KR102308075B1 - Machine-learning system for operating guided missiles - Google Patents

Abstract

The present invention relates to a missile guidance command learning system and a method for acquiring guidance data for learning a missile guidance command and learning the missile guidance command based thereon. According to an embodiment of the present invention, a missile guidance command learning system comprises: an image information collection unit which collects image information generated from a searcher; a pre-processing unit which receives target information, sets a position of a target in the image information collected from the image information collection unit, and pre-processes the image data; an image data selection unit which selects arbitrary image data from among a plurality of pre-processed image data; and a driving command generating unit which receives the selected image data and extracts features to learn the target position and target information of the received image and generate a driving command for a guided missile.

Description

Missile guidance command learning system {MACHINE-LEARNING SYSTEM FOR OPERATING GUIDED MISSILES}

The present invention relates to a missile guidance command learning system, and to a system capable of acquiring data necessary for guidance and adjustment of a missile, learning it using artificial intelligence, and generating a guided missile control command.

Conventional general guided missiles perform a task of tracking a target after acquiring a target through an optical device, are given a mission, are loaded with information, and are fired according to a preset procedure. Thereafter, the launched guided missile detects or acquires a target according to an operation concept for target detection, a guidance method, and the like.

For such guided missiles, the developer developed the driving command for missile guidance with an algorithm and mounted it on the guided missile. A number of repeated tests are required to develop an accurate driving command. Therefore, it is difficult to perform many tests.

On the other hand, with the development of artificial intelligence (AI) technology, artificial intelligence technology is being introduced in many fields, and in the case of automobiles, artificial intelligence technology is actively being introduced into autonomous driving. It is necessary to introduce such artificial intelligence technology to the field of guided missiles.

According to an embodiment of the present invention, an object of the present invention is to provide a missile guidance command learning system for acquiring guidance data for learning the missile guidance command and learning the missile guidance command based on this.

In addition, according to an embodiment of the present invention, an object of the present invention is to provide an apparatus and method for acquiring guidance data for learning a missile guidance command.

However, the object of the present invention is not limited to the above, and other objects not mentioned will be clearly understood by those skilled in the art from the following description.

Missile guidance command learning system according to an embodiment of the present specification includes an image information collection unit for collecting image information generated from a searcher; a pre-processing unit for receiving target information, setting a position of a target in the image information collected from the image information collecting unit, and pre-processing the image data; an image data selection unit for selecting arbitrary image data from among a plurality of pre-processed image data; and a driving command generator for receiving the selected image data, extracting features, learning the target position and target information of the input image, and generating a driving command for the guided missile.

In addition, the system includes a driving command collecting unit for collecting a driving command for transmitting the guided control direction of the guided missile to the driving device; a driving command modifying unit receiving the driving command collected by the driving command collecting unit and adjusting the collected driving command to a driving command corresponding to an image arbitrarily selected by the image data selecting unit; a driving command comparison unit for extracting error data by comparing the driving command adjusted by the driving command correction unit with the driving command generated by the driving command generating unit; and a weight calculation unit that calculates a weight based on the error data extracted by the drive command comparison unit and transmits it to the drive command generation unit.

Preferably, the driving command generating unit receives the weight transmitted from the weight calculation unit and updates the generated driving command.

Preferably, the target information is characterized in that it includes an expected region of interest (ROI) of the target, an expected temperature according to the target distance, and distance information to the target.

Preferably, the preprocessor sets the position information of the target in the image received from the image information collecting unit based on the expected ROI of the target, and the pixel of the image based on the expected temperature according to the target distance and the distance to the target information It is characterized in that the value is corrected.

Preferably, the driving command generator extracts a feature corresponding to a target from image data using a convolutional layer, classifies the extracted feature value using a neural network, and , characterized in that the driving command is generated.

Preferably, the weight calculator calculates the weight using an error backpropagation algorithm.

