 11a in the Az direction. A process of overlapping in the direction is performed (step ST1 in FIG. 4).
In the example of FIG. 6, four signal data 201 to 204 are overlaid in the Az (2) direction.

The layout setting unit 13a selects P points having a size of M × (N / P) as signal data of a local region in the Az (2) direction from among a plurality of signal data superimposed in the Az (2) direction. Each of the signal data 205 is selected, and each of the signal data 205 in the local region of the selected P point is stored in the local memory 12a (step ST2 in FIG. 4).
The signal data of the local region of the selected P point is signal data having a size that can be processed by one two-dimensional Fourier transform.

The Fourier transform unit 14a performs two-dimensional Fourier transform in the Az (2) direction on each of the signal data in the local area of the point P stored in the local memory 12a (step ST3 in FIG. 4).
In preparation for the next Fourier transform processing in the Rg (2) direction, the layout setting unit 13a stores the memory so that the signal data in the local region of the P point respectively converted by the Fourier transform unit 14a is continuous in the Rg (2) direction. A layout replacement process is performed, and the signal data of the local region at the point P after the replacement process is stored in the main memory 2 (step ST4 in FIG. 4).

Next, the data dividing unit 11a has a local area size of M × (as local area signal data in the Az (1) direction from the local area signal data after replacement processing stored in the main memory 2. (N / P) is selected for each point P signal data (step ST5 in FIG. 4), and each of the selected point P signal data is Q-divided in the Rg direction.
The layout setting unit 13a uses the Q point signal whose local region size is Q × (N / P) as signal data of the local region in the Rg (2) direction from the signal data divided by the data dividing unit 11a. Each data is selected, and each local area of the selected Q point is stored in the local memory 12a (step ST6 in FIG. 4).

The Fourier transform unit 14a performs two-dimensional Fourier transform in the Rg (2) direction on each of the signal data in the local area of the Q point stored in the local memory 12a (step ST7 in FIG. 4). That is, the Fourier transform unit 14a performs N / P times Q-point Fourier transform.
In preparation for the next Fourier transform process in the Rg (1) direction, the layout setting unit 13a replaces the memory layout so that the signal data of the Q points respectively converted by the Fourier transform unit 14a are continuous in the Rg (1) direction. The processing is performed, and the signal data of the local region at the point Q after the replacement processing is stored in the main memory 2 (step ST8 in FIG. 4).

Next, the layout setting unit 13a uses the local region size (M / M) as the local region signal data in the Rg (1) direction from the local region signal data after replacement processing stored in the main memory 2. Each of the Q point signal data of Q) × (N / P) is selected, and each of the selected Q point signal data is stored in the local memory 12a (step ST9 in FIG. 4).
The Fourier transform unit 14a performs two-dimensional Fourier transform in the Rg (1) direction on each of the signal data in the local area of the Q point stored in the local memory 12a (step ST10 in FIG. 4). That is, the Fourier transform unit 14a performs M / Q point Fourier transform in the Rg (1) direction.

The layout setting unit 13a prepares for the next Fourier transform processing in the Az (1) direction, and arranges the memory layout so that the M / Q point signal data respectively converted by the Fourier transform unit 14a are continuous in the Az (1) direction. The M / Q point signal data whose local area size after the replacement process is (M / Q) × (N / P) is stored in the main memory 2 (step ST11 in FIG. 4).
Next, the layout setting unit 13a uses the local area size (M / M) as the local area signal data in the Az (1) direction from the local area signal data after replacement processing stored in the main memory 2. Each of the N / P point signal data of Q) × (N / P) is selected, and each of the selected N / P point signal data is stored in the local memory 12a.

The Fourier transform unit 14a performs two-dimensional Fourier transform in the Az (1) direction on each of the N / P point local region signal data stored in the local memory 12a (step ST12 in FIG. 4). That is, the Fourier transform unit 14a performs N / P point Fourier transform in the Az (1) direction.
The signal processing unit 16a is a frequency domain signal that is signal data that is two-dimensionally Fourier-transformed in the Az (2) direction, Rg (2) direction, Rg (1) direction, and Az (1) direction by the Fourier transform unit 14a. The signal processing A is performed on the signal to reproduce the synthetic aperture radar image (step ST13 in FIG. 4).

