Image training set generation method and device based on data enhancement

Abstract

The application provides a kind of based on data enhancement's image training set generation method and device, method, comprising: obtaining image training data, image training data includes multiple frames training image and the specific target feature corresponding to each frame training image, image training data includes positive sample data and negative sample data;According to specific target feature, each frame training image is marked with feature, and first data enhancement strategy is used to the feature marked area Data enhancement is carried out, and the negative sample data is updated;According to positive sample data and the updated negative sample data, obtain image training set.The specific target of each frame training image in the training image data is enhanced in the region of the application, to solve the overfitting problem in the model training process due to insufficient training data, it is favorable to adjust the difficulty degree of recognition network training, improve the generalization ability of deep learning classification network.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image recognition, and in particular to an image training set generation method and device based on data enhancement. BACKGROUND

[0002] In the field of computer vision, a target detection model based on deep learning is usually used for decision-making. In order to achieve an ideal training effect of the target detection model, an image training set containing a large number of image samples needs to be used in the model training process. In the case of insufficient image samples, data enhancement means is needed to expand the image training set.

[0003] At present, data enhancement is mostly used to enhance the entire image to obtain more image sample data. Although this data enhancement method can effectively expand the training data set, it does not significantly help to improve the recognition accuracy of the target recognition model for specific targets, thereby resulting in certain use limitations of the existing target detection model. SUMMARY

[0004] The present application provides an image training set generation method and device based on data enhancement, which solves the problem of poor recognition accuracy of specific objects in the prior art, improves the learning ability of the network for specific objects, and greatly improves the convergence speed.

[0005] The present application provides an image training set generation method based on data enhancement, comprising: obtaining image training data, wherein the image training data comprises a plurality of training images and specific target features corresponding to each frame of the training images, and the image training data comprises positive sample data and negative sample data; according to the specific target features, performing feature labeling on each frame of the training images, and performing data enhancement on the feature labeling area using a first data enhancement strategy, and updating the negative sample data; obtaining an image training set according to the positive sample data and the updated negative sample data.

[0006] According to the image training set generation method based on data enhancement provided by the present application, the first data enhancement strategy comprises at least one of occlusion, contrast transformation, noise, blur, color transformation, erasing and filling.

[0007] According to the image training set generation method based on data enhancement provided by the present application, the feature labeling of each frame of the training images according to the specific target features comprises: using a bounding box labeling method to label the specific target in each frame of the training images according to the specific target features; or,

[0008] According to the specific target features, the specific targets in each frame of the training images are marked by using a point connection marking method.

[0009] According to the image training set generation method based on data enhancement provided by the application, after the feature marking of each frame of the training images, the method further comprises: according to the feature marking area, using a second data enhancement strategy to perform data enhancement on the area outside the feature marking area in each frame of the training images.

[0010] According to the image training set generation method based on data enhancement provided by the application, the second data enhancement strategy comprises at least one of flipping, rotating, cropping, deforming and zooming; and / or,

[0011] The second data enhancement strategy comprises at least one of noise, blur, color transformation, erasing and filling.

[0012] According to the image training set generation method based on data enhancement provided by the application, the acquisition of the image training data comprises: acquiring image data; performing specific target identification on the image data to obtain image training data; and dividing the image training data into positive sample data and negative sample data.

[0013] The application further provides an image training set generation device based on data enhancement, comprising: a data acquisition module, which acquires image training data, wherein the image training data comprises multiple frames of training images and specific target features corresponding to each frame of the training images, and the image training data comprises positive sample data and negative sample data; a first data enhancement module, which performs feature marking on each frame of the training images according to the specific target features, performs data enhancement on the feature marking area by using a first data enhancement strategy, and updates the negative sample data; and an image generation module, which obtains an image training set according to the positive sample data and the updated negative sample data.

[0014] The application further provides an electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the image training set generation method based on data enhancement according to any one of the above when executing the program.

[0015] The application further provides a non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program is executable by a processor to implement the steps of the image training set generation method based on data enhancement according to any one of the above.

[0016] The application further provides a computer program product comprising a computer program, wherein the computer program is executable by a processor to implement the steps of the image training set generation method based on data enhancement according to any one of the above.

