KR20190054782A - Apparatus and method for enhancing learning capacility for machine learning - Google Patents

Abstract

The present invention relates to an apparatus and a method to enhance the learning performance of machine learning, capable of improving/enhancing the learning performance of machine learning by generating sturdy learning images through image synthesis. The present invention includes: a learning data module outputting learning data of a three-dimensional CAD model including at least one weapon object; a learning performance enhancing module generating a plurality of two-dimensional weapon object images having different angles, distances, and directions based on the type of the three-dimensional weapon object included in the learning data, and generating a learning image by synthesizing the two-dimensional weapon object images with at least one background image and digital noise; and a high-speed cluster computing module performing the recognition machine learning of the weapon object and pattern learning in accordance with a neural network operation and a neural network model which are provided by a neural network operation framework and a neural network model part.

Description

[0001] APPARATUS AND METHOD FOR ENHANCING LEARNING CAPACITY FOR MACHINE LEARNING [0002]

The present invention relates to an apparatus and method for enhancing learning performance of machine learning capable of enhancing / enhancing learning performance of machine learning by generating various and robust learning images through image synthesis.

In general, machine learning (machine learning) is a technology that studies and constructs a system and its algorithms that perform data learning based on data, perform learning and prediction directly, and improve their performance.

Deep learning techniques for image analysis used in the development of existing machine learning systems require tens of thousands to millions of pieces of learning image data, far exceeding the required amount of learning data required for machine learning.

However, in reality, it is not easy to acquire large-capacity learning data due to a variety of factors other than some standard benchmarking data sets which are partially disclosed. Especially, in the case of learning images related to weapon systems such as trams and airplanes, the source of images is very limited due to the specificity of the field, and there are limitations in collecting / acquiring image data for machine learning learning.

It is an object of the present invention to provide an apparatus and method for enhancing learning performance of machine learning capable of variously generating images for machine learning learning in which image sources are very limited and there are many limitations in acquisition / acquisition.

Another object of the present invention is to provide an apparatus and method for enhancing learning performance of machine learning capable of enhancing data learning performance by generating various / robust synthetic learning images to be applied to machine learning learning.

According to an aspect of the present invention, there is provided an apparatus for enhancing wet performance of machine learning, the apparatus comprising: a learning data module for outputting learning data of a 3D CAD model including at least one inorganic object; A plurality of 2D inorganic object images having different angles, perspectives, and directions are generated based on the type of the 3D inorganic object included in the learning data, and the generated plurality of 2D inorganic object images are combined with one or more background images and digital noise Learning performance enhancement module for generating images; And a fast cluster computing module for performing pattern learning according to a neural network model and a neural network operation provided in the neural network model unit and the neural network operation framework, and machine learning for recognizing the inorganic object, on the generated learning image.

According to an embodiment of the present invention, the background image includes a background image of a real background image or a geometric pattern, and the digital noise may include learning disturbance noise or random noise.

According to an embodiment of the present invention, the learning performance enhancement module divides an inorganic object to perform machine learning from learning data of a 3D CAD model to recognize the type, A plurality of 2D learning images can be automatically generated by adjusting the perspective and direction differently.

According to an embodiment of the present invention, the learning performance enhancement module can divide an inorganic object by removing a background image when the learning data of the 3D CAD model includes a unique background image.

According to an embodiment of the present invention, the learning performance enhancing module is configured to classify one or more background images associated with the generated plurality of 2D learning images and a digital image corresponding to the digital image, based on the type of the recognized inorganic object, Noise is selected from the storage medium, and the specificity may include angle, perspective and direction.

According to an embodiment of the present invention, the learning performance enhancement module determines the specificity of the digital noise type and at least one background image selected from the storage medium, and determines the specificity of each of the plurality of 2D learning images, Sequential synthesis or random synthesis may optionally be performed based on the type of specificity.

