KR20200038775A - Method and apparatus for learning neural network using multi channel image - Google Patents

Abstract

Provided are a method and a device for training an artificial neural network by using a multi-channel image. The device for training an artificial neural network by using a multi-channel image can comprise: a control part that merges at least one training image used in training an artificial neural network, and trains the artificial neural network based on the generated merged image; and a memory storing the at least one training image.

Description

METHOD AND APPARATUS FOR LEARNING NEURAL NETWORK USING MULTI CHANNEL IMAGE}

Embodiments disclosed herein relates to a method and apparatus for learning an artificial neural network using a multi-channel image, and more particularly, to a method and apparatus for learning an artificial neural network by combining a plurality of images into one multi-channel image. .

Recently, with the development of computing technology, the technology of artificial neural networks has been developed, and artificial neural networks are applied in various fields.

In the existing manufacturing sector, a person directly judged a defect in order to find a defect in the product production line, but a technique for determining whether a product is defective is developed using machine vision using an artificial neural network.

To this end, the same object is photographed while changing the optical conditions (lighting, focal length, etc.) at the same location to create an image that finds features (scratches, stains, etc.) that appear only in special optical conditions, and uses the generated multiple images You have to learn artificial neural networks.

At this time, in order to recognize the features of the artificial neural network through various images, it is necessary to collect images by learning the images in the artificial neural network according to optical conditions.

However, this method has a problem in that it takes a lot of time to learn several artificial neural networks, and there is a problem in that computing power increases exponentially because several artificial neural networks must be operated simultaneously.

In addition, there is a problem that it takes a lot of time to analyze the learning results of each artificial neural network and decide how to collect them as a final result.

In relation to this, Korean Patent Publication No. 10-2000-0087346, which is a prior art document, relates to a method for learning and managing artificial intelligence of the Internet, and in particular, registers teachers and learners through the Internet, and the teacher creates questions and test questions to be used. It can be used, and the learner receives the questions contained in the problem database and learns them, and only obtains the contents that have been accurately evaluated the learning results, and cannot solve the above-mentioned problems.

Therefore, a technique for solving the above-described problems is needed.

On the other hand, the above-mentioned background technology is the technical information acquired by the inventor for the derivation of the present invention or acquired in the derivation process of the present invention, and is not necessarily a known technology disclosed to the general public before filing the present invention. .

The embodiments disclosed herein have an object to provide a method and apparatus for learning an artificial neural network using a multi-channel image.

The embodiments disclosed herein have an object to provide a method and apparatus for learning an artificial neural network that aggregates multiple learning images into a single multi-channel image.

 The embodiments disclosed herein have an object to present a method and apparatus for learning an artificial neural network that automatically learns characteristics of various channels of a learning image.

In the embodiments disclosed in the present specification, even if a dimension of an image used in prior learning differs from a dimension of an image used in additional learning, the parameters of the pre-trained artificial neural network are set to the dimension of the image used in additional learning. It is an object of the present invention to provide a method and apparatus for learning an artificial neural network that performs learning by controlling to fit.

As a technical means for achieving the above-described technical problem, according to an embodiment, in an artificial neural network learning apparatus using a multi-channel image, a combined image is generated by merging at least one learning image used for learning of the artificial neural network. And, based on the generated combined image may include a control unit for learning the artificial neural network and a memory for storing the at least one learning image.

According to another embodiment, in a method for an artificial neural network learning apparatus to learn an artificial neural network using a multi-channel image, generating a combined image by merging at least one learning image used for learning the artificial neural network And learning the artificial neural network based on the combined image.

According to another embodiment, a computer-readable recording medium in which a program for performing an artificial neural network learning method is recorded, generating and combining a combined image by combining at least one learning image used for learning the artificial neural network And learning the artificial neural network based on the combined image.

According to another embodiment, a computer program performed by an artificial neural network learning apparatus and stored in a recording medium to perform an artificial neural network learning method, wherein at least one learning image used for learning the artificial neural network is merged to form a combined image. It may include a step of generating and learning the artificial neural network based on the generated combined image.