[0017] The application provides a data enhancement-based image training set generation method and device, which performs data enhancement on a region where a specific target is located in each frame of training image in training image data, so as to increase negative sample data in which the specific target is unclear, and then controls the mixed negative sample data and original image training data to form an image training set, so as to solve the overfitting problem in the model training process caused by insufficient training data, and facilitate adjustment of the difficulty of recognition network training and improvement of the generalization ability of the deep learning classification network. In addition, by increasing the data amount of negative samples, the time and computing power consumption caused by excessive positive sample data in model training is avoided, which is beneficial to targeted increase of network learning of specific targets, makes the model converge faster and better, greatly improves the recognition accuracy of the model trained by the image training set for specific targets, and further ensures the practicability of the model. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0019] Figure 1 is a flowchart of the data enhancement-based image training set generation method provided by the application;

[0020] Figure 2 is a structural schematic diagram of the data enhancement-based image training set generation device provided by the application;

[0021] Figure 3 is a structural schematic diagram of the electronic device provided by the application. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical scheme and advantages of the application more clear, the technical scheme in the application will be described clearly and completely in combination with the drawings in the application. Obviously, the described embodiments are some embodiments of the application, not all embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor belong to the protection scope of the application.

[0023] Figure 1 shows a flowchart of a data enhancement-based image training set generation method of the application, and the method comprises:

[0024] S01, obtain image training data, the image training data comprising a plurality of frames of training images and specific target features corresponding to each frame of training image, the image training data comprising positive sample data and negative sample data;

[0025] S02, according to the specific target feature, each frame of training image is labeled, and the first data enhancement strategy is used for data enhancement on the feature labeled area, and the negative sample data is updated;

[0026] S03, according to the positive sample data and the updated negative sample data, the image training set is obtained.

[0027] It should be noted that S0N in the present specification does not represent the order of the image training set generation method based on data enhancement, and the image training set generation method based on data enhancement of the present application will be described in detail below.

[0028] Step S01, obtaining image training data, the image training data comprising a plurality of frames of training images and specific target features corresponding to each frame of training image, the image training data comprising positive sample data and negative sample data.

[0029] In the embodiment, the image training data is obtained, comprising: obtaining image data; performing specific target recognition on the image data to obtain the image training data; dividing the image training data into positive sample data and negative sample data, wherein the positive sample data and the negative sample data respectively contain different number of training images and corresponding specific target features.

[0030] It should be noted that the obtained image data can be derived from millimeter wave radar, laser radar, detector, camera and other camera equipment based on specific target shooting image, which is not limited further herein. The image data can be understood as a collection of picture data, which can be for at least one specific target, and the picture data corresponding to each specific target is in different angles, different pixel colors and other states. The picture data is usually large, which can reach millions of levels.

[0031] In addition, the specific target includes a target scene that is difficult to identify, such as smoke generated when smoking. The specific target feature can be obtained by using a trained deep learning model to identify or detect the image data.

[0032] For example, if the obtained image data is a vehicle driver seat scene target scene image, and the scene captures the driver's fingers and the smoke on the fingers, or captures the driver's fingers and the smoke near the fingers, the smoke or smoke is the specific target identified.

[0033] In step S02, according to the specific target feature, the feature label is marked on each frame of training image, and the first data enhancement strategy is used for data enhancement on the feature labeled region, and the negative sample data is updated.

[0034] In this embodiment, according to the specific target feature, the feature label is marked on each frame of training image, including: according to the specific target feature, using the detection box marking method to mark the specific target in each frame of training image; or, according to the specific target feature, using the point connection line marking method to mark the specific target in each frame of training image. It should be noted that the training image for feature labeling can be each frame of training image in the image training data, or each frame of training image in the positive sample data, which can be determined according to the actual required amount of negative sample data, and is not limited further here.

[0035] By marking the specific target in each frame of training image, the region in the training image where the specific target is located is marked for subsequent data enhancement, so as to expand the amount of negative sample data, and further solve the problem of overfitting in model training due to insufficient training data, which is beneficial to the subsequent model training process, and greatly improves the convergence speed of the model learning specific target features.

[0036] Further, the first data enhancement strategy can adopt at least one of the color transformation methods, specifically including at least one of occlusion, contrast transformation, noise, blur, color transformation, erasing and filling. For example, the first data enhancement strategy includes occlusion, contrast transformation, noise, blur, color transformation, erasing or filling; for another example, the first data enhancement strategy includes at least two of occlusion, contrast transformation, noise, blur, color transformation, erasing and filling. By using the first data enhancement strategy to enhance the data of the feature labeled region, the learning ability of the subsequent model for the specific target is improved, so as to improve the recognition accuracy of the model.