In preparation for the inverse Fourier transform process in the Rg (1) direction, the layout setting unit 13a performs a memory layout replacement process so that the signal data at the M / Q points are continuous in the Rg (1) direction. Are stored in the main memory 2 (step ST14 in FIG. 4).
Next, the layout setting unit 13a uses the M / Q point signal data as the local area signal data in the Rg (1) direction from the local area signal data after replacement processing stored in the main memory 2. Each of the signal data of the selected M / Q points is stored in the local memory 12a.

The inverse Fourier transform unit 15a performs two-dimensional inverse Fourier transform in the Rg (1) direction on each of the signal data in the local area of the M / Q point stored in the local memory 12a (step ST15 in FIG. 4). That is, the inverse Fourier transform unit 15a performs M / Q point inverse Fourier transform in the Rg (1) direction.
In preparation for the inverse Fourier transform process in the Rg (2) direction, the layout setting unit 13a performs the memory layout replacement process so that the signal data at the M / Q points are continuous in the Rg (2) direction. Are stored in the main memory 2 (step ST16 in FIG. 4).

Next, the layout setting unit 13a uses the M / Q point signal data as the local area signal data in the Rg (2) direction from the local area signal data after replacement processing stored in the main memory 2. And each of the selected M / Q point signal data is stored in the local memory 12a (step ST17 in FIG. 4).
The inverse Fourier transform unit 15a performs two-dimensional inverse Fourier transform in the Rg (2) direction on each of the signal data in the local area of the M / Q point stored in the local memory 12a (step ST18 in FIG. 4). That is, the inverse Fourier transform unit 15a performs a Q-point inverse Fourier transform in the Rg (2) direction.
The signal processing unit 16a performs signal processing B on the time domain signal, which is signal data subjected to inverse Fourier transform in the Rg (1) direction and the Rg (2) direction by the inverse Fourier transform unit 15a, and performs synthetic aperture radar. The image is reproduced (step ST19 in FIG. 4).

The Fourier transform unit 14a performs two-dimensional Fourier transform in the Rg (2) direction on each of the local region signal data subjected to the inverse Fourier transform by the inverse Fourier transform unit 15a (step ST20 in FIG. 4). That is, the Fourier transform unit 14a performs a Q-point Fourier transform in the Rg (2) direction.
In preparation for the inverse Fourier transform process in the Rg (1) direction, the layout setting unit 13a performs a memory layout replacement process so that the signal data at the M / Q points are continuous in the Rg (1) direction. Are stored in the main memory 2 (step ST21 in FIG. 4).

Next, the layout setting unit 13a sets the local area size as (N / N) as the local area signal data in the Rg (1) direction from the local area signal data after replacement processing stored in the main memory 2. Each of the Q point signal data of (P) × (M / Q) is selected, and each of the selected Q point signal data is stored in the local memory 12a (step ST22 in FIG. 4).
The Fourier transform unit 14a performs two-dimensional Fourier transform in the Rg (1) direction on each of the signal data in the local area of the Q point stored in the local memory 12a (step ST23 in FIG. 4). That is, the Fourier transform unit 14a performs M / Q point Fourier transform in the Rg (1) direction.
The signal processing unit 16a performs signal processing C on the signal in the frequency domain, which is signal data two-dimensionally Fourier-transformed in the Rg (2) direction and the Rg (1) direction by the Fourier transform unit 14a. The image is reproduced (step ST24 in FIG. 4).

The inverse Fourier transform unit 15a performs two-dimensional inverse Fourier transform in the Rg (1) direction on each of the local region signal data in the Rg (1) direction Fourier-transformed by the Fourier transform unit 14a (step ST25 in FIG. 4). That is, the inverse Fourier transform unit 15a performs M / Q point inverse Fourier transform in the Rg (1) direction.
In preparation for the inverse Fourier transform process in the Az (1) direction, the layout setting unit 13a performs a memory layout replacement process so that the N / P point signal data is continuous in the Az (1) direction. Is stored in the main memory 2 (step ST26 in FIG. 4).