According to an embodiment of the present invention, the learning performance enhancement module includes: a 2D learning image generation module for generating a plurality of 2D learning images having different angles, perspective and direction from the learning data of the 3D CAD model; A background image selection module for selecting at least one background image or a background image of a geometric pattern from the background image storage module; A digital noise selection module for selecting at least one digital noise from the digital noise storage module; And a learning image synthesis module that synthesizes the generated plurality of 2D learning images, the selected at least one background image and digital noise, and outputs a plurality of learning images to be used in the tilting machine.

According to another aspect of the present invention, there is provided an apparatus for enhancing a learning performance of a machine learning, the apparatus comprising: a processor for dividing and detecting an inorganic object from learning data of a 3D CAD model including at least one inorganic object, A 2D learning image generation module for generating a plurality of 2D inorganic object images whose angles, perspective and direction are adjusted based on the type of the inorganic object; A background image selection module for selecting at least one background image from the first storage module; A digital noise selection module for selecting at least one digital noise from the second storage module; And a learning image composing module for generating a plurality of learning images to be used for machine learning by synthesizing a 2D inorganic object image, a background image, and digital noise based on the generated plurality of 2D inorganic object images and the at least one background image specificity, ; ≪ / RTI >

According to an embodiment of the present invention, the background image includes a background image of a real background image or a geometric pattern, and the digital noise may include learning disturbance noise or random noise.

According to an embodiment of the present invention, the 2D learning image generation module divides an inorganic object to be machine-runned from learning data of the 3D CAD model to recognize the type, and calculates an angle of the 3D inorganic object based on the recognized type , The perspective and orientation are differently adjusted to automatically generate a plurality of 2D learning images. When the learning data of the 3D CAD model includes a unique background image, the learning performance enhancing module removes the corresponding background image, Can be divided.

According to an embodiment of the present invention, the 2D learning image generation module notifies the background image selection module and the digital noise selection module of the type of the recognized inorganic object and the specificity of the adjusted 3D inorganic object, And the digital noise selection module respectively select one or more background images and digital noise associated with the generated plurality of 2D learning images from the first and second storage media based on the type and the specificity of the inorganic object, Angle, perspective and direction.

According to an embodiment of the present invention, the learning image synthesis module determines the specificity of at least one background image type selected in the background image selection module and the digital noise selection module, and determines the specificity of each of the plurality of 2D learning images, And can selectively perform sequential or random synthesis based on the specificity of the discriminated background image type.

According to an embodiment of the present invention, the learning image synthesis module synthesizes a background image with a 2D learning image and finally synthesizes digital noise when the specificity of each of the 2D learning images exceeds a threshold value, When the specificity of each of the images is less than the threshold value and the specificity of the background image type is equal to or greater than the threshold value, the 2D learning image is synthesized in the background image and finally the digital noise is synthesized. The 2D learning image, the background image, and the digital noise can be randomly synthesized.

According to another aspect of the present invention, there is provided a method for enhancing learning performance of machine learning, the method comprising: outputting learning data of a 3D CAD model including an inorganic object; And generating a plurality of 2D weapon object images by adjusting the specificity of the 3D weapon objects based on the type of the 3D weapon objects included in the learning data and outputting the generated plurality of 2D weapon object images and one or more background images and digital noise And generating a learning image, wherein the generating the learning image comprises: recognizing a type of a specific inorganic object segmented from learning data of the 3D CAD model and generating a 3D inorganic object based on the recognized type; Adjusting a distance, an angle, and a direction of the 2D learning image to generate a plurality of 2D learning images; And selecting one or more background images and digital noise associated with the plurality of 2D learning images based on the type of the recognized inorganic object and the specificity of the adjusted 3D inorganic object.

According to an embodiment of the present invention, the background image includes a background image of a real background image or a geometric pattern, and the digital noise may include learning disturbance noise or random noise.

According to an embodiment of the present invention, when the learning data of the 3D CAD model includes a unique background image, the step of generating the plurality of 2D learning images includes a step of dividing the inorganic object by removing the background image , And the specificity may include an angle, a perspective and an orientation.