[0037] It should be noted that the feature labeled region can be blurred by using a blur processing method, which can be adjusting the gray value of the feature labeled region, or sharpening the feature labeled region. In addition, the feature labeled region can also be filled by using image filling method, such as filling the feature labeled region with color, other patterns or other ways of affecting the image pixels of the feature labeled region, which is not limited further here.

[0038] In an optional embodiment, after the feature marking of each frame of training image, further comprising: according to the feature marking region, using a second data augmentation strategy to perform data augmentation on the region outside the feature marking region in each frame of training image. The second data augmentation strategy can adopt various existing data augmentation methods to perform data augmentation on the part outside the feature marking region in the training image, so as to increase the amount of negative sample data, so as to facilitate the improvement of the learning of the model to the specific target.

[0039] It should be noted that, after the feature marking region is subjected to data augmentation by the first data augmentation strategy, the region outside the feature marking region in each frame of training image can be subjected to data augmentation by the second data augmentation strategy according to the feature marking region; or, while the feature marking region is subjected to data augmentation by the first data augmentation strategy, the region outside the feature marking region in each frame of training image can be subjected to data augmentation by the second data augmentation strategy according to the feature marking region; or, before the feature marking region is subjected to data augmentation by the first data augmentation strategy, the region outside the feature marking region in each frame of training image can be subjected to data augmentation by the second data augmentation strategy according to the feature marking region.

[0040] Specifically, the second data augmentation strategy includes at least one of flipping, rotating, cropping, deforming and scaling; and / or, the second data augmentation strategy includes at least one of noise, blur, color transformation, erasing and filling. In actual selection of the second data augmentation strategy, any one of flipping, rotating, cropping, deforming or scaling can be selected, any one of noise, blur, color transformation, erasing and filling can be selected, or at least two of flipping, rotating, cropping, deforming and scaling are selected; or, at least two of noise, blur, color transformation, erasing and filling are selected; or, at least two of flipping, rotating, cropping, deforming, scaling, noise, blur, color transformation, erasing and filling are selected.

[0041] Step S03, obtaining an image training set according to the positive sample data and the updated negative sample data.

[0042] It should be noted that, the negative sample for expanding the negative sample data is obtained by the first data augmentation strategy, or the negative sample for expanding the negative sample data is obtained by the first data augmentation strategy and the second data augmentation strategy, so as to expand the negative sample data in the training image data, thereby increasing the amount of negative sample data, facilitating the subsequent training of the model by using the training image data, improving the learning ability of the model to the specific target, and improving the accuracy of the model in identifying the specific target.

[0043] In an optional embodiment, after obtaining the image training set, the method further comprises: using the image training set as input data for training, using the specific target features corresponding to the frames of the image training set as labels, training the to-be-trained network, and obtaining a target recognition model for recognizing the specific target in the image.

[0044] To sum up, the present application increases the negative sample data of the specific target by data enhancement on the region where the specific target is located in each frame of training image in the training image data, and then controls the mixed formation of the increased negative sample data and the original image training data to form an image training set, so as to solve the overfitting problem in the model training process caused by insufficient training data, which is beneficial to adjust the difficulty of the recognition network training and improve the generalization ability of the deep learning classification network. In addition, by increasing the data amount of the negative sample, the time and computing power consumption caused by too much positive sample data for model training is avoided, which is beneficial to targeted increase of the network learning of the specific target, so that the model converges faster and better, greatly improving the recognition accuracy of the model trained by the image training set for the specific target, and further ensuring the practicability of the model.

[0045] The data enhancement-based image training set generation device provided by the present application is described below, and the data enhancement-based image training set generation device described below can be correspondingly referred to the data enhancement-based image training set generation method described above.

[0046] Figure 2 A structure schematic diagram of a data enhancement-based image training set generation device is shown, the device comprises:

[0047] The data acquisition module 21 acquires image training data, the image training data comprising multiple frames of training images and specific target features corresponding to each frame of training image, and the image training data comprising positive sample data and negative sample data.

[0048] The first data enhancement module 22 performs feature labeling on each frame of training image according to the specific target features, performs data enhancement on the feature labeled region using a first data enhancement strategy, and updates the negative sample data.

[0049] The image generation module 23 obtains an image training set according to the positive sample data and the updated negative sample data.

[0050] In the embodiment, the data acquisition module 21 comprises: an acquisition unit that acquires image data; an identification unit that performs specific target identification on the image data to obtain image training data; and a division unit that divides the image training data into positive sample data and negative sample data, wherein the positive sample data and the negative sample data respectively contain different amounts of training images and corresponding specific target features.