Next, the layout setting unit 13a uses the N / P point signal data as the local area signal data in the Az (1) direction from the local area signal data after the replacement process stored in the main memory 2. And the signal data of the selected N / P points are stored in the local memory 12a.
The inverse Fourier transform unit 15a performs two-dimensional inverse Fourier transform in the Az (1) direction on each of the signal data in the local area of the N / P point stored in the local memory 12a (step ST27 in FIG. 4). That is, the inverse Fourier transform unit 15a performs an N / P point inverse Fourier transform in the Az (1) direction.
In preparation for the inverse Fourier transform process in the Az (2) direction, the layout setting unit 13a performs the memory layout replacement process so that the signal data at the Q point is continuous in the Az (2) direction. The point signal data is stored in the main memory 2 (step ST28 in FIG. 4).

Next, the layout setting unit 13a selects each of the Q-point signal data as the local area signal data in the Az (2) direction from the local area signal data after replacement processing stored in the main memory 2. And each of the selected Q point signal data is stored in the local memory 12a (step ST29 in FIG. 4).
The inverse Fourier transform unit 15a performs inverse Fourier transform in the Az (2) direction on each of the signal data in the local area of the Q point stored in the local memory 12a (step ST30 in FIG. 4). That is, the inverse Fourier transform unit 15a performs a Q-point inverse Fourier transform in the Az (2) direction.
The layout setting unit 13a performs the memory layout replacement process so that the Q-point signal data is continuous in the Rg direction as before the process shown in FIG. 4, and the Q-point signal data after the replacement process is stored in the main memory. 2 (step ST31 in FIG. 4).

  As apparent from the above, according to the first embodiment, the signals of the plurality of local regions divided by the data dividing units 11a and 11b in accordance with the processing direction of the Fourier transform process or the process direction of the inverse Fourier transform process. Layout setting units 13a and 13b for setting the memory layout, and the Fourier transform units 14a and 14b receive a plurality of signals in a plurality of local regions whose memory layout is matched with the processing direction of the Fourier transform process by the layout setting units 13a and 13b. Dimensional Fourier transform is performed, and inverse Fourier transform units 15a and 15b are configured to perform multidimensional inverse Fourier transform on signals in a plurality of local regions whose memory layout is matched to the processing direction of the inverse Fourier transform process by layout setting units 13a and 13b. Therefore, it is necessary to complete the reproduction of the synthetic aperture radar image from the observation data. An effect that it is possible to reduce the time.

In the flowchart of FIG. 4, signal processing is performed once after the signal data is converted into the frequency domain in the Az direction and the frequency domain in the Rg direction. In addition, the signal processing is performed once after the signal data is inversely converted to the time domain in the Rg direction, the signal processing is performed once after the signal data is converted again into the frequency domain in the Rg direction, An example is shown in which signal processing is performed three times in total.
When the processing direction of the Fourier transform process or the process direction of the inverse Fourier transform process is changed, the multi-dimensional Fourier transform or inverse Fourier transform is performed, and the entire process is localized. Often, the number of times signal processing is performed on signal data is not limited to three. Further, the signal processing area for the signal data is not limited to the above-described area.
As described above, in the first embodiment, the entire process is localized by performing multi-dimensional Fourier transform or inverse Fourier transform. For this reason, the image reproduction process can be completed in a short time by performing local operations on the local memories 12a and 12b as much as possible and simultaneously executing the local operations by the plurality of image regenerators 3a and 3b. .

Further, the input / output processing between the main memory 2 and the local memories 12a and 12b and the processing on the local memories 12a and 12b may be executed simultaneously, and signals in different local areas may be processed in parallel. It may be executed simultaneously.
Accordingly, on the local memories 12a and 12b, signal data transfer processing in a certain local area between the main memory 2 and the local memories 12a and 12b and signal data in a local area different from the signal data in a certain local area are performed. These processes may be executed in parallel.

In the first embodiment, an example in which the SAR image reproduction process by the image reproducer 3a is the process shown in the flowchart of FIG.
This is only an example, and when the image regenerator 3a reproduces the synthetic aperture radar image, the signal data is first divided in the Az direction by using the sub-aperture method disclosed in Patent Document 1. You may make it implement the process which the flowchart of FIG. 4 shows for every signal data.

