KR102357729B1 - Method and apparatus for performing machine learning and test on plurality of images - Google Patents

Abstract

A method of performing machine learning and testing on a plurality of images includes receiving a plurality of images and performing both learning and testing on the plurality of images, but reflecting the results of the learning and performing the test It includes the step of performing.

Description

METHOD AND APPARATUS FOR PERFORMING MACHINE LEARNING AND TEST ON PLURALITY OF IMAGES

Embodiments disclosed in the present specification relate to a method and apparatus for performing machine learning and testing on a plurality of images, in particular, performing both machine learning and testing on the same image, and reflecting the results of machine learning It relates to a method and apparatus for performing the test.

Recently, AI technology has been used in the field of machine vision, contributing greatly to factory automation. However, in order to use AI technology in the field of machine vision, it is necessary to determine the existence and degree of defects in a product or to determine the type of product. machine learning should be performed. For example, machine learning may be performed on a plurality of images obtained by photographing a good product or a defective product, or machine learning may be performed on a plurality of images obtained by photographing various types of goods.

In general machine vision technology, in order to perform a test to distinguish between good and bad products through photographed images, learning is first performed using images prepared separately for training. For example, by performing learning on a plurality of images of good or defective products, the characteristics of an image of a good product or an image of a defective product are identified, and then the image of the product is tested to determine whether the product is good or not. A method of determining whether a product is defective.

However, in the case of the existing method of separately executing learning and testing as described above, the following problems exist.

First, due to the characteristics of the production process, where the occurrence rate of defective products is extremely low, it is not easy to obtain images of defective products, and it is difficult to classify a plurality of images into images of good products and images of defective products before learning. It takes time and effort to do it separately.

Second, since the test is performed based on the result of one learning, if the environment changes, such as the brightness or location of the lighting illuminating the article after learning, the accuracy of the test result decreases.

In relation to this, Korean Patent Registration No. 10-1867475, a prior art document, discloses clustering of similar models by performing unsupervised learning on static data using an autoencoder.

On the other hand, the above-mentioned background art is technical information that the inventor possessed for the purpose of derivation of the present invention or acquired during the derivation process of the present invention, and it cannot be said that it is necessarily known technology disclosed to the general public before the filing of the present invention. .

Embodiments disclosed herein are intended to provide a machine learning and test performing method and apparatus for increasing test accuracy by reflecting the learning result in real time by performing both machine learning and testing on a plurality of images.

According to an embodiment, as a technical means for achieving the above-described technical problem, a method for performing machine learning and testing on a plurality of images includes receiving a plurality of images, learning and All of the tests are performed, but the method may include performing the test by reflecting the result of performing the learning.

According to another embodiment, as a computer program for performing a method of performing machine learning and testing on a plurality of images in a computing device, the method of performing machine learning and testing on a plurality of images includes: and performing both learning and testing on the plurality of images, and performing the test by reflecting a result of performing the learning.

According to another embodiment, as a computer-readable recording medium in which a program for performing a method of performing machine learning and testing on a plurality of images in a computing device is recorded, machine learning and testing on a plurality of images The performing method may include receiving a plurality of images and performing both learning and testing on the plurality of images, and performing the test by reflecting a result of performing the learning.

According to another embodiment, a device for performing machine learning and testing includes an input/output unit for receiving manipulation and data input related to machine learning and testing, and displaying a data processing result, and communication for transmitting and receiving data with an external device and a communication unit for performing machine learning and testing, a storage unit storing a program for machine learning and testing, and a control unit for performing machine learning and testing on a plurality of images by executing the program, wherein the control unit is configured to perform machine learning and testing through the communication unit. Both learning and testing may be performed on a plurality of received images, but the test may be performed by reflecting a result of performing the learning.

According to any one of the above-described problem solving means, both learning and testing are performed on a plurality of images, but by performing the test by reflecting the learning result, the test can be performed by reflecting the learning result in real time, and thus the environment It can be expected to increase test accuracy by adaptively responding to changes.

In addition, since there is no need to separately prepare an image for learning, it is possible to reduce time and effort required to secure an image of a defective product or to classify an image.

