A training image generation method used for training an image-based artificial intelligence model that analyzes images obtained from multi-channel one-dimensional signals, and an apparatus for executing the same

Abstract

The embodiment relates to a training image generating method and an apparatus for performing the same, which generate a training signal based on source signal information, select at least one output format from a plurality of preset output formats to determine an output format of a training image, determine an output interval for each channel of the training signal based on a length of a time interval of a waveform of the determined output format, select a scale for each axis to determine a grid scale of the training image, display a grid pattern on a two-dimensional plane according to the determined grid scale, set a reference position of a waveform content of the training signal based on at least one of the determined output interval for each channel and the determined grid scale, and display the waveform content and signal indicator of the training signal on the two-dimensional plane on which the grid pattern is displayed.

Description

Technical Field

[0001] The embodiments relate to a training image generation technique, for example, a training image generation technique used to train an image-based artificial intelligence model that analyzes an image obtained from a multi-channel one-dimensional signal and can generate a large number of training images with various image formats and signal patterns, and an apparatus for executing the same.

Background Art

[0002] Signals such as biosignals such as electrocardiograms and electroencephalograms are generally measured in the form of one-dimensional signals. At this time, the measured signal is embodied as a numerical value on the time domain measured in the channel and is generally represented in a numerical array structure of the C×T form. The measured numerical values are stored and utilized in the above-described numerical array structure.

[0003] Since the signal data numerical values are difficult for a human reader to understand as they are, they are displayed as a two-dimensional image output in which waveforms in a form easy for humans to understand are drawn for each channel on paper or on the screen of a device.

[0004] However, the form of the two-dimensional image output is diverse. In particular, even for two-dimensional image outputs used for the same purpose, the forms may be very different depending on the product.

[0005] Suppose that an artificial intelligence model for analyzing signal information shown as a two-dimensional image output is developed. If this artificial intelligence model is trained using a training image of one output form, the artificial intelligence model should not be able to analyze target images of other output forms and should have low analysis performance.

[0006] Therefore, in order for the artificial intelligence model to learn to have high analysis performance, it is necessary to prepare training images in as diverse output formats as possible. However, it is extremely difficult in reality to prepare a large number of training images for sufficient learning with only existing images of existing products in use. In particular, there is a limit that it is impossible to prepare at all for future changes in which images of new output formats are utilized in the release of new products or the update of existing products.

Summary of the Invention

Problems to be Solved by the Invention

[0007] On one aspect, in an exemplary embodiment of the present application, in order to solve the problems that occur when training a two-dimensional image with only one format without considering the diversity of formats, a large-scale training image with various changes in image format and signal pattern can be generated. A training image generation method and apparatus, for example, a training image generation technology used to train an image-based artificial intelligence model that analyzes an image obtained from a multi-channel one-dimensional signal, a method for executing the same, and an apparatus are provided for this purpose.

Means for Solving the Problems

[0008] In an embodiment of the present application, in a training image generation method used to train an image-based artificial intelligence model that analyzes an image obtained from, for example, a multi-channel one-dimensional signal and is executed by a computing device including a processor and a memory, a step of generating a training signal based on source signal information including a multi-channel one-dimensional signal; a step of selecting at least one output format from a plurality of preset output formats to determine the output format of the training image; a step of determining a per-channel output section of the training signal based on the length of the time interval of the waveform of the determined output format; a step of selecting a scale for each axis to determine the grid scale of the training image; a step of representing a grid pattern on a two-dimensional plane according to the determined grid scale; a step of setting a reference position of the waveform content of the training signal based on at least one of the determined per-channel output section and the determined grid scale; and a step of representing the waveform content and signal identifier of the training signal on a two-dimensional plane on which the grid pattern is represented. A training image generation method and a computer-readable recording medium storing a program for executing the same, or a computer program stored in a computer-readable recording medium, are provided.

[0009] In addition, in an embodiment of the present application, there is provided a training image generation apparatus including an acquisition unit that acquires a source signal; and an image generation unit that includes a processor and a memory, wherein the image generation unit receives the source signal information received by the acquisition unit and executes the above-described training image generation method.

Advantages of the Invention

[0010] On one hand, according to the embodiment of the present application, it is possible to supply a training data set used to develop an artificial intelligence that generates two-dimensional signal images in various forms from a source signal such as a multi-channel one-dimensional signal and analyzes the two-dimensional signal images with excellent performance.

[0011] By utilizing the two-dimensional signal images in various forms, problems that occur when training with only one form without considering the diversity of such forms can be solved, such as out-of-distribution problems and problems where analysis goes wrong even with a slight change in form.

[0012] The invention according to the embodiments of the present application can be extended and applied not only to signals in the medical and biological fields but also to signal analysis in other industrial fields that analyze two-dimensional images based on one-dimensional signals.

[0013] The effects of the present application are not limited to those mentioned above, and other effects not mentioned should be clearly understood by those skilled in the art from the description of the claims.

Brief Description of the Drawings

[0014] To more clearly explain the exemplary embodiments of the present application, the drawings necessary for the description of the embodiments are briefly introduced below. It should be understood that the following drawings are only for the purpose of explaining the embodiments of this specification and not for the purpose of limitation. Also, for the sake of clarity of the description, some elements to which various deformations such as exaggeration and omission are applied may be shown in the following drawings.

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Mode for Carrying Out the Invention

[0015] Hereinafter, some embodiments of the present application will be described in detail with reference to exemplary drawings. When assigning reference numerals to the components of each drawing, for the same components, the same numerals may be assigned as much as possible even if they are shown on different drawings. Further, in describing the present embodiment, when it is determined that a specific description of a related known configuration or function may obscure the gist of the present technical idea, the detailed description thereof may be omitted.

[0016] In this specification, when terms such as "including", "having", "consisting of", etc. are used, unless "only ~" is used, other parts may be added. When a component is expressed in the singular, it may include the case of including a plurality, unless otherwise explicitly stated.

[0017] Also, when describing the components of the present application, terms such as first, second, A, B, (a), (b), etc. may be used. Unless otherwise explicitly stated, these terms are merely used to distinguish the components from other components, and the essence, order, procedure, number, etc. of the components are not limited by these terms.

[0018] In this specification, "learning" or "learning" is a term that refers to executing machine learning through computing according to a procedure.

[0019] As used herein, the network refers to the neural network of a machine learning algorithm or model.

[0020] As used herein, terms such as "unit", "module", "device", or "system" are intended to refer to not only hardware but also a combination of software driven by the hardware. For example, the hardware may be data processing equipment including a CPU (Central Processing Unit), a GPU (Graphic Processing Unit), or other processors. Also, the software may refer to a running process, an object, an executable file, a thread of execution, a program, etc.

[0021] As used herein, multi-channel means one or two or more channels, and is defined not to exclude the case of one channel.

[0022] In a specific embodiment, the training image generation method used to train an image-based artificial intelligence model that analyzes a two-dimensional image obtained from a multi-channel one-dimensional signal may be executed by a computing device including at least one processor and a memory.

[0023] The computing device is configured to receive signal data from an external device (e.g., a signal measurement device) and process the same.