According to an embodiment of the present invention, the step of generating the learning image comprises: determining the specificity of the selected at least one background image type; And performing sequential or random compositing based on the specificity of each of the plurality of 2D learning images and the specificity of the determined background image type.

According to an embodiment of the present invention, performing the sequential synthesis or the random synthesis may include synthesizing a background image in a 2D learning image when the specificity of each of the 2D learning images exceeds a threshold value, ; Synthesizing a 2D learning image on a background image when the specificity of each of the 2D learning images is less than a threshold value and the specificity of a background image type is more than a threshold value, and finally synthesizing digital noise; And randomly synthesizing the 2D learning image, the background image, and the digital noise when the specificities of the 2D learning image and the background image are both equal to or less than the threshold value.

The present invention relates to a learning enhancement module for improving learning performance (robustness / generalization) of machine learning learning data by applying a learning image synthesis technique for synthesizing a real background image or a geometric pattern and digital noise to a learning data source It is possible to create various learning data environment and conditions by proposing a mass data generation structure. Therefore, the present invention has an effect of fundamentally solving the problem of insufficient learning data inherent in the existing machine learning process.

1 is a block diagram of a learning performance enhancing apparatus for machine learning according to an embodiment of the present invention;
FIG. 2 is a detailed block diagram of the learning performance enhancing module shown in FIG. 1. FIG.
3 is a flowchart showing the operation of each module according to a method for enhancing learning performance of machine learning according to an embodiment of the present invention.
4 is a diagram showing an example of a background image applied in the present invention.
5 is a diagram showing an example of digital noise applied in the present invention.
6 is a diagram illustrating an example of an image segmentation function and an image composition function provided in the learning performance enhancement module according to the embodiment of the present invention.
Figure 7 illustrates the generation of a final training image to be subjected to machine learning by synthesizing a 2D learning image, a background image, and digital noise in a learning performance enhancement module according to an embodiment of the present invention.

It is noted that the technical terms used herein are used only to describe specific embodiments and are not intended to limit the invention. The suffix " module " and " part " for the components used in the following description are given or mixed in consideration of ease of specification, and do not have their own meaning or role.

And, the singular expressions used in the specification include plural expressions unless the context clearly indicates otherwise. In this specification, " comprises " Or " include. &Quot; Should not be construed to encompass the various components or steps described in the specification, and some of the components or portions may not be included, or may include additional components or steps And the like.

Further, in the description of the technology disclosed in this specification, a detailed description of related arts will be omitted if it is determined that the gist of the technology disclosed in this specification may be obscured.

The present invention relates to a learning image synthesis technique for synthesizing a real background image or a geometric pattern and a digital noise to a learning data source in order to solve a learning data lacking inherent in an existing machine learning process, (Robustness / generalization) is proposed. By means of the above learning data mass generation structure, the present invention can generate various learning data environments and conditions.

1 is a block diagram of a learning performance enhancing apparatus for machine learning according to an embodiment of the present invention.

1, the apparatus for enhancing learning performance of a machine learning according to the present invention includes a learning data module 100, a neural network model unit 110, a neural network operation framework 120, a learning performance enhancement module 130, A high-speed cluster computing module 140, and a training model predicting unit 140.

The learning data module 100 provides a learning data source to perform a machine function. The learning data source may be three-dimensional (3D) CAD model data. The three-dimensional (3D) CAD model data may include at least one inorganic object.

The neural network modeling unit 110 may provide a predefined neural network model and the neural network computing framework 120 may provide neural network operations to perform machine learning and prediction on the training data.

The learning performance enhancement module 130 divides at least one inorganic object from the learning data source input from the image learning data module 100 using an image segmentation function to generate a plurality of 2D And generates a learning image (an inorganic object image), synthesizes the generated plurality of 2D learning images with a plurality of background images (background image of a real background or a geometric pattern) and digital noise using an image compositing function, And outputs a learning image for performing the training.