According to an embodiment of the present invention, by mounting a device capable of acquiring guidance command data of a missile, such as a drone or manned aircraft, it is possible to economically acquire a searcher image and driving data for learning a guided missile, and using this, it is possible to obtain a missile There is an effect that can provide a missile guidance command system that learns the guidance command.

In addition, according to an embodiment of the present invention, there is an effect that can improve the guidance command accuracy of the guided missile by securing the guided missile drive data and utilizing it in the artificial intelligence system.

1 is a block diagram schematically illustrating an apparatus for obtaining derived data according to an embodiment of the present invention.
2 is a block diagram schematically showing the configuration of a missile guidance command learning system according to an embodiment of the present invention.
3 is a reference diagram for explaining a method for learning a target position of a convolutional neural network.
4 is a flowchart sequentially illustrating the operation of the missile guidance command learning system according to an embodiment of the present invention.
Figure 5 shows the artificial neural network structure of the missile guidance command learning system according to an embodiment of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. First of all, it should be noted that in adding reference numerals to the components of each drawing, the same components are given the same reference numerals as much as possible even though they are indicated on different drawings. In addition, in describing the present invention, if it is determined that a detailed description of a related known configuration or function may obscure the gist of the present invention, the detailed description thereof will be omitted. In addition, preferred embodiments of the present invention will be described below, but the technical spirit of the present invention is not limited thereto and may be variously implemented by those skilled in the art without being limited thereto.

1 is a block diagram schematically illustrating an apparatus 100 for obtaining derived data according to an embodiment of the present invention.

The guidance data acquisition device 100 is mounted on a drone or manned aircraft to acquire real data such as driving commands and image information, and the acquired real data is used for learning guided missiles.

The searcher and guidance data acquisition device is a device that acquires data by installing it on a drone or manned device in order to acquire actual guided missile data. Through this device, searcher images and driving data are collected for learning guided missiles. Through this device, the actual driving command when the searcher image comes in is simultaneously collected.

The induction data acquisition device 100 includes a driving command receiving unit 10 , an image searching unit 20 , and a driving command/image collecting unit 30 .

The driving command receiving unit 10 collects a driving command for giving an actual guided control direction of the guided missile to the driving device.

Specifically, the driving command to give the actual guided control direction of the missile is collected to the driving device, and the image search unit 20 collects the searcher image used for the guided missile, and as shown in the figure, the left direction and the middle of the team color flag image. and real images in the right direction are collected.

The image search unit may include a left image search unit, a middle image search unit, and a right image search unit, and the left image search unit collects a left direction of images of a searcher (IIR or CCD) used for guided missiles.

The intermediate image search unit collects the actual searcher (IIR or CCD) image of the intermediate position of the guided missile.

The image search unit on the right collects images in the right direction of the searcher (IIR or CCD) used for guided missiles.

According to an embodiment of the present invention, the image search unit 20 may be an optical searcher, which may be implemented in an infrared tracking (heat tracking) method, an image induction method, or a laser induction method according to an operation method. The infrared searcher identifies the target by the strength and weakness of any signal, and the image searcher recognizes the shape of the target identified through the camera just as a human recognizes the target with the eyes. On the other hand, the image searcher may be implemented in a visible light method or an infrared method. The infrared method is a kind of infrared camera (thermal camera) and is also called an infrared image (IIR, Imaging Infra-Red) method. The infrared imaging method has the advantage that it can be used in more diverse situations because the shape of the target can be checked even in bad weather or at night. The laser searcher is a missile that only chases a specific laser light. In the laser aiming method, the missile does not emit any signals on its own, but at least an ally must send out some signal to guide the missile, so it is semi-active induction. It belongs to the method, and it is also called SAL (Semi-Active Laser) induction method.

By mounting the guidance data acquisition device as described above on a drone, etc., it is possible to easily acquire the actual searcher image and output value driving data required for guided control of the guided missile. It is possible to secure the drive command data of the guided missile.