According to any one of the above-described problem solving means, an artificial neural network learning method and apparatus using a multi-channel image may be presented.

According to any one of the above-described problem solving means, by combining a plurality of learning images into a single multi-channel image, it is possible to present an artificial neural network learning method and apparatus capable of learning an artificial neural network without losing data for each image. .

According to any one of the above-described problem solving means, it is possible to present an artificial neural network learning method and apparatus capable of rapidly learning a plurality of learning images simultaneously through one combined image combining characteristics of various channels of the learning image.

According to any one of the above-described problem solving means, even if the dimension of the image used in the prior learning and the dimension of the image used in the additional learning are different, the parameters of the pre-trained artificial neural network are controlled to learn images having different dimensions. This possible artificial neural network learning method and apparatus can be presented.

Effects obtained in the disclosed embodiments are not limited to the above-mentioned effects, and other effects not mentioned are apparent to those skilled in the art to which the embodiments disclosed from the following description belong. It can be understood.

1 is a block diagram showing an artificial neural network learning apparatus according to an embodiment.
2 is a flow chart for explaining an artificial neural network learning method according to an embodiment.

Hereinafter, various embodiments will be described in detail with reference to the accompanying drawings. The embodiments described below may be embodied in various different forms. In order to more clearly describe the features of the embodiments, detailed descriptions of the matters well known to those skilled in the art to which the following embodiments pertain are omitted. In the drawings, parts irrelevant to the description of the embodiments are omitted, and like reference numerals are assigned to similar parts throughout the specification.

Throughout the specification, when a component is "connected" to another component, this includes not only "directly connected" but also "connected with other components in between". In addition, when a configuration is said to "include" a configuration, this means that unless otherwise stated, other configurations may be excluded and other configurations may be further included.

Hereinafter, exemplary embodiments will be described in detail with reference to the accompanying drawings.

Before explaining this, however, the meanings of terms used below are defined first.

'Channel' is a component constituting the vector value of each pixel on the image, and each pixel may be composed of at least one channel according to characteristics.

'Artificial neural network' is an information processing technology that complexly controls inputs and outputs in detail by engineeringly imitating the advanced information processing mechanism of the biological nervous system, and based on the input layer that sends signals from switches or sensors. It is a network in which three types of neuron (neural cell) models are composed of a hidden layer that adjusts the mutual relationship while prioritizing inputs and outputs, and an output layer that calculates and outputs the necessary control amount based on this.

In addition, each 'layer' constituting the artificial neural network may be composed of at least one 'node'. And the nodes of each layer can form a connection relationship with the nodes of the next layer.

The 'parameter' is a value determined through learning of the artificial neural network, and a connection relationship between layers or nodes constituting the artificial neural network may be controlled.

Meanwhile, in the following description, it is assumed that the artificial neural network is determined by performing learning using a learning image in advance.

In addition to the terms defined above, terms that require explanation will be described separately below.

The artificial neural network learning apparatus 10 to be described below may train an artificial neural network to inspect or measure an object using an image obtained by being coupled to or connected to an inspection device using, for example, a machine vision.

Here, machine vision means to automate the industry through cameras (visual recognition), CPUs, and SWs, in order to inspect or measure objects, in place of existing methods that humans judge with the naked eye.

That is, the artificial neural network learning apparatus 10 may train the artificial neural network by using at least one learning image so that the artificial neural network can recognize the image acquired through the machine non-dedicated camera.

And the learned artificial neural network matches the pattern of the object from the image acquired by the camera, fits points, lines or faces, distinguishes colors, or measures objects (gauging). It provides location information for guides, reads 1D and 2D barcodes displayed on objects, or performs optical character reading (OCR). Through this process, an object can be inspected, measured, or read to realize industrial automation.

1 is a configuration diagram for explaining an artificial neural network learning apparatus 10 according to an embodiment.