[0024] FIG. 1 is a flowchart of a training image generation method used to train an image-based artificial intelligence model that analyzes an image obtained from a multi-channel one-dimensional signal according to one aspect of the present application.

[0025] Referring to FIG. 1, the training image generation method (hereinafter, the training image generation method) includes a step (S100) of generating / modulating a training signal based on source signal information.

[0026] FIG. 2 is a schematic diagram of source signal information according to an embodiment of the present application, and FIG. 3 is a schematic diagram of a training signal according to an embodiment of the present application.

[0027] Referring to FIG. 2, the source signal information may include one-dimensional signals for each channel measured through one or more channels worn on the body of the subject. For example, the input source signal may be an array in the form of channel × time (C×T). The source signal may have a numerical array. As shown in FIG. 2, the source signal information may be composed of 12 one-dimensional electrocardiogram signals measured through 12 lead channels.

[0028] The source signal information may include an analog source signal and measured values of the source signal (for example, digital values).

[0029] In addition, the source signal information may further include source additional information other than signals such as subject information of the source signal (identification information, age, gender, read information, etc.), source scale identifier, source channel identifier, and source grid scale.

[0030] Such a source signal is utilized as a source for generating a training signal. A plurality of training signals may be generated from a single source signal.

[0031] In one embodiment, the step (S100) may include a step of selecting at least one transformation function from a plurality of preset transformation functions; and a step of transforming the source signal into a training signal using the selected transformation function.

[0032] A transformation function is a function that transforms an input signal into another signal. Such signal transformation may be treated as a preprocessing step in the training image generation process. The transformation function may be, for example, a signal preprocessing function aimed at various forms of noise removal or random modification of the input signal itself.

[0033] Each of the plurality of transformation functions consists of one or more transformation elements that execute the transformation function. In some embodiments, the transformation element may indicate transforming the signal attribute of the input signal. One or more transformation elements that make up each of n (n is a natural number greater than or equal to 1) transformation functions (f n 1 ) are predefined and stored. Incidentally, here, the subscript n is a character indicating the existence of n functions, and the superscript is a number for distinguishing from other functions described later.

[0034] In a specific embodiment, the transformation element vector (P n 1 ) of the transformation function (f n 1 ) may include m factors, and each factor is indicated by the transformation element P nm 1 (where n is the transformation function identifier and m is the element identifier). The transformation elements (P nm 1 ) may each correspond to a unique input information processing attribute. When a signal is input to the transformation function (f n 1 ), each transformation element (P nm 1 ) becomes the set value of the signal processing attribute and the signal is transformed. For example, when the transformation function (f n 1 ) is a noise removal function, the transformation elements (P nm 1 ) correspond to the signal processing attributes that define (set) certain characteristics of the noise removal operation.

[0035] The signal processing attributes may include, for example, a method of processing, modifying, or removing (replacing the signal magnitude with a fixed numerical value such as 0) the signal magnitude, waveform, frequency range, frequency distribution, signal rising time, time range, and / or other signal attributes. And this may be performed separately for each channel, or may be commonly applied to a channel group or the entire channels.

[0036] In a specific embodiment, the deformation elements of the deformation function may be embodied by hyperparameters. When one deformation function includes a plurality of deformation elements, the plurality of deformation elements may be represented by a hyperparameter vector. Here, the vector value is the deformation element value. Then, each deformation function may be associated with its own hyperparameter vector.

[0037] Such a deformation function may be used to generate a training signal (S100). Each of the plurality of deformation functions (f n 1 ) may receive a source signal (for example, an electrocardiogram signal, X) and generate a deformed signal X' = f n 1 (X, P nm 1 ). Also, when the deformation element value is changed, the signal attribute is embodied by the changed value in a corresponding manner, and finally a signal with the changed signal attribute is generated (S100).

[0038] In one embodiment, the step of selecting the at least one deformation function may select any one of the plurality of deformation functions according to a preset first-1 probability distribution for the set of the plurality of deformation functions, or may select two or more of the plurality of deformation functions.

[0039] The first-1 probability distribution defines the probability of selecting a specific deformation function (which may be plural) to generate a training image from the entire plurality of deformation functions.

[0040] In one embodiment, the first-first probability distribution may be defined by a method of selecting one deformation function out of n. For example, the first-first probability distribution may have a multinomial distribution. In this case, the source signal is singly deformed.

[0041] In another embodiment, the first-first probability distribution may be defined by a method of selecting two or more deformation functions out of n. For example, the first-first probability distribution may have a binomial distribution. In this case, different conversions may be repeatedly selected and used to generate a training signal. Then, the source signal is multiply deformed using a number of selected deformation functions.

[0042] In an alternative embodiment, the first-first probability distribution may be one obtained by optimizing an arbitrarily specified initial probability distribution through training. This will be described in more detail below.

[0043] The selected deformation function may be used as is, or the deformation element values of the selected deformation function may be changed to generate a training signal (S100). The selected deformation function may be input with new values for at least one deformation element as input arguments to adjust the function of the deformation function.

[0044] In one embodiment, the step of deforming the source signal into a training signal using the selected deformation function may include: changing at least one deformation element value among the deformation elements constituting the selected deformation function to a new value; and generating a signal reflecting the signal attributes changed according to the changed deformation element value as the training signal. The change of the deformation element value may be performed based on a probability distribution.

[0045] In one embodiment, first and second probability distributions may be set for each individual deformation element. When there are m deformation elements, m first and second probability distributions are defined and assigned. For each of the deformation elements of the deformation function selected according to the first probability distribution, the existing deformation element values may be changed to new values selected according to the first and second probability distributions preset for themselves. The changed new values are reflected in the source signal to generate a new training signal.

[0046] The first and second probability distributions define a probability distribution in which individual values are specified from the entire range of values that each deformation element of the deformation function can have. In a specific embodiment, the first and second probability distributions may include a Gaussian (normal) distribution, a gamma distribution, an exponential distribution, a uniform distribution, a chi-square distribution as a continuous probability distribution, a binomial distribution, a negative binomial distribution, an initial lower distribution, a Poisson distribution as a discrete probability distribution, and a multinomial distribution and / or a multivariate distribution for a number of deformation elements. The first and second probability distributions may be set to be the same or different for each deformation element. Also, in addition to the parametric approach method of extracting deformation elements in a state where the probability distribution of the specific form described above is defined, deformation elements may be randomly extracted without assuming a specific distribution.

[0047] In one embodiment, when the deformation element is embodied in numerical data, the first and second probability distributions may be a Gaussian (normal) distribution, a gamma distribution, an exponential distribution, a uniform distribution, or a chi-square distribution, which are continuous probability distributions. Then, the new value of the deformation element is a value selected from the continuous probability distribution.

[0048] In one embodiment, when the deformation element is embodied in binary variable data, the first and second probability distributions may be a binomial distribution, a negative binomial distribution, an initial lower distribution, or a Poisson distribution, which are discrete probability distributions. The binary variable data may be expressed by a first binary value indicating "yes" or a second binary value indicating "no". The new value of the deformation element is a value selected according to the discrete probability distribution.