In one embodiment, the learning performance enhancement module 130 recognizes the type of the divided 3D weapon object and automatically generates a plurality of 2D learning images by adjusting the angle, perspective, and direction of the recognized 3D weapon object . The learning performance enhancement module may determine a background image to be combined with the generated plurality of 2D learning images and a digital noise based on the type of the recognized inorganic object, the angle of the adjusted 3D inorganic object, and the perspective and direction.

The synthesis may be based on predetermined conditions (random / rule based). The dividing function divides the digital image into a plurality of pixel sets and divides the image so as to allow meaningful analysis.

The fast cluster computing module 140 applies the learning image generated through the learning performance enhancement module 130 to the neural network model provided by the neural network model unit 120 and the neural network operation provided by the neural network operation framework 110, Machine learning can be performed on the learning image.

The training-model predicting unit 150 may determine whether to use the training model for the learning im- ages based on the treadmill and prediction performed in the fast cluster computing module 140. [

FIG. 2 is a detailed block diagram of the learning-performance-enhancing module 130 shown in FIG.

2, the learning performance enhancement module 130 includes a 2D learning image generation module 10 for generating a plurality of 2D learning images having arbitrary angles, perspective, and directions by disassembling a learning data source (3D CAD model) A background image selection module 20 for selecting at least one background image or a background image of a geometric pattern from the background image storage module 20a and at least one digital noise from the digital noise storage module 30b , A uniform distribution or a Gaussian distribution) of a plurality of 2D learning images, a plurality of the learning images to be used in the treadmill by synthesizing the generated plurality of 2D learning images and the selected at least one background image and digital noise, And a learning image synthesis module 40 for outputting an image.

The 2D learning image generation module 10 is capable of generating various 2D learning images having different angles, perspective and directions from each other by dividing and detecting (boundary detection) learning data, which is a 3D CAD model, have. The decomposition may include removing the background from the training data already having its own background image, extracting the image, or splitting the image. The division includes use of learning data, which is a 3D CAD model, in a plurality of areas.

The background image storage module 20a may store a plurality of background images, which may include a real background image (eg, landscape) and a geometric pattern image (eg, a pattern, a shape, and combinations thereof) have.

The digital noise storage module 30a may store a plurality of digital noises, and the plurality of digital noises may include learning disturbance noise and random noise (normal / uniform / poisson and others).

The background image storage module 20a and the digital noise storage module 30a are separately shown for convenience of explanation, but are not limited thereto. That is, a plurality of background images and a plurality of digital noises may be stored in different storage areas, respectively, in one memory. In this case, the memory may not be shown in FIG.

The learning image compositing module 40 may include an image compositing function to synthesize the generated plurality of 2D learning images with the selected at least one background image and digital noise. In this case, the learning image synthesis module 30 can synthesize or synthesize a 2D learning image, a background image, and noise sequentially.

The operation of the apparatus for enhancing learning performance of machine learning according to an embodiment of the present invention will now be described with reference to the accompanying drawings. In particular, since the feature of the present invention resides in the learning performance enhancement model 130, it will be mainly described.

 FIG. 3 is a flowchart illustrating an operation of a learning performance enhancing apparatus for machine learning according to an embodiment of the present invention. FIG. 4 is an example of a background image, and FIG. 5 is an example of digital noise.

Referring to FIG. 3, the learning performance enhancement module 130 may receive a 3D CAD model, which is a learning data source to be used for machine learning, from the learning data module 100. The 3D CAD model is a 3D image related to the weapon system, and may be an individual weapon object image such as a tank, an aircraft, a gun, a position, etc., or a combination thereof (S100).

The learning performance enhancing module 130 recognizes / divides at least one image to be used for machine learning from a learning data source (3D CAD model) using an image segmentation function, and transforms and rearranges the divided images to generate a plurality of 2D learning images (S110). The segmentation may include segmentation of a weapon object image or background image. The modification of the image may include modifying the size enlargement / reduction, angle, perspective and direction of the divided inorganic object. The relocation may include moving and / or replacing a partitioned object image or a background image.