The first embodiment shows an example in which the image regenerator 3a includes a local memory 12a and the image regenerator 3b includes a local memory 12b.
This is merely an example, and each of the image reproducers 3a and 3b may include a plurality of local memories having different memory sizes, and the plurality of local memories may be arranged in the order of the memory sizes.
In this case, the Fourier transform units 14a and 14b may perform Fourier transform hierarchically in accordance with the memory sizes of a plurality of local memories.
For example, the data division units 11a and 11b divide the signal data into local area sizes corresponding to the memory size of the first-stage local memory, and the Fourier transform units 14a and 14b use Az (corresponding to the division number of the signal data. 2) Perform Fourier transform in direction and Rg (2) direction.
Further, the data dividing units 11a and 11b divide the signal data of the divided local region size into local region sizes corresponding to the memory size of the second stage local memory, and the Fourier transform units 14a and 14b divide the signal data. Fourier transforms in the Az (2) direction and Rg (2) direction corresponding to the numbers are performed, and Fourier transforms in the Az (1) direction and Rg (1) direction corresponding to the divided local region sizes are performed.

In the first embodiment, the signal processing device includes the image regenerators 3a and 3b. However, the signal processing device may be realized by a general-purpose computer, for example.
When the signal processing apparatus is realized by a general-purpose computer, one or more cores are allocated to each image regenerator among the CPU multicores included in the general-purpose computer, and the data management unit 1 and the image are stored on the main memory of the CPU. A management unit 4 is configured. Further, the cache memory and the register provided in the general-purpose computer may correspond to the local memory of each image player.
When two or more general-purpose computers are connected by communication, the main memory included in any one general-purpose computer corresponds to the data management unit 1 and the image management unit 4, and each general-purpose computer includes. The cache memory and the register may correspond to the local memory of each image player.
Further, when each general-purpose computer includes a GPU, the main memory of the GPU may correspond to a local memory, and the core included in the GPU may constitute an image regenerator.

In the first embodiment, the memory layout setting process by the layout setting units 13a and 13b, the Fourier transform process by the Fourier transform units 14a and 14b, the inverse Fourier transform process by the inverse Fourier transform units 15a and 15b, and the signal processing units 16a and 16b. In this example, each of the image reproduction processes is executed on the local memories 12a and 12b.
This is merely an example, memory layout setting processing by the layout setting units 13a and 13b, Fourier transform processing by the Fourier transform units 14a and 14b, inverse Fourier transform processing by the inverse Fourier transform units 15a and 15b, and signal processing units 16a and 16b. As long as at least one of the image reproduction processes is performed on the local memories 12a and 12b.

  In the present invention, any constituent element of the embodiment can be modified or any constituent element of the embodiment can be omitted within the scope of the invention.

  The present invention is suitable for an image regenerator, a signal processing device, and an image reproducing method for reproducing a synthetic aperture radar image from observation data obtained by a synthetic aperture radar.

  DESCRIPTION OF SYMBOLS 1 Data management part, 2 Main memory, 3a, 3b Image player, 4 Image management part, 5 Main memory, 11a, 11b Data division part, 12a, 12b Local memory, 13a, 13b Layout setting part, 14a, 14b Fourier transform 15a, 15b Inverse Fourier transform unit, 16a, 16b Signal processing unit, 21 Data management circuit, 22 Data division circuit, 23 Layout setting circuit, 24 Fourier transform circuit, 25 Inverse Fourier transform circuit, 26 Signal processing circuit, 27 image Management circuit, 28 main recording circuit, 29 local recording circuit, 31 memory, 32 processor, 101 signal data, 102, 103 coefficient data, 104-108 dividing line, 201-204 signal data, 205 local area signal data, 301- 303 signal data, 30 Signal data of the local area.