[0049] In one embodiment, when the deformation element is embodied as a categorical variable, the first to second probability distributions may be a multinomial probability distribution. Then, the new value of the deformation element is a value selected according to the multinomial probability distribution.

[0050] In some embodiments, the set of deformation elements forming the deformation function may include at least some deformation elements having a correlation relationship with each other. When the deformation element changed to a new value has a correlation relationship with other deformation elements, the first to second probability distributions set for the other deformation elements having the correlation relationship may be a multivariate probability distribution. Then, the values of the other deformation elements having the correlation relationship are changed to new values selected according to a predefined multivariate probability distribution.

[0051] And in addition to the parametric approach method of extracting the deformation element in the state where the probability distribution of the specific form is defined as described above, the deformation element may be randomly extracted without assuming a specific distribution.

[0052] In an alternative embodiment, the first to second probability distributions may be those obtained by optimizing an arbitrarily specified initial probability distribution through training. This will be described in more detail below.

[0053] As shown in FIG. 3, a part of the generated training signal is displayed on the two-dimensional plane frame of the training image.

[0054] Referring to FIG. 1 again, the training image generation method includes a step (S200) of selecting at least one output format from a plurality of preset output formats to determine the output format of the training image.

[0055] FIG. 4 is a schematic diagram of an output format according to an embodiment of the present application.

[0056] Referring to FIG. 4, the output format defines a structure for arranging the output components of the training image on a two-dimensional plane frame. The plurality of output formats may be associated with output components on different sides of each other. In a specific embodiment, the plurality of output formats may include an output format defining the type of waveform, an output format defining the array position of the waveform, an output format defining the position of the waveform display area, and / or an output format defining the position of the additional information display area.

[0057] The type of waveform may include a channel number.

[0058] The array position of the waveform indicates in what order and at what positions the waveforms for each channel are arranged, and may include the arrangement order and the like.

[0059] The position of the waveform display area may include the coordinate range of the area (for example, a grid area) where the waveform is displayed.

[0060] The position of the additional information display area includes the coordinate range of the area where additional information other than the waveform is displayed. The additional information may include subject information of the waveform (for example, age, gender), waveform analysis, scale information of the waveform, and the like.

[0061] The waveform display area and the additional information display area may be distinguished from each other, or at least a part of them may overlap with each other.

[0062] Each of the plurality of output formats consists of one or more format elements that constitute the arrangement structure of the output format. Therefore, similar to the deformation function based on the deformation element, the output format may also be expressed as a function consisting of format elements. One or more format elements that make up each of the n (n is a natural number greater than or equal to 1) output format functions (f n 2 ) are predefined and stored. Incidentally, here, the subscript n is a character indicating the existence of n functions, and the superscript is a number for distinguishing from other functions described above and below.

[0063] In a specific embodiment, the output format function (f n 2 ) has a format vector (P n 2 ) that may include m factors, and each factor is indicated by a format element P nm 2 (where n is the output format function identifier and m is the element identifier). The format element (P nm 2 ) corresponds to the placement object defined in the output format, that is, the output component.

[0064] In one embodiment, the format elements may include, for example, the channel number, channel order, position of the waveform display area, position of the additional display area, vertical interval between each waveform, horizontal interval between individual waveforms, and / or length of the time interval of the waveform. For example, the format element of the output format that defines the position of the waveform display area may include the position of the waveform display area on a two-dimensional plane.

[0065] When the format element has a specific value, the output element corresponding to the format element is embodied in the output format with the specific value and represented on the training image.

[0066] Similar to the transformation function, the format elements of the output format function may be embodied by hyperparameters. When one output format function includes a number of transformation elements, the number of format elements may be represented by a hyperparameter vector. Here, the vector value is the format element value. Then, each output format function may be associated with its own hyperparameter vector.

[0067] In one embodiment, the step of selecting at least one output format function (S200) may select any one output format function according to a preset second-1 probability distribution for the preset set of multiple output formats, or select two or more output format functions among the multiple output format functions.

[0068] The probability distribution of the second - 1 defines the probability of selecting a specific output format function (which may be multiple) to generate a training image from the entire plurality of output format functions. Selecting an output format function according to the probability distribution of the second - 1 is similar to selecting a transformation according to the probability distribution of the first - 1.

[0069] In one embodiment, the probability distribution of the second - 1 may be defined in a way that selects one output format function out of n. For example, the probability distribution of the second - 1 may have a multinomial distribution.

[0070] In another embodiment, the probability distribution of the second - 1 may be defined in a way that selects two or more output format functions out of n. For example, the probability distribution of the second - 1 may have a binomial distribution. In this case, different output format functions may be selected repeatedly and used to generate a training image.

[0071] In an alternative embodiment, the probability distribution of the second - 1 may be one that optimizes an arbitrarily specified initial probability distribution through training. This will be described in more detail below.

[0072] The selected output format may be used as it is, or the values of the format elements of the selected output format may be changed to determine the output format for the training image (S200). The selected output format function may be input with new values for at least one format element as input arguments to adjust the function of the output format function.

[0073] In one embodiment, the step (S200) of selecting at least one output format from a plurality of preset output formats and determining the output format of the training image may include: changing at least one format element value among the format elements forming the selected output format to a new value; and determining the output format having the changed format element value as the output format of the training image. The value of the format element may be changed based on probability.

[0074] In one embodiment, a second-second probability distribution may be set for each individual format element. When there are m format elements, m second-second probability distributions are defined and assigned. For each of the format elements of the output format function selected according to the second-first probability distribution, the existing format element value may be changed to a new value selected according to the second-second probability distribution preset for itself. The output format reflected by the changed new value is used for generating the training image.

[0075] The second-second probability distribution defines a probability distribution in which individual values are specified from the entire range of values that each format element of the output format function can have. In a specific embodiment, the second-second probability distribution may include a Gaussian (normal) distribution, a gamma distribution, an exponential distribution, a uniform distribution, a chi-square distribution as a continuous probability distribution, a binomial distribution, a negative binomial distribution, an initial lower distribution, a Poisson distribution, and a multinomial distribution and / or a multivariate distribution for a number of deformation elements as a discrete probability distribution. The same or different second-second probability distributions may be set for each format element. Also, in addition to the parametric approach method of extracting format elements in a state where the probability distribution of the specific form described above is defined, the format elements may be randomly extracted without assuming a specific distribution.

[0076] In one embodiment, when the format element is embodied in numerical data, the second-second probability distribution may be a Gaussian (normal) distribution, a gamma distribution, an exponential distribution, or a chi-square distribution, which is a continuous probability distribution. Then, the new value of the format element is a value selected by the continuous probability distribution.

[0077] In one embodiment, when the formal element is embodied in binary variable data, the second-second probability distribution may be a binomial distribution, a negative binomial distribution, an initial lower distribution, or a Poisson distribution, which are discrete probability distributions. The binary variable data may be represented by a second binary value indicating "yes" or a second binary value indicating "no". The new value of the formal element is a value selected according to the discrete probability distribution.

[0078] In one embodiment, when the formal element is embodied in a categorical variable, the second-second probability distribution may be a multinomial probability distribution. Then, the new value of the formal element is a value selected according to the multinomial probability distribution.

