KR102272411B1 - Method and apparatus for learning artificial nearal network to improve the target recognition of simulation-image database in sar image - Google Patents

Abstract

An artificial neural network learning method according to one embodiment of the present invention may include: a step of acquiring an actual measurement synthetic aperture radar image; a step of constructing a simulation image database including an image generated by using a very high frequency technique; a step of generating a similar image by inputting the actual measurement synthetic aperture radar image to a cycle GAN; and a step of learning a neural network of a CNN structure using a plurality of images included in the similar image and the simulation image database. Therefore, the present invention is capable of allowing a simulation-like image to be generated.

Description

Artificial neural network learning method and apparatus for increasing the identification rate of simulation image database of synthetic aperture radar images

The present invention relates to an artificial neural network learning method and apparatus for increasing the identification rate of a simulation image database of a synthetic aperture radar image, and more particularly, using a cycle for increasing the identification rate of a simulation image database in a synthetic aperture radar image. It is about measurement image learning technique.

ATR (Automatic Target Recognition) technology for automatically detecting and recognizing a target using SAR (Synthetic Aperture Radar) has been developed. However, for target identification, it is practically difficult to construct a database of targets extracted from measurement images. Acquiring images of targets at various elevations and azimuths is not an easy task. In addition, the characteristics of the target reflected by various surrounding clutter (eg, an artificial object such as a building, a natural object, or the ground surface) shows a different aspect from the reflection characteristic of the target alone. In order to supplement the insufficient database, it is simulated with an electromagnetic wave numerical analysis (Method of Moments, Finite Element Mehtod, Finite Difference Time Domain, High Frequency Methods) based on a CAD model created with indirect information such as an Internet model or a photo, Research on the composition is in progress. For full-wave methods such as MoM, FEM, and FDTD, the resolution of the SAR measurement equipment and the numerical analysis of the X-band are very limited with the current computational power (eg, CPU time and memory). Therefore, most of the methods using the very high frequency technique are being performed, but the accuracy is difficult to guarantee. In addition, the accuracy of CAD model production is also recognized as a limiting factor in producing images using electromagnetic wave numerical analysis at high frequencies.

The simulation image database generated through the electromagnetic wave numerical analysis has limitations in performance in automatically identifying the target. In order to solve this problem, target identification research using the scattering point of the image is being conducted and a complex preprocessing technique is being performed, but it is difficult to guarantee identification performance.

According to an embodiment of the present invention, a simulation-like image may be generated using an interval between cycles for the measured SAR image.

In one embodiment, the artificial neural network learning method performed by the computing device includes the steps of: acquiring a measured synthetic aperture radar image; building a simulation image database including an image generated using a very high frequency technique; It may include generating a similar image by inputting a spherical radar image to a cycle GAN, and learning a neural network of a CNN structure using the similar image and a plurality of images included in the simulation image database. .

In an embodiment, the step of constructing the simulation image database includes constructing the simulation image database by using a Shooting and Bouncing Rays (SBR) technique for a target extracted using the measured synthetic aperture radar image. can do.

In an embodiment, the step of constructing the simulation image database may include generating a simulation image database including a plurality of target images having a specified elevation angle and a specified azimuth interval for each target using a very high frequency technique. .

In an embodiment, the method may include identifying a target included in the actually measured synthetic aperture radar image by using the learned neural network of the CNN structure.

An artificial neural network learning apparatus according to another embodiment of the present invention includes a processor, wherein the processor acquires a measured synthetic aperture radar image, and builds a simulation image database including an image generated using a very high frequency technique, A similar image may be generated by inputting the measured synthetic aperture radar image to a cycle GAN, and a neural network of a CNN structure may be trained using the similar image and a plurality of images included in the simulation image database.

In an embodiment, the simulation image database may be constructed using a Shooting and Bouncing Rays (SBR) technique with respect to a target extracted using the measured synthetic aperture radar image.

In an embodiment, the processor may generate a simulation image database including a plurality of target images having a designated elevation angle and a designated azimuth angle interval for each target by using a very high frequency technique.

In an embodiment, the artificial neural network learning apparatus may identify a target included in the actually measured synthetic aperture radar image by using the learned neural network of the CNN structure.