Claims (9)

A data dividing unit that divides observation data obtained by the synthetic aperture radar into signals of a plurality of local regions;
A layout setting unit that sets a memory layout of signals of a plurality of local regions divided by the data dividing unit in accordance with a processing direction of Fourier transform processing or a processing direction of inverse Fourier transform processing;
A Fourier transform unit that multi-dimensionally Fourier transforms signals of a plurality of local regions in which the memory layout is matched to the processing direction of the Fourier transform process by the layout setting unit;
An inverse Fourier transform unit for performing multidimensional inverse Fourier transform on signals of a plurality of local regions in which the memory layout is matched to the processing direction of the inverse Fourier transform process by the layout setting unit;
An image reproduction comprising: a signal processing unit for reproducing a synthetic aperture radar image from a local region signal subjected to multidimensional Fourier transform by the Fourier transform unit or a local region signal subjected to multidimensional inverse Fourier transform by the inverse Fourier transform unit vessel. Observation data from the synthetic aperture radar is stored in the main memory,
The data dividing unit, the dividing the observation data stored by the main memory, stored in b Karumemori divided observation data as the signal of the local region,
One or more processes among the memory layout setting process by the layout setting unit, the Fourier transform process by the Fourier transform unit, the inverse Fourier transform process by the inverse Fourier transform unit, and the image reproduction process by the signal processing unit are described above. 2. The image regenerator according to claim 1, wherein the image regenerator is implemented on a local memory. Observation data from the synthetic aperture radar is stored in the main memory,
The data dividing unit, the dividing the observation data stored by the main memory, stored in b Karumemori divided observation data as the signal of the local region,
A memory layout setting process by the layout setting unit, a Fourier transform process by the Fourier transform unit, an inverse Fourier transform process by the inverse Fourier transform unit, and an image reproduction process by the signal processing unit are performed on the local memory. The image regenerator according to claim 1.   The layout setting unit is stored in the local memory before setting a memory layout of a plurality of local region signals when the processing direction of the Fourier transform process or the processing direction of the inverse Fourier transform process is changed. 4. The image regenerator according to claim 3, wherein a local area signal is transferred to the main memory. There are multiple local memories,
Of the plurality of local memories, the local region signal transfer process between any one of the local memories and the main memory, and the local region signal different from the local region signal to be subjected to the transfer process, the 1 5. The image regenerator according to claim 4, wherein processing on two local memories and different local memories are executed in parallel.   The Fourier transform unit superimposes a plurality of local region signals divided by the data division unit, and performs a two-dimensional Fourier transform process on the superimposed local region signals. The image regenerator as described. A data management unit for managing observation data by a synthetic aperture radar;
An image regenerator that reproduces a synthetic aperture radar image from observation data managed by the data management unit;
An image management unit for managing a synthetic aperture radar image reproduced by the image regenerator,
The image player is
A data dividing unit that divides observation data managed by the data management unit into a plurality of local region signals;
A layout setting unit that sets a memory layout of signals of a plurality of local regions divided by the data dividing unit in accordance with a processing direction of Fourier transform processing or a processing direction of inverse Fourier transform processing;
A Fourier transform unit that multi-dimensionally Fourier transforms signals of a plurality of local regions in which the memory layout is matched to the processing direction of the Fourier transform process by the layout setting unit;
An inverse Fourier transform unit for performing multidimensional inverse Fourier transform on signals of a plurality of local regions in which the memory layout is matched to the processing direction of the inverse Fourier transform process by the layout setting unit;
A signal processing unit for reproducing a synthetic aperture radar image from a local region signal multidimensionally Fourier transformed by the Fourier transform unit or a local region signal multidimensionally inverse Fourier transformed by the inverse Fourier transform unit. A signal processing device.   The signal processing apparatus according to claim 7, wherein a plurality of the image regenerators are mounted, and the plurality of image regenerators execute a synthetic aperture radar image reproduction process in parallel. The data dividing unit divides the observation data from the synthetic aperture radar into a plurality of local region signals,
The layout setting unit sets the memory layout of the signals of the plurality of local regions divided by the data dividing unit according to the processing direction of the Fourier transform process or the processing direction of the inverse Fourier transform process,
The Fourier transform unit performs multi-dimensional Fourier transform on the signals of a plurality of local regions in which the memory layout is matched with the processing direction of the Fourier transform process by the layout setting unit,
The inverse Fourier transform unit performs multidimensional inverse Fourier transform on the signals of a plurality of local regions in which the memory layout is matched with the processing direction of the inverse Fourier transform process by the layout setting unit,
An image reproduction method in which a signal processing unit reproduces a synthetic aperture radar image from a local region signal subjected to multidimensional Fourier transform by the Fourier transform unit or a local region signal subjected to multidimensional inverse Fourier transform by the inverse Fourier transform unit.

Images