2 is a block diagram schematically showing the configuration of the missile guidance command learning system 200 according to an embodiment of the present invention.

The missile guidance command learning system 200 performs data learning using an artificial intelligence learning algorithm such as a convolutional neural network based on data collected through the guidance data acquisition device 100 described above with reference to FIG. It is possible to generate the driving command data of the guided missile in the situation.

The missile guidance command learning system trains the neural network based on the data collected from the previous searcher and guidance data learning system. In the learning process, three randomly adjusted image data are input, output the final guidance command, and compare it with the actual driving command to create a CNN network that minimizes the error value.

For this purpose, the missile guidance command learning system 200 includes a driving command collecting unit 31, an image information collecting unit 32, a driving command modifying unit 201, a preprocessing unit 203, an image data selecting unit 205, and driving. It includes a command generation unit 207 , a driving command comparison unit 209 , and a weight calculation unit 211 .

The driving command collecting unit 31 collects driving commands for transmitting the guided control direction of the guided missile to the driving device, and the image information collecting unit 32 collects image information generated from the searcher.

Specifically, the image information collecting unit 32 includes a left image collecting unit, an intermediate image collecting unit, and a right image collecting unit, the left image collecting unit inputs the collected left searcher image, and the middle image collecting unit collects the collected intermediate searcher image. Enter The right image collecting unit inputs the collected right navigator image.

The driving command collecting unit 31 receives the actually collected driving command.

The pre-processing unit 203 receives target information, sets the position of the target in the image information collected from the image information collection unit 32, and pre-processes the image data.

The first input information input to the preprocessor 203 is the target expected ROI, the second input information is the expected temperature according to the target distance, and the third input information is distance information to the target. .

The pre-processing unit 203 performs pre-processing using the target information possessed by the guided missile, and sets the position information of the target in the searcher image using the target expected ROI, which is the first input information.

The target information input to the preprocessor 203 includes an expected region of interest (ROI) of the target, an expected temperature according to the target distance, and distance to the target information, and the image information collecting unit 32 based on the expected ROI of the target ), sets the position information of the target in the image received from the ), and corrects the pixel value of the image based on the expected temperature according to the target distance and the distance to the target information.

According to an embodiment of the present invention, the search unit may collect an image by using an IR image, and this is because most targets such as a tank, which are the main targets of a guided missile, emit heat above a certain temperature. can be used to pre-process the collected images. Here, if the expected temperature value of the target is th, and the pixel values of the image are x and y, the image pixel values processed by the preprocessor are as shown in Equation 1 below.

The image data selection unit 205 selects arbitrary image data from among a plurality of pre-processed image data.

At this time, the driving command correction unit 201 receives the driving command collected by the driving command collecting unit 31 , and adjusts the collected driving command to the driving command corresponding to the image arbitrarily selected by the image data selection unit 205 . do.

The driving command correction unit 201 corrects the actually collected driving command into a command rotated to match the randomly returned image.

The driving command generating unit 207 receives the selected image data, extracts features, learns the target position and target information of the input image, and generates a driving command for the guided missile.

The driving command generating unit 207 outputs a driving command through the CNN network.

Thereafter, the driving command comparison unit 209 compares the actual driving command with the driving command issued through CNN.

That is, error data is extracted by comparing the driving command adjusted by the driving command correction unit 201 and the driving command generated by the driving command generating unit 207 .

Based on the error data extracted from the drive command comparison unit 209 , the weight calculation unit 211 calculates and transmits the weights to the drive command generation unit, and the drive command generation unit 207 transmits the weights transmitted from the weight calculation unit 211 . The generated driving command is updated by using the weight.

Meanwhile, the driving command generator 207 may be implemented as a convolutional neural network (CNN) according to an embodiment of the present invention. That is, a feature corresponding to a target is extracted from image data using a convolutional layer, and the extracted feature value is classified using a neural network, and then a driving command is generated.