1 is a diagram for explaining a method of learning an artificial neural network using an actually measured synthetic aperture radar image.
2 is a diagram for explaining the configuration and operation of an artificial neural network learning apparatus according to an embodiment of the present invention.
3 is a diagram for explaining the configuration of a processor of an artificial neural network learning apparatus according to an embodiment of the present invention.
4 is a flowchart of an artificial neural network learning method according to an embodiment of the present invention.
5 and 6 are diagrams for explaining an experimental example of a simulation result of an artificial neural network learned according to the prior art shown in FIG. 1 .
7 is a diagram for explaining an artificial neural network learning method according to an embodiment of the present invention.
8 is a diagram for explaining an experimental example of a simulation result of an artificial neural network learned according to an embodiment of the present invention.
9 is a view for explaining the effect of the present invention.

The terms used in the present embodiments have been selected as currently widely used general terms as possible while considering the functions in the present embodiments, which may vary depending on the intention or precedent of a person skilled in the art, the emergence of new technology, etc. have. In addition, in a specific case, there is a term arbitrarily selected by the applicant, and in this case, the meaning will be described in detail in the relevant part. Therefore, the terms used in the present embodiments should be defined based on the meaning of the term and the contents throughout the present embodiments, rather than the simple name of the term.

Since the present embodiments may have various changes and may have various forms, some embodiments will be illustrated in the drawings and described in detail. However, this is not intended to limit the present embodiments to a specific disclosed form, and it should be understood to include all modifications, equivalents and substitutes included in the spirit and scope of the present embodiments. The terms used herein are used only for description of the embodiments, and are not intended to limit the present embodiments.

Unless otherwise defined, terms used in the present embodiments have the same meanings as commonly understood by those of ordinary skill in the art to which the present embodiments belong. Terms such as those defined in commonly used dictionaries should be interpreted as having meanings consistent with the meanings in the context of the related art, and unless explicitly defined in the present embodiments, they have an ideal or excessively formal meaning. should not be interpreted.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS In the following detailed description of the present invention, reference is made to the accompanying drawings which show by way of illustration specific embodiments in which the present invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the present invention. It should be understood that the various embodiments of the present invention are different but need not be mutually exclusive. For example, certain shapes, structures, and characteristics described herein may be implemented with changes from one embodiment to another without departing from the spirit and scope of the present invention. In addition, it should be understood that the location or arrangement of individual components within each embodiment may be changed without departing from the spirit and scope of the present invention. Accordingly, the following detailed description is not to be taken in a limiting sense, and the scope of the present invention should be taken as encompassing the scope of the claims and all equivalents thereto. In the drawings, like reference numerals refer to the same or similar elements throughout the various aspects.

Hereinafter, various embodiments of the present invention will be described in detail with reference to the accompanying drawings in order to enable those of ordinary skill in the art to easily practice the present invention.

Hereinafter, a method of learning an artificial neural network using a measured synthetic aperture radar image will be described in detail with reference to FIG. 1 .

1 shows an existing deep learning technique for training and identifying an actual measurement image using a simulation image database 20 in a deep learning network. The simulation image database may configure the simulation image database 20 by using Shooting and Bouncing Rays (SBR), which is one of the very high frequency techniques 10 . The configuration of the simulation image database 20 may be configured at intervals of 2 degrees in elevation and 2 degrees in azimuth for each target. The deep learning network 40 may use VGG. The test of the deep learning network 40 is performed through the identification of the SAR image 30 acquired through the actual measurement with the SAR image database constructed by simulation.

However, there is a problem in that it is difficult to guarantee the accuracy of the simulation image database constructed using the above-described very high frequency technique. In the case of a deep learning network trained according to an artificial neural network learning method according to some embodiments of the present invention, which will be described later, a higher identification rate for the prediction result 50 may be guaranteed than in the embodiment shown in FIG. 1 .

2 is a diagram for explaining the configuration and operation of an artificial neural network learning apparatus according to an embodiment of the present invention.

In an embodiment, the artificial neural network learning server 100 may include a memory 101 , a processor 102 , a communication module 103 , and an input/output interface 104 .

The memory 101 is a computer-readable recording medium and may include a random access memory (RAM), a read only memory (ROM), and a permanent mass storage device such as a disk drive. In addition, the memory 101 may temporarily or permanently store program codes and settings for controlling the artificial neural network learning server 100 , point cloud data, and surface information.