In one embodiment, the 2D learning image generation module 20 of the learning performance enhancement module 130 may generate a plurality of 2D learning images based on the detailed settings input by the operator. The detailed settings may include, but are not limited to, an enlargement / reduction, an angle, a perspective, an orientation, a rearrangement position, and the like of an inorganic object to be divided. In this case, the learning performance enhancement module 130 may display a separate user interface (UI) for user selection or provide a menu including a plurality of selection items.

In another embodiment, the learning performance enhancement module 130, more specifically, the 2D learning image generation module 20, may acquire, from the learning data source (3D CAD model) through the object recognition technique, A plurality of 2D learning images can be generated by differently changing the angle, perspective, and direction according to the type of the recognized weapon object after recognizing the aircraft, the aircraft, the ball, the center, and the like. For example, in the case of a tank, it does not create an image that shows the back side, but in the case of an aircraft, it produces an image that shows the back side. Particularly, when the object is recognized from the learning data source (3D CAD model), the 2D learning image generation module 20 can provide the type of the recognized object to the background image selection module 20 and the digital noise selection module 3 .

Also, the learning performance enhancement module 130 can select at least one background image and digital noise corresponding to the plurality of 2D learning images generated by the 2D learning image generation module 20 using the image selection function (S120 , S130). The selection means selecting a plurality of background images stored in a storage medium (e.g., a storage module) and at least one background image matching the recognized object type among digital noises.

That is, in one embodiment, the background image selection module 20 of the learning performance enhancement module 130 determines the type of the inorganic object based on the type (eg, tram) of the inorganic object input from the 2D learning image generation module 20 At least one related background image may be selected from the background image storage module 20a. The background image may include an image with a real background image and / or a geometric pattern, as shown in FIG. The background image may be an image that takes into account time (e.g., morning, afternoon, or night) and weather (cloudy, bright, rain, snow and others).

In one embodiment, the digital noise selection module 30 of the learning-performance enhancement module 130 generates a digital noise image by using the type of the inorganic object (e.g., tram) input from the 2D learning image generation module 20, At least one digital noise relating to the digital noise can be selected from the digital noise storage module 30a. The digital noise may be noise that interferes with machine learning of the recognized object, and may include predetermined learning disturbance noise or random noise (e.g., normal, uniform, poisson noise) as shown in FIG.

A plurality of 2D learning images generated / selected in the learning performance enhancement module 130, at least one background image and digital noise may be synchronized with each other. That is, in FIG. 2, each of the plurality of 2D learning images, at least one background image, and digital noise may be input to the learning image synthesis module 40 in synchronism with each other. 6, the learning performance enhancing module 130 generates various 2D learning images by synthesizing the plurality of 2D learning images with at least one background image and digital noise using the image compositing function, (S140). At this time, different background images and digital noise may be combined with different 2D learning images, and one or more background images and digital noise that are identical to each other in a 2D learning image may be synthesized.

The generated various 2D learning images are provided to the rear end 140, 150 of FIG. 1 and used for machine learning. The synthesis may be based on predetermined conditions (random / rule based).

FIG. 7 shows an example of generating a final learning image to be used in machine learning by synthesizing a 2D learning image, a background image, and digital noise in a learning performance enhancing module according to an embodiment of the present invention.

In one embodiment, the learning image composition module 40 of the learning performance enhancement module 130 determines the specificity (angle, perspective and direction) of each 2D training image (e.g., an inorganic object) The digital noise can be synthesized in a specific order or randomly synthesized based on each type of digital noise.

For example, a first method of synthesizing a background image to a 2D learning image and then synthesizing a digital noise to a synthesized image, and a first method of synthesizing a digital noise to a background image And a second method of synthesizing the 2D training image after the synthesis. The first synthesis method is applied when the specificity (angle, perspective and direction) of the generated 2D learning image (e.g., an inorganic object) is greater than a preset threshold value, so that accurate composition with the background image is required.