The artificial neural network learning apparatus 10 may be embodied as a computer or a portable terminal, a television, a wearable device, or the like that connects to a remote server through a network N or connects to other terminals and servers. Here, the computer includes, for example, a laptop equipped with a web browser (WEB Browser), a desktop (desktop), a laptop (laptop), and the like, and the portable terminal is, for example, a wireless communication device that guarantees portability and mobility. , PCS (Personal Communication System), PDC (Personal Digital Cellular), PHS (Personal Handyphone System), PDA (Personal Digital Assistant), GSM (Global System for Mobile communications), IMT (International Mobile Telecommunication) -2000, CDMA (Code All kinds of handhelds such as Division Multiple Access (2000), W-Code Division Multiple Access (W-CDMA), Wireless Broadband Internet (Wibro), Smart Phone, Mobile Worldwide Interoperability for Microwave Access (WiMAX), etc. (Handheld) -based wireless communication device. In addition, the television may include Internet Protocol Television (IPTV), Internet Television (TV), terrestrial TV, and cable TV. Furthermore, the wearable device is a type of information processing device that can be directly worn on the human body, for example, a watch, glasses, accessories, clothing, shoes, etc., or connects to a remote server through a network directly or through another information processing device or other terminal. And can be connected.

Referring to FIG. 1, the artificial neural network learning apparatus 10 according to an embodiment may include an input / output unit 110, a control unit 120, a communication unit 130, and a memory 140.

The input / output unit 110 may include an input unit for receiving input from a user, and an output unit for displaying information such as a result of performing a task or a state of the artificial neural network learning apparatus 10. For example, the input / output unit 110 may include an operation panel for receiving user input, a display panel for displaying the screen, and the like.

Specifically, the input unit may include devices capable of receiving various types of user input, such as a keyboard, a physical button, a touch screen, a camera, or a microphone. Also, the output unit may include a display panel or a speaker. However, the present invention is not limited thereto, and the input / output unit 110 may include a configuration supporting various input / output.

The control unit 120 controls the overall operation of the artificial neural network learning device 10, and may include a processor such as a CPU. The control unit 120 may control other components included in the artificial neural network learning apparatus 10 to perform an operation corresponding to a user input received through the input / output unit 110.

For example, the controller 120 may execute a program stored in the memory 140, read a file stored in the memory 140, or store a new file in the memory 140.

The controller 120 may generate a combined image by merging at least one learning image used for learning of an artificial neural network.

According to an embodiment, the controller 120 may extract at least one learning image to be trained in the artificial neural network from a file including at least one learning image.

For example, the controller 120 may extract a learning image from a file including a plurality of learning images based on user input, and the number of extracted learning images and each learning image may be color or grayscale. (Grayscale) can be identified.

Also, according to an embodiment, the controller 120 may identify the number of channels of the extracted at least one learning image.

According to an embodiment, when at least one learning image is in a pixel format, the controller 120 may identify the number of channels constituting a pixel value on each learning image for each at least one training image.

For example, the controller 120 may identify the number of channels of the learning image by identifying an RGB channel, which is a channel of the learning image, and an alpha channel indicating transparency, based on the pixel values on the learning image.

According to another embodiment, when the at least one learning image is in a planar format, the controller 120 may identify the number of channels through the number of planners of each learning image for each at least one training image.

For example, the controller 120 may identify a planner of an R channel on a learning image, a planar of a G channel, and a planner of a B channel.

At this time, the controller 120 may identify the number of channels for other image format formats other than the image format format of the above-described embodiment. For example, the controller 120 may identify the format format of the image based on the metadata of the image, and identify the channels constituting the data according to the identified format format, thereby identifying the number of channels of the learning image. .

Thereafter, the controller 120 may generate a multi-channel combined image by merging each channel of at least one learning image.

For example, the controller 120 may have values (1, 4, 6) for each of the 3 channels of the first pixel on the first learning image and values for each of the 3 channels of the first pixel on the second learning image ( 5, 7, 10) can be combined in parallel to generate a combined image including first pixel values (1, 4, 6, 5, 7, 10) composed of information about 6 channels.