The processor 102 may be configured to process instructions of a computer program by performing basic arithmetic, logic, and input/output operations. The instructions may be provided to the processor 102 by the memory 101 or the communication module 103 . For example, the processor 102 may be configured to execute instructions received according to program code stored in a recording device, such as the memory 101 . In an embodiment, the processor 102 of the artificial neural network learning server 100 acquires a measured synthetic aperture radar image, builds a simulation image database including an image generated using a very high frequency technique, and the measured synthetic aperture A radar image may be input to a cycle GAN to generate a similar image, and a neural network of a CNN structure may be trained using the similar image and a plurality of images included in the simulation image database.

The communication module 103 may provide a function for communicating with an external server through a network. For example, a request generated by the processor 102 of the artificial neural network learning server 100 according to a program code stored in a recording device such as the memory 101 is transmitted to an external server through a network under the control of the communication module 103 . can be Conversely, a control signal, command, content, file, etc. provided under the control of the processor of the external server may be received by the artificial neural network learning server 100 through the communication module 103 through the network. For example, a control signal or command of an external server received through the communication module 103 may be transmitted to the processor 102 or the memory 101 , and the content or file may be further transmitted by the artificial neural network learning server 100 . It may be stored in a storage medium that may include.

The communication method is not limited, and not only a communication method using a communication network (eg, a mobile communication network, a wired Internet, a wireless Internet, a broadcasting network) that the network may include, but also short-range wireless communication between devices may be included. For example, the network includes a personal area network (PAN), a local area network (LAN), a campus area network (CAN), a metropolitan area network (MAN), a wide area network (WAN), a broadcast database and network (BBN), may include any one or more of the networks, such as the Internet. Further, the network may include, but is not limited to, any one or more of a network topology including, but not limited to, a bus network, a star network, a ring network, a mesh network, a star-bus network, a tree or a hierarchical network, and the like. .

Also, the communication module 103 may communicate with an external server through a network. Although the communication method is not limited, the network may be a local area wireless network. For example, the network may be a Bluetooth (Bluetooth), BLE (Bluetooth Low Energy), or Wifi communication network.

The input/output interface 104 may be a means for interfacing with an input/output device. For example, the input device may include a device such as a keyboard or mouse, and the output device may include a device such as a display for displaying a communication session of an application. As another example, the input/output interface 104 may be a means for an interface with a device in which functions for input and output are integrated into one, such as a touch screen. As a more specific example, the processor 102 of the artificial neural network learning server 100 processes the command of the computer program loaded in the memory 101, and the service screen or content configured using the data provided by the external server is an input/output interface. may be displayed on the display via 104 .

In addition, in other embodiments, the artificial neural network learning server 100 may include more components than those of FIG. 1 . However, there is no need to clearly show most of the prior art components. For example, the artificial neural network learning server 100 may include a battery and a charging device for supplying power to internal components of the artificial neural network learning server, and is implemented to include at least some of the above-described input/output devices or a transceiver. It may further include other components such as a transceiver, a global positioning system (GPS) module, various sensors, and a database.

Hereinafter, the internal configuration of the processor of the artificial neural network learning server according to an embodiment of the present invention will be reviewed in detail with reference to FIG. 3 . For ease of understanding, it is assumed that the following processor is the processor 102 of the artificial neural network learning server 100 shown in FIG. 21, but in an embodiment, the processor described later when learning of the artificial neural network is performed in an external server Note that may be a processor of the above-described external server.

The processor 102 of the artificial neural network learning server according to an embodiment of the present invention includes the SAR image acquisition unit 111 , the database construction unit 112 , the similar image generation unit 113 , the neural network learning unit 114 and the target identification. part 115 . According to some embodiments, components of the processor 102 may be selectively included or excluded from the processor 102 . In addition, according to some embodiments, components of the processor 102 may be separated or merged to represent the functionality of the processor 102 .

The processor 102 and the components of the processor 102 may control the artificial neural network learning server 100 to perform the steps S110 to S150 included in the artificial neural network learning method of FIG. 4 . For example, the processor 102 and the components of the processor 102 may be implemented to execute instructions according to the code of the operating system included in the memory 101 and the code of at least one program. Here, the components of the processor 102 are expressions of different functions of the processor 102 performed by the processor 102 according to instructions provided by the program code stored in the artificial neural network learning server 100 . can take The internal configuration and specific operation of the processor 102 will be described with reference to the flowchart of the artificial neural network learning method of FIG. 4 and the embodiments of FIGS. 7 to 9 .