Effects obtainable in the disclosed embodiments are not limited to the above-mentioned effects, and other effects not mentioned are clear to those of ordinary skill in the art to which the embodiments disclosed from the description below belong. can be understood clearly.

1 is a diagram illustrating a system for performing machine learning and testing on a plurality of images according to an embodiment.
FIG. 2 is a block diagram illustrating the configuration of the computing device shown in FIG. 1 .
3 to 8 are flowcharts for explaining a method for performing machine learning and testing on a plurality of images according to embodiments.

Hereinafter, various embodiments will be described in detail with reference to the accompanying drawings. The embodiments described below may be modified and implemented in various different forms. In order to more clearly describe the characteristics of the embodiments, detailed descriptions of matters widely known to those of ordinary skill in the art to which the following embodiments belong are omitted. In addition, in the drawings, parts irrelevant to the description of the embodiments are omitted, and similar reference numerals are attached to similar parts throughout the specification.

Throughout the specification, when a component is said to be "connected" with another component, it includes not only the case where it is 'directly connected', but also the case where it is 'connected with another component in between'. In addition, when a component "includes" a component, it means that other components may be further included, rather than excluding other components, unless otherwise stated.

1 is a diagram illustrating a system for performing machine learning and testing on a plurality of images according to an embodiment. Referring to FIG. 1 , a system for performing machine learning and testing on a plurality of images according to an embodiment may include a photographing unit 10 and a computing device 100 .

The system shown in FIG. 1 is a machine vision system for determining whether the articles 1 are good or defective. The photographing unit 10 captures an image of the article 1 and transmits it to the computing device 100 . Then, the computing device 100 determines whether the photographed article 1 is a good product or a defective product by analyzing the received image. To this end, the computing device 100 learns the received images (hereinafter, used in the same sense as 'machine learning') to learn the characteristics of normal images (images in which good products are photographed) and abnormal images (images in which defective products are photographed). can figure out

The photographing unit 10 is a configuration for photographing the external appearance of the article 1 and may be implemented as a camera including an image sensor. In FIG. 1 , the photographing unit 10 is illustrated as being separate from the computing device 100 . Alternatively, the photographing unit 10 may be included in the computing device 100 . In addition, the photographing unit 10 may include a light for illuminating the article 1 when photographing in order to secure a clearer image.

The photographing unit 10 may photograph the items 1 moving through a conveyor belt, etc. one by one, and transmit the photographed images to the computing device 100 .

The computing device 100 performs machine learning and testing on the plurality of images received from the photographing unit 10 . The computing device 100 determines whether the article 1 included in the image is a good product or a defective product by analyzing the photographed image of the article 1 .

In particular, the computing device 100 may perform both learning and testing on a plurality of images, but may reflect the learning result to the test in real time by performing the test by reflecting the learning result. That is, the computing device 100 may perform both learning and testing on the same image without separating the training image and the test image from each other. Therefore, there is no need to separately prepare an image for learning, and the effect of improving user convenience can be expected. In addition, while performing the test, learning is also performed on the image that was tested at the same time, and the results of learning are reflected when performing the test on the next image, so even if the surrounding environment such as brightness or location of lighting changes, it can be adaptively In response, the effect of increasing the test accuracy can be expected.

A specific method for the computing device 100 to learn and test on a plurality of images will be described below with reference to other drawings.

FIG. 2 is a block diagram illustrating the configuration of the computing device 100 illustrated in FIG. 1 . Referring to FIG. 2 , the computing device 100 according to an embodiment may include an input/output unit 110 , a communication unit 120 , a control unit 130 , and a storage unit 140 .

The input/output unit 110 is configured to receive input from a user, such as settings related to machine learning and test execution, and output test results, and the like. According to an embodiment, the input/output unit 110 may receive a setting input from a user through input devices such as a keyboard and a mouse, and display the results of learning and testing on the screen.

The communication unit 120 is a configuration for performing communication for data transmission/reception with an external device, and may support various types of wired/wireless communication. For example, the communication unit 120 may sequentially receive images obtained by photographing the article 1 from the photographing unit 10 . The communication unit 120 may be implemented as a communication chipset supporting various communication protocols.