The convolutional layer serves to extract features from the input data as described above. The convolutional layer consists of a filter that extracts features, and an activation function that converts the filter value into a nonlinear value. combined to extract features. After the feature is extracted from the convolutional layer, the extracted feature value is input to a neural network (artificial neural intelligence network) for classification.

3 is a reference diagram for explaining a method for learning a target position of a convolutional neural network.

As shown, the CNN can be divided into an area for extracting features of an image (Feature extraction) and an area for classifying a class (Classification).

The feature extraction area is composed of several layers of Convolution Layer and Pooling Layer. The Convolution Layer is an essential element that reflects the activation function after applying a filter to the input data, and the Pooling Layer located after the Convolution Layer is an optional layer. At the end of the CNN, a Fully Connected layer for image classification is added. Between the part that extracts the features of the image and the part that classifies the image, there is a flatten layer that makes the data in the form of an image into an array. As shown, in CNN, a filter traverses input data for image feature extraction, calculates convolution, and uses the calculation result to create a feature map. In the Convolution Layer, the shape of the output data is changed according to the filter size, stride, whether padding is applied, and the max pooling size.

Meanwhile, according to an embodiment of the present invention, the weight calculator 211 may calculate a weight using an error backpropagation algorithm.

The backpropagation algorithm is a method of learning a neural network with the input and output known, and it re-updates the weight by sending an error back to the input direction through the result value again.

4 is a flowchart sequentially illustrating the operation of the missile guidance command learning system according to an embodiment of the present invention.

First, the image information generated from the searcher and the driving command for transmitting the guided control direction of the guided missile to the driving device are collected (S401, S403).

Thereafter, the target information is received, the position of the target is set in the collected image information, and the image data is pre-processed (S405). Target information input for image data pre-processing includes the expected ROI (Region of Interest) of the target, the expected temperature according to the target distance, and distance information to the target, and based on the expected ROI of the target, the image information collecting unit 32 ), sets the position information of the target in the image received from the ), and corrects the pixel value of the image based on the expected temperature according to the target distance and the distance to the target information.

Random image data is selected from among a plurality of pre-processed image data (S407). Meanwhile, the collected driving command is adjusted based on the selected image data (S409).

Thereafter, by extracting features from the selected image data, the target position and target information are learned (S411), and a driving command of the guided missile is generated through this (S413).

According to an embodiment of the present invention, machine learning for generating a driving command may be implemented as a convolutional neural network (CNN). That is, a feature corresponding to a target is extracted from image data using a convolutional layer, and the extracted feature value is classified using a neural network, and then a driving command is generated.

When the driving command is generated, the actual collected driving command is compared with the generated driving command (S415), and the error data is extracted to calculate a weight (S417).

According to an embodiment of the present invention, the weight calculation may use an error backpropagation algorithm.

Thereafter, the generated driving command is updated based on the calculated weight.

Figure 5 shows the artificial neural network structure of the missile guidance command learning system according to an embodiment of the present invention.

As shown in FIG. 5 , it is preferable to use a convolutional neural network (CNN), and includes a first layer, a third layer, and a fifth layer for performing a convolution operation.

Specifically, in the first step, the first layer performs a convolution operation, and includes a 20x20 filter, 5 channels, and 5 stride.

In the second step, the second layer performs a Max Pooling operation and includes a 5x3 filter, 1channel, and stride 5.

In the third step, the third layer performs a convolution operation, and includes a 5x5 filter, 1 channel, and stride 1.

In the fourth step, the fourth layer performs an average pooling operation, and includes a 3x3 filter, 1 channel, and stride 2.

In the fifth step, the fifth layer performs a convolution operation and includes a 5x5 filter, 1 channel, and stride 1.

After that, performing full connection in steps 6 and 7,

In step 8, the output layer is located.