When using the first synthesis method, the learning image synthesis module 40 combines, for example, n 2D learning images and m different real background images (or geometric pattern images) to obtain n × m 2D learning images And a total of n × m × p learning images can be generated by combining p different noise (non-uniform distribution, Gaussian distribution, etc.) in the generated n × m 2D learning images. Since the generated final learning image includes various angles and backgrounds and information that can be robust to noise, the learning performance can be enhanced by using the final generated learning image.

The second synthesis method is mainly used when the specificity of the generated 2D training image (e.g., inorganic object) is less than the threshold value (general angle, perspective and direction) but the specificity of the background image (or digital noise) . To this end, the 2D learning image generation module 10 and the background image selection module 20 generate information (eg, normal or specific) indicating the type of the corresponding image when transmitting a plurality of 2D learning images and a background image, respectively Information), and based on the information, the learning image synthesis module 40 can determine whether the specificity exceeds the threshold value.

On the other hand, the random synthesis is performed when the specificities of the images provided in the 2D learning image generation module 10 and the background image selection module 20 (or the digital noise selection module) are all below the threshold, Weapon object), background image, and digital noise.

In the embodiment of the present invention, for the sake of convenience of description, when a learning data source, that is, an inorganic object of the 3D CAD model is provided, the learning performance enhancing module 130 recognizes the type of the inorganic object, A plurality of 2D learning images are generated based on the type of the 2D learning image, and at least one background image and a digital image corresponding to the state information of the generated 2D learning image are selected. However, the present invention is not limited to this, and the learning performance enhancing module 130 may selectively provide a separate user interface (e.g., a setting pop-up menu) when generating a plurality of 2D learning images, Can be specified. The user interface (eg, setting pop-up menu) may include a window for inputting keyword commands, or may include a plurality of image creation types (eg, angle, perspective and direction) And may include an item list.

In some implementations, if the user interface is configured as a window into which a keyword command can be entered, the operator can enter a keyword such as " T-30 tank mountain battle ", for example, The learning image generation module 10 generates a T-30 tank image having various angles, directions and attitudes, and based thereon, the background image selection module 20 and the digital noise selection module 30 generate at least one mountain background image And digital noise.

On the other hand, when the user interface providing the pop-up menu is used, the image creation type (eg, angle, perspective and direction), the type of the background image, and the type of the digital noise May be appropriately selected to limit the types of images to be used in the rear stage (high-speed cluster computing module, training model prediction section). The above method can be appropriately selected and used according to the needs of the operator.

The learning performance enhancement module 130 does not always provide a user interface when a learning data source, that is, an inorganic object of the 3D CAD model, is input. The provision of the user interface may be activated / deactivated according to the operator's initial setting, and may be configured to disappear when there is no input from the user for a preset time even if the user interface is activated and displayed.

The fast cluster computing module 140 applies various learning images generated through the learning performance enhancement module 130 to the neural network model provided by the neural network model unit 120 and the neural network operation provided by the neural network operation framework 110, Run. That is, the high-speed cluster computing module 140 extracts features necessary for inorganic object recognition from the various 2D learning images generated by the learning-performance enhancing module 13, and extracts features necessary for recognizing the inorganic objects from the neural network model provided by the neural model unit 110, And performs pattern learning on the extracted feature using the neural network operation provided by the neural network operation.

The training model predicting unit 150 determines an inorganic object using the pattern learning result performed by the high-speed cluster computing module 140, and determines whether to apply the training model to the learning image.

As described above, the present invention provides a learning data source, i.e., a learning data source (e.g., an inorganic object) of a 3D CAD model to generate a 2D learning image having various angles, perspective and directions, An advantage of solving the lack of learning data inherent in the existing machine learning process by generating the learning image, the at least one background image (the background image of the real background image or the geometric pattern) and the final learning image synthesizing the digital noise have.