4 is a flowchart illustrating an artificial neural network learning method according to an embodiment of the present invention in time series.

In step S110, the artificial neural network learning server may acquire the actually measured synthetic aperture radar image.

In step S120, the artificial neural network learning server may build a simulation image database including images generated using the ultra-high frequency technique. In an embodiment, the artificial neural network learning server may construct the simulation image database by using a Shooting and Bouncing Rays (SBR) technique with respect to a target extracted by using the measured synthetic aperture radar image. Alternatively, in another embodiment, the artificial neural network learning server may generate a simulation image database including a plurality of target images having a specified elevation angle and a specified azimuth interval for each target by using a very high frequency technique.

In step S130, the artificial neural network learning server may generate a similar image by inputting the measured synthetic aperture radar image to the cycle GAN.

In step S140, the artificial neural network learning server may train the neural network of the CNN structure by using a plurality of images included in the similar image and simulation image database.

In step S150, the artificial neural network learning server may identify the target included in the actually measured synthetic aperture radar image by using the learned neural network of the CNN structure.

5 and 6 are diagrams for explaining an experimental example of a simulation result of an artificial neural network learned according to the prior art shown in FIG. 1 .

5 is a table showing the number of simulation image DBs used in the test, the number of DBs built through training, and the number of images used in the test. Referring to the figure, the number of images stored in the simulation image database is 23,400, the number of images stored in the training database of the artificial neural network is 6,205, and the number of images used as experimental data of the artificial neural network is 6,199.

6 shows the simulation result of FIG. 1 as a confusion matrix. It is assumed that there are four types of targets used here. Using the simulation image as the identification database, training and testing with the measured synthetic aperture radar image, it can be confirmed that the overall identification rate is approximately 20%. In addition, referring to the drawings, it can be seen that most of the targets are identified as the second target or are misidentified. In general, when using a simulation image DB, the identification rate is improved by using various pre-processing techniques and image adjustment techniques, but it is assumed that no improvement technique is used in this test.

7 is a diagram for explaining an artificial neural network learning method according to an embodiment of the present invention.

The artificial neural network learning method according to an embodiment of the present invention acquires an actual measured synthetic aperture radar image (30), builds a simulation image database including an image 10 generated using a very high frequency technique (20), A similar image is generated (210) by inputting the measured synthetic aperture radar image 30 into a cycle GAN (200), and a plurality of images included in the similar image 210 and the simulation image database ( 20) can be used to train the neural network 230 of the CNN structure. Then, the target included in the actually measured synthetic aperture radar image may be identified using the learned neural network of the CNN structure ( 240 ).

The artificial neural network learning method according to some embodiments of the present invention may use Cycle GAN, which is one of deep learning techniques, to generate a similar image and improve the identification rate. That is, in the present embodiment, a synthetic aperture radar image similar to a simulation may be generated by using the measured synthetic aperture radar image between cycles. Thereafter, the simulated synthetic aperture radar image is generated by inputting the measured synthetic aperture radar image to the inter-cycle network, and the generated pseudo synthetic aperture radar image is input to the artificial neural network of the CNN structure. The type and shape of the artificial neural network of the CNN structure are not limited. For example, in the artificial neural network learning method according to an embodiment, the above-described synthetic aperture radar image and simulation image database may be input to the VGG network to learn and test.

FIG. 8 is a diagram showing a simulation result of the embodiment of the present invention shown in FIG. 7 as a confusion matrix. Compared with the embodiment shown in FIG. 6 , it can be seen that the result of the embodiment of FIG. 8 according to an embodiment of the present invention is more improved. In other words, it can be seen that the probability of identifying the target as the original target is increased because the result, which was biased toward a specific target, is learned to be specialized in the simulation image database through the cycles.

Referring to FIG. 9 , it is possible to confirm the improved performance of the target identification rate of the artificial neural network learned according to some embodiments of the present invention. Referring to the bar, it can be seen that the overall identification rate is innovatively improved from 20% to 84%. 5 and 6, all targets are identified as No. 2, and there is a problem that targets 1, 3, and 4 are not identified at all, whereas the artificial neural network learning method according to some embodiments of the present invention is used. When used, it can be confirmed that the high identification rates of 83%, 87%, and 88% are improved by using between cycles.