Alternatively, for example, the controller 120 may combine the R channel planner of the first learning image and the R channel planner of the second learning image to generate a combined image including one R channel planner.

Through this, the controller 120 may generate a combined image in which data for each channel of each of the plurality of learning images is not lost.

In addition, the controller 120 may train the combined image to the artificial neural network, and according to an embodiment, may reconstruct the artificial neural network according to the number of channels of the combined image.

For example, the controller 120 may reconstruct the size of the artificial neural network so that the combined image can be input by comparing the number of channels input to the input layer with the artificial neural network and the number of channels of the combined image.

At this time, the controller 120 may use a preset parameter of the artificial neural network before reconstruction to the reconstructed artificial neural network.

For example, the controller 120 may use a preset parameter of the artificial neural network before reconstruction for a parameter corresponding to the reconstructed artificial neural network, and additional parameters not corresponding to the preset parameter may input a random initialization value. .

In addition, the controller 120 may train the combined image on the reconstructed artificial neural network, and may set the parameters of the reconstructed artificial neural network through learning of the combined image.

For example, the controller 120 may input a combined image into the reconstructed artificial neural network, and among the parameters of the reconstructed artificial neural network through a learning process of the reconstructed artificial neural network using the combined image, a preset parameter of the artificial neural network before reconstruction. You can update or set additional parameters.

Through this, it is possible to shorten the learning time by not reconfiguring all the parameters of the artificial neural network by reconfiguring the existing learned artificial neural network for further learning.

Thereafter, the controller 120 may recognize the image input to the artificial neural network where the combined image has been learned.

For example, if the number of input channels of the artificial neural network is different from the number of channels of the image, the controller 120 may output an error message indicating that image recognition has failed.

Or, for example, if the number of input channels of the artificial neural network is the same as the number of channels of the image, the controller 120 may input data for each channel into the input layer of the artificial neural network, and output a result recognized by the artificial neural network. can do.

The communication unit 130 may perform wired / wireless communication with other devices or networks. To this end, the communication unit 130 may include a communication module supporting at least one of various wired and wireless communication methods. For example, the communication module may be implemented in the form of a chipset.

The wireless communication supported by the communication unit 130 may be, for example, Wi-Fi (Wireless Fidelity), Wi-Fi Direct, Bluetooth (Bluetooth), UWB (Ultra Wide Band), or NFC (Near Field Communication). In addition, the wired communication supported by the communication unit 130 may be, for example, USB or HDMI (High Definition Multimedia Interface).

Various types of data such as files, applications, and programs may be installed and stored in the memory 140. The controller 120 may access and use data stored in the memory 140, or may store new data in the memory 140. Also, the control unit 120 may execute a program installed in the memory 140.

The memory 140 may store a file including at least one learning image.

2 is a flowchart illustrating an artificial neural network learning method according to an embodiment.

The artificial neural network learning method according to the embodiment shown in FIG. 2 includes steps that are processed in time series in the artificial neural network learning apparatus 10 shown in FIG. 1. Therefore, even if omitted below, the contents described above with respect to the artificial neural network learning apparatus 10 shown in FIG. 1 can be applied to the artificial neural network learning method according to the embodiment shown in FIG. 2.

First, the artificial neural network learning apparatus 10 may generate a combined image by merging at least one learning image used for learning of an artificial neural network (S2001).

To this end, the artificial neural network learning apparatus 10 may acquire at least one learning image, and may identify the number of channels for each learning image.

For example, the artificial neural network learning apparatus 10 may extract a learning image to be trained into the artificial neural network according to a user's input from a file including at least one learning image, and learn based on metadata of the extracted learning image The format of the image can be identified. Also, the artificial neural network learning apparatus 10 may identify the number of channels of the learning image by counting the arrangement of data according to the format format of the identified learning image.