The present invention provides a learning data mass generation structure capable of generating various learning images by synthesizing a background image of a real background image or a geometric pattern and digital noise on a learning data source as described above, (Robustness / generalization) can be improved.

In addition, the present invention provides an interface for creating various learning data environments and conditions when various learning images are generated by composing a background image of a real background image or a geometric pattern and a digital noise to a learning data source, It is possible to generate various learning images.

The apparatus and method for enhancing the learning performance of the machine learning according to the present invention can be applied to the embodiments described above without departing from the technical idea or essential features of the present invention. It will be understood that the invention may be practiced. Therefore, it should be understood that the above-described embodiments are to be considered in all respects as illustrative and not restrictive.

10: 2D learning image generation module 20: background image selection module
20a: background image storage module 30: design noise selection module
30a: digital noise storage module 40: learning image synthesis module
100: learning data module 110: neural network model part
120: neural network operation framework 130: learning performance enhancement module
140: high-speed cluster computing module 150: training model prediction unit

Claims (18)

A learning data module for outputting learning data of a 3D CAD model including at least one inorganic object;
A plurality of 2D inorganic object images having different angles, perspectives, and directions are generated based on the type of the 3D inorganic object included in the learning data, and the generated plurality of 2D inorganic object images are combined with one or more background images and digital noise Learning performance enhancement module for generating images; And
And a fast cluster computing module for performing pattern learning according to a neural network model and a neural network operation provided in the neural network model unit and the neural network operation framework and machine learning for recognizing the inorganic object, Learning performance enhancement device. 2. The method of claim 1,
A background image of a real background image or a background image of a geometric pattern,
Wherein the digital noise includes learning disturbance noise or random noise. The apparatus of claim 1, wherein the learning performance enhancing module
A plurality of 2D learning images are automatically adjusted by differentiating the angle, perspective and direction of the 3D inorganic object based on the recognized type by dividing the inorganic object to be subjected to machine learning from the learning data of the 3D CAD model, The learning performance of the machine learning apparatus is improved. 4. The apparatus of claim 3, wherein the learning performance enhancement module
Wherein when the learning data of the 3D CAD model includes a unique background image, the background image is removed to divide the inorganic object. 4. The apparatus of claim 3, wherein the learning performance enhancement module
Selecting one or more background images and digital noise associated with the generated plurality of 2D learning images from the storage medium based on the type of the recognized inorganic object and the specificity of the adjusted 3D inorganic object,
Wherein the singularity includes an angle, a perspective and a direction. 6. The apparatus of claim 5, wherein the learning performance enhancement module
Determining a specificity of the digital noise type from the selected one or more background images from the storage medium and selecting the sequential or random combination based on the specificity of each of the plurality of 2D learning images and the specificity of the determined background image and the digital noise type Learning performance of the machine learning apparatus. The apparatus of claim 1, wherein the learning performance enhancing module
A 2D learning image generation module for generating a plurality of 2D learning images in which an arbitrary angle, perspective and direction are different from each other by disassembling an inorganic object from learning data of a 3D CAD model;
A background image selection module for selecting at least one background image or a background image of a geometric pattern from the background image storage module;
A digital noise selection module for selecting at least one digital noise from the digital noise storage module; And
And a learning image synthesis module for synthesizing the generated plurality of 2D learning images, the selected at least one background image and digital noise, and outputting a plurality of learning images to be used in the treadmill Learning performance enhancing device. A method for segmenting and detecting an inorganic object from learning data of a 3D CAD model including at least one inorganic object and generating a plurality of 2D inorganic object images with angle, perspective and orientation adjusted based on the detected type of the inorganic object Learning image generation module;
A background image selection module for selecting at least one background image from the first storage module;
A digital noise selection module for selecting at least one digital noise from the second storage module; And