That is, as described above, according to the artificial neural network learning method according to some embodiments of the present invention, when identification is performed through a simulation image database in a synthetic aperture radar image, the measured image is simulated by using between cycles of deep learning. The identification rate can be innovatively improved by generating an image and learning the generated image with a deep learning network of CNN structure.

On the other hand, those of ordinary skill in the technical field related to the present embodiment will understand that it can be implemented in a modified form within a range that does not deviate from the essential characteristics of the above description. Therefore, the disclosed methods are to be considered in an illustrative rather than a restrictive sense. The scope of the present invention is indicated in the claims rather than the foregoing description, and all differences within the scope equivalent thereto should be construed as being included in the present invention.

The device described above may be implemented as a hardware component, a software component, and/or a combination of the hardware component and the software component. For example, devices and components described in the embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA). , a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions, may be implemented using one or more general purpose or special purpose computers. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to execution of the software. For convenience of understanding, although one processing device is sometimes described as being used, one of ordinary skill in the art will recognize that the processing device includes a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that can include For example, the processing device may include a plurality of processors or one processor and one controller. Other processing configurations are also possible, such as parallel processors.

The software may comprise a computer program, code, instructions, or a combination of one or more thereof, which configures a processing device to operate as desired or is independently or collectively processed You can command the device. The software and/or data may be any kind of machine, component, physical device, virtual equipment, computer storage medium or device, to be interpreted by or to provide instructions or data to the processing device. , or may be permanently or temporarily embody in a transmitted signal wave. The software may be distributed over networked computer systems, and stored or executed in a distributed manner. Software and data may be stored in one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and available to those skilled in the art of computer software. Examples of the computer readable recording medium include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floppy disks. - includes magneto-optical media, and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

As described above, although the embodiments have been described with reference to the limited embodiments and drawings, various modifications and variations are possible from the above description by those skilled in the art. For example, the described techniques are performed in a different order than the described method, and/or the described components of the system, structure, apparatus, circuit, etc. are combined or combined in a different form than the described method, or other components Or substituted or substituted by equivalents may achieve an appropriate result.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (9)

In the artificial neural network learning method performed by a computing device,
acquiring an actual measured synthetic aperture radar image;
generating a simulation image for the actually measured synthetic aperture radar image by using an ultra-high frequency technique, which is an electromagnetic wave numerical analysis technique, and constructing a simulation image database including the generated simulation image;
generating a similar image by inputting the measured synthetic aperture radar image to a cycle GAN; and
The similar image is used as input data of the neural network of the CNN structure, the class of the similar image is obtained as the result data of the neural network of the CNN structure, and the simulation image of the synthetic aperture radar image corresponding to the similar image is obtained. training the neural network of the CNN structure using a class as target data; containing,
How to learn artificial neural networks.
According to claim 1,
The step of building the simulation image database includes:
Constructing the simulation image database using a shooting and bouncing rays (SBR) technique, which is one of the very high frequency techniques, with respect to the target extracted using the measured synthetic aperture radar image,
How to learn artificial neural networks.
According to claim 1,
The simulation image is
Using a very high frequency technique, including a plurality of target images of a specified elevation angle and a specified azimuth interval with respect to at least one target included in the simulation image,
How to learn artificial neural networks.
delete processor; including;
The processor is
Acquire an actual measured synthetic aperture radar image, generate a simulation image for the measured synthetic aperture radar image using an ultra-high frequency technique that is an electromagnetic wave numerical analysis technique, and build a simulation image database including the generated simulation image, A similar image is generated by inputting the measured synthetic aperture radar image to a cycle GAN, the similar image is used as input data of the neural network of the CNN structure, and the class of the similar image is used as input data of the neural network of the CNN structure. Learning the neural network of the CNN structure by acquiring the result data and using the class of the simulation image for the actually measured synthetic aperture radar image corresponding to the similar image as the target data,
Artificial neural network learning device.
6. The method of claim 5,
The processor is
constructing the simulation image database by using a Shooting and Bouncing Rays (SBR) technique, which is one of the very high frequency techniques, with respect to the target extracted using the measured synthetic aperture radar image;
Artificial neural network learning device.
6. The method of claim 5,
The simulation image is
Using a very high frequency technique, including a plurality of target images of a specified elevation and a specified azimuth interval with respect to at least one target included in the simulation image, Artificial neural network learning apparatus.
delete A computer program stored in a recording medium for executing the method of any one of claims 1 to 3 using a computer.

Images