In addition, the artificial neural network learning apparatus 10 may generate a multi-channel combined image by combining each channel of at least one learning image in parallel.

For example, the artificial neural network learning apparatus 10 combines RGB values, which are three channels of the first learning image, and CMYK values, which are four channels of the second learning image, for two training images of the pixel format, and generates seven. A combined image with channels can be created.

And the artificial neural network learning apparatus 10 may determine whether to reconstruct the artificial neural network based on the number of channels of the combined image (S2002), and if the number of channels of the combined image and the number of channels at the entrance of the artificial neural network is different, It can be reconstructed (S2003).

To this end, the artificial neural network learning apparatus 10 may determine whether the combined image generated in step S2001 can be input to the pre-trained artificial neural network, and reconstruct the artificial neural network according to the number of channels of the combined image.

According to an embodiment, when the number of channels that can be input to the input layer of the artificial neural network is less than the number of channels of the combined image, the artificial neural network learning apparatus 10 reconstructs to expand the size of the artificial neural network, and the channel of the combined image is input to the input layer. It can be done.

At this time, the artificial neural network learning apparatus 10 may use a preset parameter of the artificial neural network among parameters of the artificial neural network reconstructed in step S2003.

For example, the artificial neural network learning apparatus 10 may use parameters of an existing artificial neural network while performing reconstruction to expand the size of the artificial neural network. In addition, in the case of additional parameters added to the reconstructed artificial neural network, the artificial neural network learning apparatus 10 may set values of additional parameters through learning of a combined image to be described later.

According to another embodiment, if the number of channels that can be input to the input layer of the artificial neural network is greater than or equal to the number of channels of the combined image, the artificial neural network learning apparatus 10 reconstructs to reduce the size of the artificial neural network or converts the number of channels of the combined image. The channel of the combined image can be input to the input layer.

For example, the artificial neural network learning apparatus 10 may reduce the size of the artificial neural network to delete some parameters, or merge a dummy channel consisting of 0, which is a meaningless value in the combined image, and input the artificial neural network.

In addition, the artificial neural network learning apparatus 10 may train the artificial neural network reconstructed in step S2003 using the combined image generated in step S2001 (S2004).

For example, the artificial neural network learning apparatus 10 may input a combined image to the reconstructed artificial neural network in step S2002 to perform learning, and add the parameters of the artificial neural network reconstructed through learning, except for the preset parameter. The parameters can be set through learning the combined image.

Meanwhile, the artificial neural network learning apparatus 10 may recognize an image through an artificial neural network in which learning is completed.

To this end, the artificial neural network learning apparatus 10 may identify the number of channels of the image to be recognized and identify whether the number of channels of the combined image learned in the artificial neural network in step S2004 is the same.

For example, if the number of channels is different, the artificial neural network learning apparatus 10 may output an error message for image input.

Or, for example, if the number of channels is the same, the artificial neural network learning apparatus 10 may input an image through an input layer of the artificial neural network, and provide a result value recognized by the artificial neural network.

Through steps S2001 to S2004, the artificial neural network learning apparatus 10 may train an artificial neural network to inspect or measure an object using an image acquired through a machine non-dedicated camera, and the artificial neural network is a machine non-dedicated camera. Match the object's pattern from the image obtained through, fitting points, lines or faces, distinguishing colors, gauging objects, and providing location information for robot guides Or, it may read 1D and 2D barcodes displayed on objects or perform optical character reading (OCR). Through this process, objects can be inspected, measured, or read to enable industrial automation.

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

The functionality provided within components and '~ units' may be combined into a smaller number of components and '~ units', or separated from additional components and '~ units'.

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

The artificial neural network learning method according to the embodiment described with reference to FIG. 2 may also be implemented in the form of a computer-readable medium that stores instructions and data executable by a computer. At this time, instructions and data may be stored in the form of program code, and when executed by a processor, a predetermined program module may be generated to perform a predetermined operation. Also, the computer-readable medium can be any available medium 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 volatile and non-volatile implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. It can include both volatile, removable and non-removable media. For example, computer recording media are magnetic storage media such as HDDs and SSDs, optical recording media such as CDs, DVDs and Blu-ray Discs, or are accessible over a network. It may be a memory included in the server.