A learning image synthesis module for synthesizing a 2D inorganic object image, a background image, and digital noise based on the generated plurality of 2D inorganic object images and the specificity of at least one background image to generate a plurality of learning images to be used for machine learning; And the learning performance of the machine learning apparatus is improved. 9. The method of claim 8,
A background image of a real background image or a background image of a geometric pattern,
Wherein the digital noise includes learning disturbance noise or random noise. The method of claim 8, wherein the 2D learning image generation module
An inorganic object to be subjected to machine learning is divided from learning data of the 3D CAD model to recognize the type,
A plurality of 2D learning images are automatically generated by differently adjusting the angle, perspective and direction of the 3D inorganic object based on the recognized type,
Wherein the learning performance enhancement module divides the inorganic object by removing the background image if the learning data of the 3D CAD model includes a unique background image. The method of claim 8, wherein the 2D learning image generation module
A background image selection module and a digital noise selection module are informed of the type of the recognized inorganic object and the specificity of the adjusted 3D inorganic object,
Wherein the background image selection module and the digital noise selection module select one or more background images and digital noise associated with the generated plurality of 2D learning images from the first and second storage media, respectively, based on the type and the specificity of the inorganic entity In addition,
Wherein the singularity includes an angle, a perspective and a direction. The method of claim 8, wherein the learning image synthesis module
The background image selection module determines the specificity of at least one background image type selected by the digital noise selection module,
Wherein the sequential synthesis or the random synthesis is selectively performed based on the specificity of each of the plurality of 2D learning images and the specificity of the discriminated background image type. 13. The apparatus of claim 12, wherein the learning image synthesis module
When the specificity of each of the 2D learning images exceeds a threshold value, synthesizes a background image to a 2D learning image and finally synthesizes a digital noise,
When the specificity of each of the 2D learning images is less than the threshold value and the specificity of the background image type is equal to or greater than the threshold value, the 2D learning image is synthesized with the background image,
Wherein the 2D learning image, the background image, and the digital noise are randomly synthesized when the specificities of the 2D learning image and the background image are all equal to or less than the threshold value. Outputting learning data of a 3D CAD model including an inorganic object; And
Generating a plurality of 2D inorganic object images by adjusting the specificity of the 3D inorganic objects based on the type of the 3D inorganic objects included in the learning data, synthesizing the generated plurality of 2D inorganic object images with one or more background images and digital noise And generating a learning image,
The step of generating the learning image
Recognizing a type of a specific inorganic object segmented from learning data of the 3D CAD model and generating a plurality of 2D learning images by differently adjusting angles, perspectives and directions of the 3D inorganic object based on the recognized types; And
Selecting one or more background images and digital noise associated with the plurality of 2D learning images based on the type of the recognized inorganic object and the specificity of the adjusted 3D inorganic object; How to enhance learning performance. 15. The method of claim 14,
A background image of a real background image or a background image of a geometric pattern,
Wherein the digital noise includes learning disturbance noise or random noise. 15. The method of claim 14, wherein generating the plurality of 2D learning images comprises:
If the learning data of the 3D CAD model includes a unique background image, dividing the inorganic object by removing the background image,
Wherein the specificity includes an angle, a perspective and a direction. 15. The method of claim 14, wherein generating the training image comprises:
Determining the specificity of the selected at least one background image type; And
And performing sequential synthesis or random synthesis based on the specificity of each of the plurality of 2D learning images and the specificity of the discriminated background image type. 18. The method of claim 17, wherein performing the sequential or random synthesis comprises:
Synthesizing a background image in a 2D learning image and finally synthesizing a digital noise when the specificity of each of the 2D learning images exceeds a threshold value;
Synthesizing a 2D learning image on a background image when the specificity of each of the 2D learning images is less than a threshold value and the specificity of a background image type is more than a threshold value, and finally synthesizing digital noise; And
And randomly synthesizing the 2D learning image, the background image, and the digital noise when the specificities of the 2D learning image and the background image are both equal to or less than the threshold value.

Images