[0079] In some embodiments, the set of formal elements forming the output format function may include at least some formal elements having a correlation relationship with each other. When the formal element changed to a new value has a correlation relationship with other formal elements, the second-second probability distribution set for the other formal elements having the correlation relationship may be a multivariate probability distribution. Then, the values of the other formal elements having the correlation relationship are changed to new values selected according to a predefined multivariate probability distribution.

[0080] And in addition to the parametric approach method of extracting formal elements in the state where the probability distribution of the specific form is defined as described above, formal elements may be randomly extracted without assuming a specific distribution.

[0081] In an alternative embodiment, the second-second probability distribution may be one optimized through training from an arbitrarily specified initial probability distribution. This will be described in more detail below.

[0082] In this way, it is supplied as an output format for generating the training image in an output format having a formal element value changed according to the second-second probability distribution (S200).

[0083] Referring back to FIG. 1, the training image generation method includes a step (S300) of determining the output section for each channel of the training signal based on the length of the time interval of the waveform in the determined output format, and a step (S400) of selecting a scale for each axis to determine the grid scale of the training image.

[0084] The step (S300) of determining the output section for each channel of the training signal may include a step of selecting at least one of the start point and the end point of the output section for each channel to be output to the training image among the total lengths of the received source signals; and a step of calculating the output section for each channel based on the length of the time interval of the waveform and the selected point among the format elements of the selected output format.

[0085] Generally, only a part of the total measured signals is output to the image. The total length of the output section is determined as the time interval of the waveform among the format attributes of the output format determined in step (S200) (S300).

[0086] The start point or the end point is selected from the range to which the entire time interval of the determined waveform belongs among the entire section of the generated training signal.

[0087] The selection range of the start point may be the range from the rising point of the training signal to the point extended in the negative time direction by only the time interval of the waveform determined in step (S200) from the falling point of the training signal. In an example, when the entire section of the training signal is [0, Ω] and the time interval of the waveform is w, the distribution of the start point may be [0, Ω - w]. In this case, the end point is the point where the determined time interval of the waveform is extended in the positive time direction from the selected start point.

[0088] The selection range of the end point is opposite to the selection range of the start point, and the selection range of the end point may be a range up to a point where the falling edge point of the training signal and the time interval of the waveform determined from the rising edge point of the training signal are extended in the positive time direction. In the above example, the distribution of the end point may be [w, Ω]. In this case, the start point is a point where the time interval of the determined waveform is extended in the negative time direction from the selected end point.

[0089] In one embodiment, the start point or the end point may be a point selected according to a third probability distribution preset from the selection range for each point. The third probability distribution defines a probability distribution in which individual values are specified from the selection range for each point.

[0090] The third probability distribution may be a Gaussian (normal) distribution, a uniform distribution, or a chi-square distribution as a continuous probability distribution. When the start point is selected by a uniform probability distribution or the like, the end point may be automatically selected and the output interval may be determined from the training signal.

[0091] In some embodiments, the plurality of channels may include synchronized channels where the output intervals of at least some of the channels match each other. In some other embodiments, the plurality of channels may be asynchronous channels with different output intervals from each other.

[0092] By the operation of such a step (S300), the waveform content to be displayed in the training image is determined for the entire section of the training signal.

[0093] The step (S400) of determining the grid scale of the training image may be determined by selecting the horizontal axis unit scale and / or the vertical axis unit scale of the coordinate system for displaying the training signal.

[0094] FIG. 5 is a schematic diagram of the grid scale according to an embodiment of the present application.

[0095] Referring to FIG. 5, on the grid scale, the horizontal axis (or x-axis) represents time, and the vertical axis (or y-axis) represents the signal measurement value.

[0096] In some embodiments, the step of selecting the horizontal axis unit scale and / or the vertical axis unit scale may include: selecting any one of the plurality of horizontal axis unit scales according to a preset fourth-1 probability distribution for the entire plurality of preset horizontal axis unit scales; and / or selecting any one of the plurality of vertical axis unit scales according to a preset fourth-2 probability distribution for the entire plurality of preset vertical axis unit scales.

[0097] Similar to the first-1 probability distribution, the fourth-1 and fourth-2 probability distributions may have a multinomial distribution. Then, the training signal is displayed on the training image in the selected horizontal axis unit and / or vertical axis unit.

[0098] Referring to FIG. 1 again, the training image generation method includes a step (S500) of representing a grid pattern on a two-dimensional plane according to the grid scale determined in step (S400).

[0099] FIG. 6 shows the grid pattern according to an embodiment of the present application.

[0100] Referring to FIG. 6, the grid pattern according to the grid scale is a pattern in which grids are arranged in grid scale units for each axis determined in step (S400) (S500).

[0101] In one embodiment, the step (S500) of representing the grid pattern on a two-dimensional plane may include: selecting any one of a plurality of preset grid pattern forms.

[0102] The grid pattern format defines a grid pattern using the unique display line hierarchy and / or the unique display line design of the pattern. The hierarchy indicates, for example, large intervals, medium intervals, small intervals, etc.

[0103] Similar to the output format function, each of the plurality of grid pattern formats consists of one or more pattern elements that make up the grid pattern format. Therefore, similar to the output format function, the grid pattern format may also be represented as a function consisting of pattern elements. One or more pattern elements that make up each of n (n is a natural number greater than or equal to 1) grid pattern format functions (f n 3 ) are predefined and stored. Incidentally, here, the subscript n is a character indicating the existence of n functions, and the superscript is a number for distinguishing from other functions described above and below.

[0104] In a specific embodiment, the pattern vector (P n 3 ) of the grid pattern format function (f n 3 ) may include m factors, and each factor is indicated by the pattern element P nm 3 (here, n is the grid pattern format function identifier, and m is the element identifier). The pattern element (P nm 2 ) corresponds to the pattern component defined by the grid pattern format function.

[0105] In one embodiment, the pattern element may include the pattern of the display line of the grid pattern (various display forms for representing virtual lines such as solid lines / dotted lines / double lines), thickness, hue, etc. The hue may be embodied by RGB values, CMYK channel-specific values, etc.

[0106] When the pattern element has a specific value, the pattern element corresponding to the pattern element is embodied on the grid pattern format with the specific value and represented on the training image.

[0107] Similar to the output form function, the pattern elements of the lattice pattern form function may be embodied by hyperparameters. When one lattice pattern form function includes a large number of pattern elements, the large number of pattern elements may be represented by a hyperparameter vector. Here, the vector value is the pattern element value. Then, each lattice pattern form function may be associated with its own hyperparameter vector.

[0108] In one embodiment, the step of selecting any one of the plurality of lattice pattern forms may select any one of the lattice pattern forms according to a preset fifth - 1 probability distribution.

[0109] The fifth - 1 probability distribution defines the probability of selecting a specific lattice pattern form to generate a training image from the whole of a plurality of lattice pattern form functions. Selecting the output form function according to the fifth - 1 probability distribution is the same as selecting a transformation according to the first - 1 probability distribution.

[0110] In one embodiment, the fifth - 1 probability distribution may be defined in a way of selecting 1 out of n lattice pattern form functions. For example, the fifth - 1 probability distribution may have a multinomial distribution.