The controller 130 is a configuration including at least one processor such as a CPU, and controls the overall operation of the computing device 100 . In particular, the controller 130 may perform machine learning and testing on a plurality of images by executing a program stored in the storage 140 . A detailed operation of the controller 130 performing machine learning and testing on a plurality of images will be described in detail below.

Various types of programs and data may be stored in the storage unit 140 . In particular, a program for performing machine learning and testing on a plurality of images may be stored in the storage 140 , and a plurality of images to be learned and tested may also be stored.

According to an embodiment, the controller 130 may perform learning and testing on a plurality of images received through the communication unit 120 , and may perform the test by reflecting the learning result. In this case, the communication unit 120 may receive images of the article 1 one by one from the photographing unit 10 in order, and the control unit 130 may perform learning and testing one by one in the order in which the images are received.

The controller 130 may perform learning and testing on each of the plurality of images one by one, and may perform a test on the next image by reflecting the previously learned results. For example, when learning and testing is performed on any one image (first image), the control unit 130 reflects the learning result for the corresponding image (first image) for the next image (second image). do the test In addition, the controller 130 may also perform learning on the next image (second image), and reflect the result when performing a test on the next image (third image).

In order for the control unit 130 to perform a test on an image, it should know in advance a test criterion (hereinafter referred to as a 'test criterion'), for example, the characteristics of a normal image (an image of a good product). These test criteria may be previously stored in the storage unit 140 , or the control unit 130 may directly grasp and store the images in the storage unit 140 in the process of learning the images.

Hereinafter, an embodiment in which the controller 130 directly grasps the test criterion in the process of learning the images will be described.

The controller 130 starts learning one by one for the plurality of images in order. At first, since the test standard is not prepared, the controller 130 only learns the image. If a predetermined criterion (hereinafter, referred to as a 'preparation criterion') is satisfied while learning the images one by one in order, the controller 130 determines that the preparation for setting the test criterion is complete, Both learning and testing are performed for the , but the test is performed by reflecting the learning results performed until the preparation criteria are satisfied and the learning results for the images in the subsequent order.

In detail, if it is determined that the preparation criterion is satisfied, the controller 130 sets the test criterion based on the learning result performed until the preparation criterion is satisfied, and then tests the images in the following order. In this case, the control unit 130 continuously updates the test criteria based on the results while also performing the test and learning for the images in the subsequent order. That is, when performing the test on the images in the subsequent order, the controller 130 performs the test by reflecting the learning results of the other images performed just before the test execution.

A method for the controller 130 to determine whether the preparation criterion is satisfied is as follows.

The controller 130 may determine whether the preparation criterion is satisfied based on the time the learning has been performed so far or the number of images for which the learning has been performed so far. This is because the test criteria such as the characteristic of a normal image can be grasped only when learning is performed for at least a certain time or learning is performed on at least a certain number of images.

For example, the controller 130 may check the time for which the learning has been performed so far, and if the checked time exceeds a preset value, it may determine that the preparation criterion is satisfied. Alternatively, for example, the controller 130 may check the number of images for which learning has been performed so far, and if the checked number exceeds a preset value, it may be determined that the preparation criterion is satisfied.

In addition, the controller 130 may determine whether the preparation for setting the test criterion is completed (whether the preparation criterion is satisfied) in various ways.

When the preparation criterion is satisfied, a method for the controller 130 to perform both learning and testing on images of a subsequent order is as follows.

If it is determined that the preparation criterion is satisfied, the controller 130 determines the characteristics of the normal image based on the learning result performed until the preparation criterion is satisfied, and sets it as the test criterion. Here, the test criterion is described as an example of a characteristic of a normal image, but otherwise, the characteristic of an abnormal image may be the test criterion.

The controller 130 performs a test on any one image (the first image) based on the identified characteristics of the normal image. That is, the controller 130 determines whether the first image is a normal image or an abnormal image. When it is determined that the first image is an abnormal image, the controller 130 may display a guide indicating that a defective product has been detected through the input/output unit 110 .