In addition, the artificial neural network learning method according to the embodiment described with reference to FIG. 2 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, object-oriented programming language, assembly language, or machine language. . In addition, the computer program may be recorded on a tangible computer-readable recording medium (eg, memory, hard disk, magnetic / optical medium, or solid-state drive (SSD), etc.).

Therefore, the artificial neural network learning method according to the embodiment described with reference to FIG. 2 may be implemented by executing a computer program as described above 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 a 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 mounted in other suitable ways.

Here, the processor is capable of processing instructions within the computing device, such as for displaying graphical information for providing a graphical user interface (GUI) on an external input or output device, such as a display connected to a high-speed interface. Examples include instructions stored in memory or storage devices. In other embodiments, multiple processors and / or multiple buses may be used with multiple memories and memory types 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 computing devices. In one example, the memory may be comprised of volatile memory units or a collection thereof. As another example, the memory may be composed of non-volatile memory units or a collection thereof. The memory may also be other types of computer readable media, such as magnetic or optical disks.

And the storage device can provide a large storage space for the computing device. The storage device may be a computer readable medium or a configuration including such a medium, and may include, for example, devices within a storage area network (SAN) or other configurations, and may include floppy disk devices, hard disk devices, optical disk devices, Or a tape device, flash memory, or other similar semiconductor memory device or device array.

The above-described embodiments are for illustration only, and those skilled in the art to which the above-described embodiments belong can easily be modified into other specific forms without changing the technical idea or essential characteristics 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 distributed manner, and similarly, components described as distributed may be implemented in a combined form.

The scope to be protected through the present specification is indicated by the claims, which will be described later, rather than the detailed description, and should be interpreted to include all modified or modified forms derived from the meaning and scope of the claims and equivalent concepts thereof. .

10: Artificial neural network learning device
110: input unit
120: control unit
130: communication unit
140: memory

Claims (12)

In the artificial neural network learning apparatus using a multi-channel image,
A control unit for generating a combined image by merging at least one learning image used for learning the artificial neural network, and learning the artificial neural network based on the combined image; And
And a memory for storing the at least one learning image. According to claim 1,
The control unit,
An artificial neural network learning apparatus that identifies a channel of each learning image for each of the at least one learning image, and combines the identified channels in parallel to generate a multi-channel combined image. According to claim 1,
The control unit,
An artificial neural network learning apparatus for reconstructing the artificial neural network using preset parameters of the artificial neural network according to the number of channels of the combined image. The method of claim 3,
The control unit,
An artificial neural network learning apparatus for learning the combined image on a reconstructed artificial neural network including the preset parameter. The method of claim 4,
The control unit,
An artificial neural network learning apparatus for setting parameters of the reconstructed artificial neural network through learning of the combined image. In the method of learning the artificial neural network using the multi-channel image artificial neural network learning device,
Generating a combined image by merging at least one learning image used for learning the artificial neural network; And
And learning the artificial neural network based on the generated combined image. The method of claim 6,
The step of generating the combined image,
Identifying a channel of each learning image for each of the at least one learning image; And
And combining the identified channels in parallel to generate a multi-channel combined image. The method of claim 6,
The step of learning the artificial neural network,
And reconstructing the artificial neural network using preset parameters of the artificial neural network according to the number of channels of the combined image. The method of claim 8,
The artificial neural network learning method,
And learning the combined image on a reconstructed artificial neural network including the preset parameter. The method of claim 9,
The step of learning the combined image,
And setting parameters of the reconstructed artificial neural network through learning of the combined image. A computer-readable recording medium on which a program for performing the method according to claim 6 is recorded. A computer program carried out by an artificial neural network learning device and stored in a medium to perform the method according to claim 6.

Images