[0111] In an alternative embodiment, the fifth - 1 probability distribution may be one obtained by optimizing an arbitrarily specified initial probability distribution through training. This will be described in more detail below.

[0112] The selected lattice pattern form may be used as it is, or the value of the pattern element of the selected lattice pattern form may be changed to determine the lattice pattern form for the training image (S500). The selected lattice pattern form may input new values for at least one form element as input arguments to adjust the function of the lattice pattern form function.

[0113] In one embodiment, the step (S500) of representing the lattice pattern on a two-dimensional plane may include: adjusting the pattern element values of the selected lattice pattern format; and determining the lattice pattern format having the adjusted pattern element values as the lattice pattern format of the training image. The change of the value of the pattern element may be performed based on probability.

[0114] In one embodiment, a fifth-second probability distribution may be set for each individual pattern element. When there are m pattern elements, m fifth-second probability distributions are defined and assigned. For each pattern element of the lattice pattern format function selected according to the fifth-first probability distribution, the existing pattern element value may be changed to a new value selected according to the fifth-second probability distribution preset for itself. The lattice pattern format reflected by the changed new value is used for generating the training image.

[0115] The fifth-second probability distribution defines a probability distribution in which individual values are specified from the entire range of values that each pattern element of the lattice pattern format function can have. In a specific embodiment, the fifth-second probability distribution may include a Gaussian probability distribution, a uniform probability distribution, a continuous uniform probability distribution, a normal distribution, a chi-square distribution, a binomial probability distribution, a multinomial probability distribution, and / or a multivariate probability distribution. The same or different fifth-second probability distributions may be set for each pattern element.

[0116] In one embodiment, when the pattern element is implemented with numerical data, the fifth-second probability distribution may be a Gaussian (normal) distribution, a gamma distribution, an exponential distribution, a uniform distribution, or a chi-square distribution as a continuous probability distribution. Then, the new value of the pattern element is a value selected by the continuous probability distribution.

[0117] In one embodiment, when the pattern element is embodied in binary variable data, the 5-2 probability distribution may be a binomial distribution, a negative binomial distribution, an initial lower distribution, or a Poisson distribution, which are discrete probability distributions. The binary variable data may be represented by a second binary value indicating "yes" or a second binary value indicating "no". The new value of the pattern element is a value selected according to a discrete probability distribution.

[0118] In some embodiments, a set of pattern elements forming a lattice pattern form function may include at least some pattern elements having a correlation relationship with each other. When a pattern element changed to a new value has a correlation relationship with other pattern elements, the 5-2 probability distribution set for the other pattern elements having the correlation relationship may be a multivariate probability distribution. Then, the values of the other pattern elements having the correlation relationship are changed to new values selected according to a predefined multivariate probability distribution.

[0119] Then, the lattice pattern is represented on a two-dimensional plane in a lattice pattern form obtained by adjusting the selected lattice pattern form.

[0120] In an alternative embodiment, the 5-2 probability distribution may be one optimized from an arbitrarily specified initial probability distribution through training. This will be described in more detail below.

[0121] Referring to FIG. 1 again, the training image generation method includes: setting a reference position of waveform content of a training signal determined in step (S300) based on at least one of an output format determined in step (S200), an output interval for each channel determined in step (S300), and a lattice scale determined in step (S400) (step (S600)); and representing the waveform content and signal identifier of the training signal on a two-dimensional plane on which a lattice pattern is represented in step (S500) (step (S700)).

[0122] In step (S200), the positions of the training signals for each channel and the positions of the output object arrangement structures are calculated on the two-dimensional plane forming the frame of the training image. The content of the waveform displayed in the training image is determined through step (S300).

[0123] Coordinates of the measured values for the waveform content in the training signal are calculated (S600) according to the grid scale determined in step (S400). The coordinates of the measured values are positions on the two-dimensional plane, which define at which positions the waveform content of each channel is arranged. The coordinates of the measured values are calculated as coordinate values based on the grid pattern.

[0124] The calculated measured values of the training signal may be utilized as the reference coordinates of the waveform content of the training signal. FIG. 7 is a schematic diagram of the result of representing the waveform content and signal identifiers of a training signal on a two-dimensional plane on which a grid pattern is marked according to an embodiment of the present application.

[0125] Referring to FIG. 7, in step (S700) of representing the waveform content and signal identifiers of the training signal on a two-dimensional plane on which the grid pattern is marked, the waveform content and signal identifiers for each channel of the training signal may be represented on the two-dimensional plane on which the pattern is marked based on the reference position.

[0126] In one embodiment, step (S700) of representing the waveform content for each channel of the training signal on the two-dimensional plane on which the grid pattern is marked based on the reference position may include: defining a representation function for graphically expressing the waveform content of the training signal for each channel based on the reference position; and representing the waveform content and signal identifiers of the training signal using the defined representation function.

[0127] The representation function (f n 4 ) is based on the reference coordinates of the training signal and one or more representation vectors (P nm 4) is based on. The notation vector (P n 4 ) may include m factors, and each factor is a notation element P nm 4 indicated by (where n is the notation function identifier and m is the element identifier). Incidentally, here, the subscript n is a character indicating the existence of n functions, and the superscript is a number for distinguishing from other functions described above).

[0128] The reference coordinates of the training signal may include the coordinates of the measured values of the waveform content of the training signal calculated according to the grid pattern, such as the coordinates on the horizontal axis and the vertical axis of the signal.

[0129] The notation element (P nm 4 ) defines the waveform design and the signal identifier. The notation element consists of a first group related to the waveform design and a second group related to the signal identifier design.

[0130] The notation elements of the first group may include the waveform pattern, thickness, hue, etc. The hue may be embodied by RGB numerical values, numerical values for each CMYK channel, etc.

[0131] The signal identifier includes a channel identifier and a scale identifier. The notation elements of the second group may include the position, pattern, font, hue, thickness of the character (or line), etc. of the identifier. The position of the identifier is indicated by the relative position from the represented training signal.

[0132] Such a definition may be made according to the user's input or may be preset.

[0133] The waveform content of the training signal may be represented on a two-dimensional plane using the notation function, and the channel identifier and the scale identifier may be represented within a certain distance from the waveform of the training signal (S700).

[0134] When the notation element value is adjusted, the waveform content of the training signal, the channel identifier, and the scale identifier may be represented on the two-dimensional plane reflecting the adjusted notation element value (S700).

[0135] In one embodiment, the step of representing the waveform content and the signal identifier of the training signal using the defined notation function may include: adjusting the value of the notation element of the notation function; and representing the waveform content of the training signal for each channel on the two-dimensional plane on which the grid pattern is represented, reflecting the adjusted notation element value. The change of the value of the notation element may be performed based on probability.

[0136] In one embodiment, a sixth probability distribution may be set for each individual notation element. When there are m notation elements, m fifth probability distributions are defined and assigned. For each notation element of the notation function selected according to the sixth probability distribution, the existing notation element value may be changed to a new value selected according to a sixth probability distribution preset for itself. The training image is represented on the two-dimensional plane in a notation format reflecting the changed new value.