The controller 130 performs learning as well as a test on the first image. The controller 130 updates the characteristics of the normal image based on the result of learning the first image. Therefore, when performing a test on an image (second image) in the order following the first image, the controller 130 determines whether the second image is a normal image based on the characteristics of the normal image to which the learning result for the first image is reflected can be judged In this way, the controller 130 may continuously update the test criteria by performing a test and learning at the same time for each image.

Meanwhile, in the above, it has been described that the control unit 130 updates the test criteria (characteristics of a normal image) whenever the learning for one image is performed. may not be effective in Therefore, instead of updating the test criterion for each image, the test criterion may be updated only when a preset predetermined criterion (hereinafter, referred to as an 'update criterion') is satisfied. Hereinafter, such an embodiment will be described.

When it is determined that the preparation criterion is satisfied, the controller 130 determines the characteristics of the normal image based on the previously performed learning result. Here, the test criterion is described as an example of a characteristic of a normal image, but otherwise, the characteristic of an abnormal image may be the test criterion.

The controller 130 performs a test on any one image (the first image) based on the identified characteristics of the normal image. That is, the controller 130 determines whether the first image is a normal image or an abnormal image. When it is determined that the first image is an abnormal image, the controller 130 may display a guide indicating that a defective product has been detected through the input/output unit 110 .

The controller 130 performs learning as well as a test on the first image. When learning of the first image is completed, the controller 130 determines whether a preset update criterion is satisfied. In detail, the control unit 130 checks the learning time after the time when the characteristic identification of the normal image is completed or the time when the update of the characteristic of the normal image is completed, and if the checked time exceeds a preset value, the update criterion can be considered to be satisfied. Alternatively, the controller 130 checks the number of images for which learning is performed after the time point when the characterization of the normal image is completed or the time point when the update of the characteristics of the normal image is completed, and if the checked number exceeds a preset value, update it It can be judged that the criteria are satisfied.

In this way, the controller 130 reflects the learning result and updates the test criterion by reflecting the learning result only when a certain criterion (updating criterion) is satisfied, so that the efficiency can be improved while reflecting the learning result in real time.

Hereinafter, a method of performing machine learning and testing on a plurality of images according to embodiments will be described with reference to FIGS. 3 to 8 . The method for performing machine learning and testing according to the embodiments shown in FIGS. 3 to 8 includes steps that are time-series processed by the computing device 100 shown in FIGS. 1 and 2 . Therefore, even if omitted below, the content described above with respect to the computing device 100 shown in FIGS. 1 and 2 is also in the method for performing machine learning and testing according to the embodiments shown in FIGS. 3 to 8 . can be applied.

3 is a flowchart illustrating a method for performing machine learning and testing on a plurality of images according to an embodiment.

Referring to FIG. 3 , in step 301 , the computing device 100 receives a plurality of images. In this case, the computing device 100 may receive a plurality of images at once, or may receive target images one by one while performing learning and testing.

In step 302 , the computing device 100 performs both learning and testing on the plurality of images, but performs the test by reflecting the learning result. In other words, the computing device 100 performs a test on the next image by reflecting the learning result on the previous image while sequentially performing learning and testing on each of the plurality of images.

4 is a flowchart for explaining detailed steps included in step 302 of FIG. 3 . Referring to FIG. 4 , in step 401 , the computing device 100 performs learning on a plurality of received images one by one.

In step 402 , the computing device 100 determines whether a next image to be learned exists, ends the process if the next image does not exist, and proceeds to step 403 if the next image exists.

In step 403 , the computing device 100 determines whether a preset criterion (preparation criterion) is satisfied, returns to step 401 if the preparation criterion is not satisfied, and proceeds to step 404 if the preparation criterion is satisfied. Here, the 'preparation criterion' is a criterion for judging whether the preparation for setting the criterion (test criterion) for the test is completed as described above. That is, when the preparation criterion is satisfied, the computing device 100 determines that preparation for setting the test criterion is complete. The computing device 100 determines whether the preparation criterion is satisfied based on the time the learning has been performed so far or the number of images for which the learning has been performed so far, the detailed process will be described with reference to FIGS. 7 and 8 below. .

In step 404, the computing device 100 performs both learning and testing on the remaining images, but reflects the learning results of other images to perform the test.