Even if all the components constituting the embodiment of the present invention described above are described as being combined or operated in combination, the present invention is not necessarily limited to this embodiment. That is, within the scope of the object of the present invention, all the components may operate by selectively combining one or more. In addition, although all of the components may be implemented as one independent hardware, some or all of the components are selectively combined to perform some or all of the functions of one or a plurality of hardware programs It may be implemented as a computer program having In addition, such a computer program is stored in a computer readable media such as a USB memory, a CD disk, a flash memory, etc., read and executed by a computer, thereby implementing the embodiment of the present invention. The computer program recording medium may include a magnetic recording medium, an optical recording medium, a carrier wave medium, and the like.

In addition, all terms including technical or scientific terms have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs, unless otherwise defined in the detailed description. Commonly used terms such as terms defined in the dictionary should be interpreted as being consistent with the contextual meaning of the related art, and are not interpreted in an ideal or excessively formal meaning unless explicitly defined in the present invention.

The above description is merely illustrative of the technical idea of the present invention, and various modifications, changes, and substitutions are possible within the range that does not depart from the essential characteristics of the present invention by those of ordinary skill in the art to which the present invention pertains. will be. Accordingly, the embodiments disclosed in the present invention and the accompanying drawings are for explaining, not limiting, the technical spirit of the present invention, and the scope of the technical spirit of the present invention is not limited by these embodiments and the accompanying drawings. . The protection scope of the present invention should be interpreted by the following claims, and all technical ideas within the scope equivalent thereto should be construed as being included in the scope of the present invention.

10: drive command receiving unit
20: image search unit
30: drive command / image receiving unit
100: induction data acquisition device
200: missile guidance command learning system

Claims (10)

an image information collection unit that collects image information generated from the searcher;
a pre-processing unit for receiving target information, setting a position of a target in the image information collected from the image information collecting unit, and pre-processing the image data;
an image data selection unit for selecting arbitrary image data from among a plurality of pre-processed image data; and
and a driving command generating unit that receives the selected image data and extracts features to learn the target position and target information of the input image and generates a driving command for the guided missile,
The target information is a missile guidance command learning system, characterized in that it includes an expected ROI (Region of Interest) of the target, an expected temperature according to the target distance, and distance information to the target. According to claim 1,
The image information collecting unit, missile guidance command learning system, characterized in that the input of the image of the left searcher, the image of the middle searcher, and the image of the right searcher. According to claim 1,
a driving command collecting unit for collecting a driving command for transmitting the guided control direction of the guided missile to the driving device;
a driving command modifying unit receiving the driving command collected by the driving command collecting unit and adjusting the collected driving command to a driving command corresponding to an image arbitrarily selected by the image data selecting unit; and
A missile guidance command learning system further comprising a; a drive command comparison unit for extracting error data by comparing the drive command adjusted by the drive command correction unit and the drive command generated by the drive command generation unit. 4. The method of claim 3,
Missile guidance command learning system further comprising a; weight calculation unit that calculates a weight based on the error data extracted from the drive command comparison unit and transmits it to the drive command generation unit. 5. The method of claim 4,
The drive command generating unit receives the weight transmitted from the weight calculation unit and updates the generated drive command. delete According to claim 1, wherein the pre-processing unit,
setting the location information of the target in the image received from the image information collecting unit based on the expected ROI of the target,
Missile guidance command learning system, characterized in that correcting the pixel value of the image based on the estimated temperature according to the target distance and distance to the target information. The method of claim 1, wherein the driving command generator comprises:
Extracting a feature corresponding to a target from image data using a convolutional layer, classifying the extracted feature value using a neural network, and generating a driving command missile guidance command learning system. 5. The method of claim 4, wherein the weight calculator comprises:
Missile guidance command learning system, characterized in that the weight is calculated using an error backpropagation algorithm. According to claim 1,
The collected image is a missile guidance instruction learning system, characterized in that the image information generated by the infrared image method.

Images