[0137] The sixth probability distribution defines a probability distribution in which individual values are specified from the entire range of values that each notation element of the notation function can have. In a specific embodiment, the sixth probability distribution may include a Gaussian (normal) distribution, a uniform distribution, a chi-square distribution as a continuous probability distribution, a binomial distribution, a negative binomial distribution, an initial lower distribution, a Poisson distribution, and a multinomial distribution as a discrete probability distribution, and / or a multivariate probability distribution for a number of deformation elements. The same or different sixth probability distributions may be set for each notation element. In addition to the parametric approach method of extracting notation elements in a state where the probability distribution of the specific form described above is defined, the notation elements may be randomly extracted without assuming a specific distribution.

[0138] In one embodiment, when the notation element is embodied in numerical data, the sixth probability distribution may be a Gaussian (normal) distribution, a uniform distribution, a continuous uniform probability distribution, a normal distribution, or a chi-squared distribution as a continuous probability distribution. Then, the new value of the notation element is a value selected from the continuous probability distribution.

[0139] In one embodiment, when the notation element is embodied in binary variable data, the sixth probability distribution may be a binomial distribution, a negative binomial distribution, an initial lower distribution, or a Poisson distribution, which are discrete probability distributions. The binary variable data may be represented by a second binary value indicating "yes" or a second binary value indicating "no". The new value of the notation element is a value selected according to the discrete probability distribution.

[0140] In one embodiment, when the notation element is embodied in a categorical variable, the sixth probability distribution may be a multinomial probability distribution. Then, the new value of the notation element is a value selected according to the multinomial probability distribution.

[0141] In some embodiments, the set of notation elements forming the notation function may include at least some notation elements having a correlation relationship with each other. When the notation element changed to a new value has a correlation relationship with other notation elements, the sixth probability distribution set for the other notation elements having the correlation relationship may be a multivariate probability distribution. Then, the values of the other notation elements having the correlation relationship are changed to new values selected according to a predefined multivariate probability distribution. Then, the waveform content and signal identifier of the training signal are represented on a two-dimensional plane in a notation format having the changed value (S700).

[0142] In an alternative embodiment, the sixth probability distribution may be one obtained by optimizing an arbitrarily specified initial probability distribution through training. This will be described in more detail below.

[0143] In addition, the training image generation method further includes a step (S800) of further indicating the additional information and / or scale information of the training signal.

[0144] FIG. 8 is a schematic diagram showing the indication of additional information according to an embodiment of the present application.

[0145] Referring to FIG. 8, the additional information of the training signal is the original attributes of the training signal, similar to the source additional information. The additional information of the training signal may be embodied in text meaning age, gender, measurement location, measurement time, or waveform analysis.

[0146] In one embodiment, the step of indicating the additional information of the training signal may be a step of selecting and indicating an additional information text arbitrarily selected from a preset additional information text. The indication position of the additional information text is based on the output format determined in step (S300).

[0147] In some other embodiments, the additional information of the training signal may be the source additional information. In this case, the additional information of the training signal and the additional information of the source signal are the same.

[0148] The scale information is information describing the grid scale determined in step (S400), and may be embodied in a symbol, pattern, or text meaning the grid scale. The indication position of the scale information is also based on the output format determined in step (S200).

[0149] In one embodiment, the step of indicating the scale information may include a step of selecting any one of a plurality of preset scale indication methods; and a step of indicating the scale information by the selected scale indication method.

[0150] In one embodiment, the step of selecting any one of a plurality of preset scale notation methods may be to select any one of the plurality of preset scale notation methods according to a seventh probability distribution.

[0151] The seventh probability distribution may be defined by a method of selecting 1 out of n. For example, the seventh probability distribution may have a multinomial distribution.

[0152] Also, the training image generation method may further include a step (S900) of further transforming the generated training image.

[0153] The step (S900) may include a step of selecting at least one image augmentation function from a plurality of preset image augmentation functions; and a step of further transforming the training image using the selected image augmentation function.

[0154] The image augmentation function may consist of a number of augmentation elements. The augmentation elements may include types of transformation, frequency, intensity, and / or epoch number.

[0155] The step of selecting the augmentation function is the same as selecting a transformation function. In one embodiment, the step of selecting the augmentation function is to select any one of the plurality of image augmentation functions according to an eighth probability distribution preset for the set of the plurality of image augmentation functions, or select two or more of the plurality of image augmentation functions.

[0156] The eighth probability distribution defines the probability of selecting a specific image augmentation function (which may be plural) to generate a training image from the entirety of the plurality of image augmentation functions.

[0157] In one embodiment, the eighth probability distribution may be defined by a method of selecting one image enhancement function out of n. For example, the eighth probability distribution may have a multinomial distribution.

[0158] In another embodiment, the eighth probability distribution may be defined by a method of selecting two or more image enhancement functions out of n. For example, the eighth probability distribution may have a binomial distribution.

[0159] Further, the training image generation method may further include a step (S1000) of converting the data of the generated training image into a tensor in the form of W×H×C'. Here, W represents the width of the image, H represents the height of the image, and C' represents the number of color channels (in the case of monochrome, C' = 1). When a multi-channel one-dimensional signal in the form of a C×T real number array is input as the source signal, it is also possible to generate a training image by such a training image generation method and output the result converted into a three-dimensional real number array in the form of W*h*C.

[0160] In the alternative embodiment, the probability distribution may be one obtained by optimizing an arbitrarily specified initial probability distribution through training.

[0161] As described above, the hyperparameter vector may include binomial data, multinomial data, and numerical data.

[0162] The binomial data is selected according to the binomial distribution. As described above, the selection according to the binomial distribution may be performed two or more times. The label data for the binomial data is implemented by a multilabel choice in which "yes" is selected for one or more items.

[0163] Multinomial data is selected according to the multinomial distribution. As described above, the selection according to the multinomial distribution may be performed alone. Then, the label data for the multinomial data is embodied by single label choice.

[0164] Numerical data is selected according to a probability distribution having a limited or unlimited range. Such probability distributions may include Gaussian (normal) distribution, gamma distribution, exponential distribution, uniform distribution, chi-square distribution, etc. as continuous probability distributions. As described above, the selection according to the uniform distribution is to select real numbers in a limited or unlimited range.

[0165] The elements embodied in the data format correspond to the hyperparameters of the machine learning model and may be subject to optimization.

[0166] Such hyperparameters may be optimized in the process of learning an image-based artificial intelligence model that analyzes multi-channel one-dimensional signal images by utilizing the generated training images.

[0167] The hyperparameters are optimized under the following objectives: a) improvement in accuracy in the text data set of the trained artificial intelligence, b) improvement in robustness against domain shift and adversarial attack, c) improvement in the embedding quality of the latent vector (i.e., minimization of the distance in the latent space between the same concepts).

[0168] The optimization of the hyperparameters may be performed through optimization algorithms such as Grid search, Random search, Gaussian process, Tree-structed Parzen Estimator (TPE), etc., but is not limited thereto.

[0169] According to another aspect of the present application, the training image generation method may be performed by components configured to receive and process data.

[0170] FIG. 9 is a block diagram of an apparatus for executing the training image generation method according to another aspect of the present application.