If the computing device 100 determines that the preparation criterion is satisfied, both learning and testing are performed on the images in the subsequent order, but the learning result performed until the preparation criterion is satisfied and the learning result of the images in the subsequent order are reflected to perform the test. In detail, when it is determined that the preparation criterion is satisfied, the computing device 100 sets a test criterion based on the learning result performed until the preparation criterion is satisfied, and then tests the images in the following order. In this case, the computing device 100 continuously updates the test criteria based on the results while also performing testing and learning on the images in the subsequent order. That is, when the computing device 100 performs the test on the images in the subsequent order, the test is performed by reflecting the learning results of other images performed just before the test execution.

A detailed process in which the computing device 100 performs a test by reflecting the learning result of another image in step 404 will be described below with reference to FIGS. 5 and 6 .

5 is a flowchart for explaining detailed steps included in step 404 of FIG. 4 . Referring to FIG. 5 , in step 501 , the computing device 100 determines the characteristics of a normal image based on a previous learning performance result.

In operation 502 , the computing device 100 detects an abnormal image by performing a test on the image based on characteristics of the normal image. The computing device 100 may determine whether each image is a normal image or an abnormal image by comparing the characteristics of the normal image identified in step 501 with each image.

In step 503 , the computing device 100 performs learning on the image for which the test is performed in step 502 . In step 504 , the computing device 100 updates the characteristic of the normal image based on the result of performing the learning in step 503 .

In step 505 , the computing device 100 determines whether the next image exists, returns to step 502 if the next image exists, and ends the process if the next image does not exist.

On the other hand, as described above, in updating the characteristics of the normal image, the computing device 100 does not update whenever learning of one image is completed, but only updates when a preset condition (update condition) is satisfied. By doing this, it is possible to increase the efficiency while reflecting the learning results in real time. This embodiment will be described below with reference to FIG. 6 .

6 is a flowchart illustrating detailed steps included in step 404 of FIG. 4 . Referring to FIG. 6 , in step 601 , the computing device 100 determines the characteristics of a normal image based on the previous learning performance result.

In operation 602, the computing device 100 detects an abnormal image by performing a test on the image based on the characteristics of the normal image. The computing device 100 may determine whether each image is a normal image or an abnormal image by comparing the characteristics of the normal image identified in step 601 with each image.

In step 603 , the computing device 100 performs learning on the image for which the test is performed in step 602 . In step 604 , the computing device 100 determines whether there is a next image, and if there is no next image as a result of the determination, the process is terminated. Conversely, if there is a next image, the computing device 100 proceeds to step 605 .

In operation 605, the computing device 100 determines whether a preset criterion (update criterion) is satisfied. In detail, the computing device 100 checks a time at which learning is performed after a time point at which characterization of a normal image is completed or a time point at which an update on the characteristics of a normal image is completed, and updates if the checked time exceeds a preset value It can be judged that the criteria are satisfied. Alternatively, the computing device 100 checks the number of images for which learning has been performed after the time point at which the characteristic identification of the normal image is completed or the time point at which the update of the characteristic of the normal image is completed, and if the checked number exceeds a preset value It can be determined that the update criteria are satisfied.

As a result of the determination in step 605 , if it is determined that the preset criterion is not satisfied, the computing device 100 returns and performs step 602 again. Conversely, if it is determined that the preset criterion is satisfied, the computing device 100 proceeds to step 606 .

In step 606 , the computing device 100 updates the features of the normal image based on the result of learning the images after the features of the normal image were last updated. After performing step 606, the computing device 100 returns and performs step 602 again.

Hereinafter, a detailed method in which the computing device 100 determines whether the preparation criterion is satisfied in operation 403 of FIG. 4 will be described with reference to FIGS. 7 and 8 .