[0171] Referring to FIG. 9, the apparatus includes an acquisition unit 10; and an image generation unit 100.

[0172] The acquisition unit 10 acquires a source signal from a device capable of measuring a source signal such as a multi-channel one-dimensional signal. For example, an electrocardiogram signal can be directly or indirectly acquired from an electrocardiogram measuring device that is attached to a part of the body of the subject and measures a multi-channel one-dimensional signal of the subject (user), for example, an electrocardiogram signal.

[0173] The acquisition unit 10 may be connected to receive information from an electrocardiogram measuring device that measures an electrocardiogram signal of the subject via a sensor attached to a part of the body of the subject. Then, the acquisition unit 10 can also directly acquire the electrocardiogram signal from the electrocardiogram measuring device.

[0174] The sensor may be attached to a part of the body of the subject and measure the electrocardiogram signal of the subject (user). The electrocardiogram signal acquired from the sensor and the electrocardiogram signal measuring device may be converted into a digital signal via an analog-digital converter (ADC). Further, an electrocardiogram measuring device (not shown) can measure a biological signal when it is determined that the user's body has been in contact for a predetermined time or more via a touch panel. According to another embodiment, the acquisition unit 10 can acquire not only an electrocardiogram signal (raw signal) but also an electrocardiogram image output and visualized on paper or an image based on the already obtained electrocardiogram signal.

[0175] The acquisition unit 10 may be connected to receive information from an electrocardiogram measuring device that measures an electrocardiogram signal of a subject via a sensor attached to a part of the subject's body. Then, the acquisition unit 10 can also directly acquire the electrocardiogram signal from the electrocardiogram measuring device.

[0176] Alternatively, the acquisition unit 10 may be connected to communicate with an external device by wired or wireless electrical means. Then, the acquisition unit 10 can also acquire from electrocardiogram signal data that has been previously acquired or stored in the external device. The external device acquires electrocardiogram signal data by being connected to an electrocardiogram measuring device itself or via another external device connected to the electrocardiogram measuring device. Therefore, the acquisition of electrocardiogram signal data from the external device by the acquisition unit 10 may be treated as an indirect acquisition of the electrocardiogram signal.

[0177] Source signal information measured via one or more channels by the acquisition unit 10 is acquired. The source signal information may include an analog source signal, digital source signal information, or a source image displaying the source signal.

[0178] The image generation unit 100 is a computing device including a processor and a memory. When receiving the source signal information received by the acquisition unit 10, it may execute the steps (S100 to S800) of the training image generation method in FIG. 1.

[0179] In one embodiment, the image generation unit 100 may be implemented on a server. The acquisition unit 10 may be a device (e.g., a user terminal or a signal input device) connected to the server to input data.

[0180] In this case, the server is a number of computer systems or computer software implemented by a network server, which can configure and provide various information on a website. Here, the network server refers to a computer system and computer software (network server program) that is connected to a subordinate device capable of communicating with other network servers via a computer network such as a private intranet or the Internet, accepts a work execution request, executes the work accordingly, and provides the execution result. However, in addition to such a network server program, it should be understood as a broad concept including a series of application programs operating on the network server and various databases sometimes built inside. For example, when including various databases, the server is configured to use external database information such as the cloud. In this case, the server can connect to an external database server (for example, a cloud server) through operation for data communication.

[0181] The operations of the training image generation method according to the embodiments described above and the apparatus for executing the same may be at least partially implemented by a computer program and may be recorded on a computer-readable recording medium. For example, it may be implemented together with a program product composed of a computer-readable medium including program code, which may be executed by a processor for performing any or all of the described steps, operations, or processes.

[0182] The computer may be a desktop computer, laptop computer, notebook, smartphone, or similar computing device, or any device that may be integrated. The computer is a device having one or more alternative and special-purpose processors, memories, storage spaces, and networking components (either wireless or wired). The computer can execute, for example, an operating system compatible with Microsoft Windows, Apple OSX (registered trademark) or iOS, a Linux distribution, or Google's Android OS.

[0183] The computer-readable recording medium includes all types of recording devices in which computer-readable data is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage devices, etc. Also, the computer-readable recording medium may be distributed in a computer system connected via a network, and computer-readable code may be stored and executed in a distributed manner. Also, functional programs, codes, and code segments for implementing this embodiment should be easily understood by those skilled in the technical field to which this embodiment belongs.

[0184] As described above, the present invention has been described with reference to the embodiments shown in the drawings, but these are merely exemplary, and it should be understood that those skilled in the art can make various modifications and variations based on this. However, such modifications should be considered to be within the technical protection scope of the present invention. Therefore, the true technical protection scope of the present invention should be determined by the technical idea of the appended claims.

Industrial Applicability

[0185] The present invention relates to a training image generation technique used for training an image-based artificial intelligence model and an apparatus for executing the same, and can be applied to signal analysis in various industrial fields that analyze two-dimensional images based on signals in the medical bio field and other one-dimensional signals.

Claims

1. In a training image generation method executed by a computing device including a processor and a memory, generating a training signal based on source signal information; selecting at least one output format from a plurality of preset output formats to determine the output format of the training image; determining an output section for each channel of the training signal based on the length of the time interval of the waveform of the determined output format; selecting a scale for each axis to determine the grid scale of the training image; marking a grid pattern on a two-dimensional plane according to the determined grid scale; setting a reference position of the waveform content of the training signal based on at least one of the determined output section for each channel and the determined grid scale; and marking the waveform content and signal identifier of the training signal on a two-dimensional plane on which the grid pattern is marked A training image generation method comprising:

2. The step of generating a training signal based on the source signal information includes: selecting at least one transformation function from a plurality of preset transformation functions; and transforming the source signal into a training signal using the selected transformation function, The transformation function is composed of one or more transformation elements indicating transformation of the signal processing attributes of the input signal, and each transformation element corresponds to the signal processing attribute. The training image generation method according to claim 1.

3. Transforming the signal processing attribute includes processing, changing, or removing one or more of signal magnitude, waveform, frequency range, frequency distribution, signal rising time, and signal time range, The transformation of the signal processing attribute is characterized in that it is performed for each channel, for a channel group, or for the entire channel. The training image generation method according to claim 2.

4. The step of selecting at least one transformation function from the plurality of transformation functions includes: selecting any one transformation function from the plurality of transformation functions according to a preset first-1 probability distribution for the set of the plurality of transformation functions, or selecting two or more transformation functions from the plurality of transformation functions, When selecting one transformation function out of N, the first-1 probability distribution is a multinomial distribution, The training image generation method according to claim 2, characterized in that when selecting two or more deformation functions out of N, the first - 1 probability distribution is a binomial distribution.

5. The step of transforming the source signal into a training signal using the selected deformation function includes: changing at least one deformation element value among the deformation elements forming the selected deformation function to a new value; and then generating a signal reflecting the signal processing attributes changed according to the changed deformation element value as a training signal, The training image generation method according to claim 2, characterized in that the new value is a value selected according to a preset first - 2 probability distribution for the deformation element.