Referring to FIG. 7 , in step 701 , the computing device 100 checks the time for which learning has been performed so far. In step 702, the computing device 100 determines whether the checked time exceeds a preset value, and if it exceeds the preset value, proceeds to step 404 of FIG. 4, and if it is less than or equal to the preset value, step 401 of FIG. do it again

Referring to FIG. 8 , in step 801 , the computing device 100 checks the number of images for which learning has been performed so far. In step 802, the computing device 100 determines whether the checked number exceeds a preset value, and if it exceeds the preset value, proceeds to step 404 of FIG. 4, and if it is less than or equal to the preset value, step 401 of FIG. do it again

In this way, the computing device 100 identifies the standards for the test by performing learning on the images until the preparation criteria are satisfied, and performs learning on the subsequent images and simultaneously grasps the learning results up to the previous image. A test can be performed based on the established test criteria.

According to the embodiments described above, both learning and testing are performed on a plurality of images, but by performing the test by reflecting the learning result, the test can be performed by reflecting the learning result in real time, and thus the environment change It can be expected to have the effect of increasing the test accuracy by adaptively responding to the

In addition, since there is no need to separately prepare an image for learning, it is possible to reduce time and effort required to secure an image of a defective product or to classify an image.

The term '~ unit' used in the above embodiments means software or hardware components such as field programmable gate array (FPGA) or ASIC, and '~ unit' performs certain roles. However, '-part' is not limited to software or hardware. '~unit' may be configured to reside on an addressable storage medium or may be configured to refresh one or more processors. Accordingly, as an example, '~' indicates components such as software components, object-oriented software components, class components, and task components, and processes, functions, properties, and procedures. , subroutines, segments of program patent code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays, and variables.

The functions provided in the components and '~ units' may be combined into a smaller number of elements and '~ units' or separated from additional components and '~ units'.

In addition, components and '~ units' may be implemented to play one or more CPUs in a device or secure multimedia card.

The method for performing machine learning and testing according to the embodiments described with reference to FIGS. 3 to 8 may also be implemented in the form of a computer-readable medium for storing instructions and data executable by a computer. In this case, the instructions and data may be stored in the form of program codes, and when executed by the processor, a predetermined program module may be generated to perform a predetermined operation. In addition, computer-readable media can be any available media that can be accessed by a computer, and includes both volatile and nonvolatile media, removable and non-removable media. In addition, the computer-readable medium may be a computer recording medium, which is a volatile and non-volatile and non-volatile storage medium implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules, or other data. It may include both volatile, removable and non-removable media. For example, the computer recording medium may be a magnetic storage medium such as HDD and SSD, an optical recording medium such as CD, DVD, and Blu-ray disc, or a memory included in a server accessible through a network.

Also, the method for performing machine learning and testing according to the embodiments described with reference to FIGS. 3 to 8 may be implemented as a computer program (or computer program product) including instructions executable by a computer. The computer program includes programmable machine instructions processed by a processor, and may be implemented in a high-level programming language, an object-oriented programming language, an assembly language, or a machine language. . In addition, the computer program may be recorded in a tangible computer-readable recording medium (eg, a memory, a hard disk, a magnetic/optical medium, or a solid-state drive (SSD), etc.).

Accordingly, the method for performing machine learning and testing according to the embodiments described with reference to FIGS. 3 to 8 may be implemented by executing the above-described computer program by a computing device. The computing device may include at least a portion of a processor, a memory, a storage device, a high-speed interface connected to the memory and the high-speed expansion port, and a low-speed interface connected to the low-speed bus and the storage device. Each of these components is connected to each other using various buses, and may be mounted on a common motherboard or in any other suitable manner.

Here, the processor may process a command within the computing device, such as, for example, to display graphic information for providing a graphic user interface (GUI) on an external input or output device, such as a display connected to a high-speed interface. Examples are instructions stored in memory or a storage device. In other embodiments, multiple processors and/or multiple buses may be used with multiple memories and types of memory as appropriate. In addition, the processor may be implemented as a chipset formed by chips including a plurality of independent analog and/or digital processors.

Memory also stores information within the computing device. As an example, the memory may be configured as a volatile memory unit or a set thereof. As another example, the memory may be configured as a non-volatile memory unit or a set thereof. The memory may also be another form of computer readable medium, such as, for example, a magnetic or optical disk.

In addition, the storage device may provide a large-capacity storage space to the computing device. A storage device may be a computer-readable medium or a component comprising such a medium, and may include, for example, devices or other components within a storage area network (SAN), a floppy disk device, a hard disk device, an optical disk device, or a tape device, a flash memory, or other semiconductor memory device or device array similar thereto.