6. When the deformation element is embodied by numerical - type data, the new value of the deformation element is a value selected according to a continuous probability distribution, When the deformation element is embodied by binary variable data, the new value of the deformation element is a value selected according to a discrete probability distribution, The training image generation method according to claim 5, characterized in that when the deformation element is embodied by a categorical variable, the new value of the deformation element is a value selected according to a multinomial probability distribution.

7. The set of deformation elements forming the deformation function includes at least some deformation elements having a correlation relationship with each other. The training image generation method according to claim 5, characterized in that when the deformation element changed to a new value has a correlation relationship with other deformation elements, the values of the other deformation elements having the correlation relationship are changed to new values selected according to a predefined multivariate probability distribution.

8. The step of selecting at least one output format out of the preset plurality of output formats to determine the output format of the training image includes: selecting any one output format according to a preset second - 1 probability distribution for the set of the preset plurality of output formats, or selecting two or more output formats out of the plurality of output formats, When selecting 1 output format out of N, the second - 1 probability distribution is a multinomial distribution, The training image generation method according to claim 1, characterized in that when selecting two or more deformation functions out of N, the second - 1 probability distribution is a binomial distribution.

9. The step of determining the output format of the training image by selecting at least one output format from a plurality of preset output formats is as follows: The step of changing at least one of the format element values of the format elements forming the selected output format to a new value; and The step of determining the output format having the changed format element value as the output format of the training image, Each of the plurality of output formats consists of one or more format elements that define the layout structure of the output format, The training image generation method according to claim 8, wherein the new value is a value selected according to a preset second - 2 probability distribution for the format element.

10. When the format element is embodied by numerical data, the new value of the format element is a value selected according to a continuous probability distribution, When the format element is embodied by binary variable data, the new value of the format element is a value selected according to a discrete probability distribution, The training image generation method according to claim 9, wherein when the format element is embodied by a categorical variable, the new value of the format element is a value selected according to a multinomial probability distribution.

11. The set of format elements forming the output format includes at least some format elements having a correlation relationship with each other, The training image generation method according to claim 9, wherein when the format element changed to a new value has a correlation relationship with other format elements, the values of the other format elements having the correlation relationship are changed to new values selected according to a predefined multivariate probability distribution.

12. The step of determining the channel - specific output interval of the training signal based on the length of the time interval of the waveform of the determined output format is as follows: The step of selecting at least one of the start point and the end point of the channel - specific output interval to be output to the training image from the total length of the received source signal; and The training image generation method according to claim 1, characterized by including the step of calculating the channel - specific output interval based on the length of the time interval of the waveform and the selected point among the format elements of the selected output format.

13. The start point or the end point is a point selected according to a preset third probability distribution from the selection range for each point, The third probability distribution defines a probability distribution in which individual values are specified from the selection ranges for the start point and the end point. The selection range of the start point is a range up to a point that is extended in the negative time direction by a time interval of a waveform determined from the rising point of the training signal and the falling point of the training signal. The training image generation method according to claim 12, wherein the selection range of the end point is a range up to a point that is extended in the positive time direction by a time interval of a waveform determined from the falling point of the training signal and the rising point of the training signal.

14. The step of selecting the scale for each axis to determine the grid scale of the training image includes: selecting any one of a plurality of horizontal axis unit scales according to a preset fourth-1 probability distribution for the entire preset plurality of horizontal axis unit scales; or selecting any one of a plurality of vertical axis unit scales according to a preset fourth-2 probability distribution for the entire preset plurality of vertical axis unit scales, The training image generation method according to claim 1, wherein the fourth-1 and fourth-2 probability distributions have a multinomial distribution.

15. The step of representing the grid pattern on the two-dimensional plane according to the determined grid scale includes: selecting any one of the plurality of grid pattern forms according to a preset fifth-1 probability distribution for the preset set of the plurality of grid pattern forms; The training image generation method according to claim 1, wherein the fifth-1 probability distribution is a multinomial distribution.

16. The step of representing the grid pattern on the two-dimensional plane according to the determined grid scale includes: adjusting the pattern element values of the selected grid pattern form; and determining the grid pattern form having the adjusted pattern element values as the grid pattern form of the training image, Each of the plurality of grid pattern forms consists of one or more pattern elements that define the grid pattern by the display line hierarchy of the pattern or the unique display line design of the pattern. The training image generation method according to claim 15, wherein the adjusted value is a value selected according to a preset fifth-second probability distribution for the formal element.

17. When the pattern element is embodied in numerical data, the adjusted value of the pattern element is a value selected according to a continuous probability distribution. The training image generation method according to claim 16, wherein when the pattern element is embodied in binary variable data, the adjusted value of the pattern element is a value selected according to a discrete probability distribution.

18. The set of pattern elements having the lattice pattern form includes at least some pattern elements having a correlation relationship with each other. The training image generation method according to claim 16, wherein when the pattern element adjusted to a new value has a correlation relationship with other pattern elements, the values of the other pattern elements having the correlation relationship are adjusted to new values selected according to a predefined multivariate probability distribution.

19. The training image generation method according to claim 1, wherein the reference position of the waveform content of the training signal includes the coordinates of the measured value of the training signal calculated as coordinate values based on the lattice pattern.

20. The step of representing the waveform content and signal identifier of the training signal on a two-dimensional plane on which a lattice pattern is represented includes: defining a representation function for graphically representing the waveform content of the training signal for each channel based on the reference position; and representing the waveform content and signal identifier of the training signal using the defined representation function. The representation function is based on the reference coordinates of the training signal and one or more representation elements, and the representation elements define the design of the waveform or the design of the signal identifier. The training image generation method according to claim 1.

21. The step of representing the waveform content and signal identifier of the training signal using the defined representation function includes: adjusting the values of the representation elements of the representation function; and representing the waveform content of each channel of the training signal on the two-dimensional plane on which the lattice pattern is represented, reflecting the adjusted representation element values. The training image generation method according to claim 20, wherein the adjusted value is a value selected according to a sixth probability distribution preset for the notation element.

22. When the notation element is embodied by numerical data, the adjusted value of the notation element is a value selected according to a continuous probability distribution. When the notation element is embodied by binary variable data, the adjusted value of the notation element is a value selected according to a discrete probability distribution. The training image generation method according to claim 21, wherein when the notation element is embodied by categorical variable data, the adjusted value of the notation element is a value selected according to a multinomial probability distribution.

23. The set of notation elements forming the notation function includes at least some notation elements having a correlation relationship with each other. The training image generation method according to claim 21, wherein when a notation element adjusted to a new value has a correlation relationship with other notation elements, the values of the other notation elements having the correlation relationship are adjusted to new values selected according to a predefined multivariate probability distribution.

24. The training image generation method according to claim 1, wherein the method is used to train an image-based artificial intelligence model that analyzes an image obtained from a multi-channel one-dimensional signal.

25. A computer program stored in a medium so as to be combined with hardware and execute the training image generation method according to any one of claims 1 to 24.

26. A training image generation device, comprising: An acquisition unit that acquires a source signal; and An image generation unit including a processor and a memory, The training image generation device, wherein the image generation unit receives the source signal information received by the acquisition unit and executes the training image generation method according to any one of claims 1 to 24.

Images