The above-described embodiments are for illustration, and those of ordinary skill in the art to which the above-described embodiments pertain can easily transform into other specific forms without changing the technical idea or essential features of the above-described embodiments. you will understand Therefore, it should be understood that the above-described embodiments are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a dispersed form, and likewise components described as distributed may be implemented in a combined form.

The scope to be protected through this specification is indicated by the claims described below rather than the detailed description, and it should be construed to include all changes or modifications derived from the meaning and scope of the claims and their equivalents. .

10: photographing unit 100: computing device
110: input/output unit 120: communication unit
130: control unit 140: storage unit

Claims (14)

In a method for performing machine learning and testing on a plurality of images,
receiving a plurality of images; and
For at least some of the plurality of images, both learning and testing are performed on each image, but learning is also performed on a specific image that has been tested, and the result of learning on the specific image is reflected performing a test on the image;
The step of performing the test is
determining whether a preset first criterion is satisfied while learning is performed on the received images one by one; and
If the determination result satisfies the first criterion, all of the learning and testing are performed on the remaining images, but the test is performed by reflecting the learning results performed until the first criterion is satisfied and the learning results for the remaining images; including,
The step of performing a test by reflecting the learning results performed until the first criterion is satisfied and the learning results for the remaining images,
identifying a characteristic of a normal image based on a learning result performed until the first criterion is satisfied;
determining whether the first image is a normal image based on the identified characteristics of the normal image;
performing learning on the first image;
determining whether a preset second criterion is satisfied; and
As a result of the determination, if the second criterion is satisfied, the characteristic of the normal image is updated based on a result of performing learning on the first image, and if the determination result does not satisfy the second criterion, based on the characteristic of the normal image A method, comprising: performing testing and learning on the next image. delete delete delete According to claim 1,
The step of determining whether the first criterion is satisfied,
Checking the time of performing the learning so far; and
and determining that the first criterion is satisfied if the checked time exceeds a preset time. According to claim 1,
The step of determining whether the first criterion is satisfied,
checking the number of images for which learning has been performed so far; and
and determining that the first criterion is satisfied if the checked number exceeds a preset number. A computer-readable recording medium in which a program for performing the method according to claim 1 is recorded. A computer program stored in a medium for performing the method according to claim 1 performed by an apparatus for performing machine learning and testing. A device for performing machine learning and testing, comprising:
an input/output unit for receiving manipulation and data input related to machine learning and testing, and displaying data processing results;
a communication unit for performing communication for transmitting and receiving data with an external device;
a storage unit in which a program for machine learning and test execution is stored; and
A control unit for performing machine learning and testing on a plurality of images by executing the program,
The control unit performs both learning and testing on each image for at least some of the plurality of images received through the communication unit, but also performs learning on the specific image on which the test is performed, and for the specific image The test is performed on the next image by reflecting the learning result,
In performing the test, the control unit
While learning is performed on the received images one by one, it is determined whether a preset first criterion is satisfied. The test is performed by reflecting the learning results performed up to the time and the learning results for the remaining images,
In performing the test, the control unit reflects the learning results performed until the first criterion is satisfied and the learning results for the remaining images,
Determine the characteristics of the normal image based on the learning result performed until the first criterion is satisfied, determine whether the first image is a normal image based on the identified characteristics of the normal image, and learn about the first image After performing , it is determined whether a preset second criterion is satisfied, and if the determination result satisfies the second criterion, the characteristic of the normal image is updated based on the result of learning the first image, and the result of the determination If the second criterion is not satisfied, testing and learning for the next image are performed based on the characteristics of the normal image. delete delete delete 10. The method of claim 9,
In determining whether the control unit satisfies the first criterion,
The apparatus of claim 1 , wherein a time for which learning has been performed so far is checked, and if the checked time exceeds a preset time, it is determined that the first criterion is satisfied. 10. The method of claim 9,
In determining whether the control unit satisfies the first criterion,
Checking the number of images for which learning has been performed so far, and determining that the first criterion is satisfied if the checked number exceeds a preset number.

Images