[0051] It should be noted that the acquired image data can be derived from images captured by millimeter wave radar, laser radar, detectors, cameras and other image capturing devices based on specific targets, and the source of the image data is not further limited here. The image data can be understood as a collection of picture data, which can be for at least one specific target, and corresponds to pictures of each specific target in different angles, different pixel colors, etc. The amount of picture data is usually large, up to millions.

[0052] In addition, the specific target includes a target that is difficult to identify in a target scene, such as smoke generated when smoking. The specific target identification of the image data can be identified or detected using a trained deep learning model to obtain the specific target features.

[0053] The first data augmentation module 22 includes a marking unit that marks each frame of training image according to the specific target features; a first data augmentation unit that performs data augmentation on the feature marked region using a first data augmentation strategy; and a data updating unit that updates the negative sample data.

[0054] Further, the marking unit includes a marking subunit configured to mark the specific target in each frame of training image according to the specific target features using a bounding box marking method, or to mark the specific target in each frame of training image according to the specific target features using a point connection marking method. It should be noted that the training images for feature marking can be each frame of training image in the image training data, or each frame of training image in the positive sample data, which can be determined according to the actual required amount of negative sample data, and is not further limited here.

[0055] By marking the specific target in each frame of training image, the region in the training image where the specific target is located can be data-augmented based on the marking, so as to expand the amount of negative sample data, thereby solving the problem of overfitting in model training due to insufficient training data, and facilitating the model to learn the specific target features and greatly improving the convergence speed in the subsequent model training process.

[0056] Further, the first data augmentation strategy can use at least one of the color transformation methods, specifically including at least one of occlusion, contrast transformation, noise, blur, color transformation, erasing and filling. For example, the first data augmentation strategy includes occlusion, contrast transformation, noise, blur, color transformation, erasing or filling; for another example, the first data augmentation strategy includes at least two of occlusion, contrast transformation, noise, blur, color transformation, erasing and filling. By using the first data augmentation strategy to perform data augmentation on the feature marked region, the learning ability of the subsequent model for the specific target is improved, thereby improving the recognition accuracy of the model.

[0057] It should be noted that the feature mark region can be blurred in a blur processing manner. The blur processing can be adjusting the gray value of the feature mark region or sharpening the feature mark region. In addition, the feature mark region can also be filled in an image filling manner, such as filling the feature mark region with color, other patterns, or other ways of affecting the image pixels of the feature mark region, which is not limited further herein.

[0058] In an optional embodiment, the device further comprises a second data enhancement module configured to perform data enhancement on the region outside the feature mark region in each frame of training image according to the feature mark region and using a second data enhancement strategy. The second data enhancement strategy can use various existing data enhancement methods to perform data enhancement on the part outside the feature mark region in the training image, thereby increasing the amount of negative sample data to facilitate the learning of the model on the specific target.

[0059] It should be noted that the data enhancement on the region outside the feature mark region in each frame of training image according to the feature mark region and using the second data enhancement strategy can be performed after the data enhancement on the feature mark region using the first data enhancement strategy, simultaneously with the data enhancement on the feature mark region using the first data enhancement strategy, or before the data enhancement on the feature mark region using the first data enhancement strategy.

[0060] Specifically, the second data enhancement strategy includes at least one of flipping, rotating, cropping, deforming, and scaling; and / or, the second data enhancement strategy includes at least one of noise, blur, color transformation, erasing, and filling. In actual selection of the second data enhancement strategy, any one of flipping, rotating, cropping, deforming, or scaling can be selected as a strong strategy, any one of noise, blur, color transformation, erasing, and filling can be selected, or at least two of flipping, rotating, cropping, deforming, and scaling are selected; or at least two of noise, blur, color transformation, erasing, and filling are selected; or at least two of flipping, rotating, cropping, deforming, scaling, noise, blur, color transformation, erasing, and filling are selected.

[0061] The image generation module 23 comprises an image generation unit configured to obtain an image training set according to the positive sample data and the updated negative sample data. It should be noted that the negative sample used for expanding the negative sample data is obtained by the first data expansion strategy, or the negative sample used for expanding the negative sample data is obtained by the first data expansion strategy and the second data expansion strategy, so as to expand the negative sample data in the training image data, thereby increasing the data amount of the negative sample, facilitating subsequent training of the model by using the training image data, improving the learning ability of the model to the specific target, and improving the accuracy of the model in identifying the specific target.

[0062] In an optional embodiment, the apparatus further comprises a training module configured to use the image training set as input data for training, use the specific target features corresponding to each frame of the training image in the image training set as labels, train the network to be trained, and obtain a target recognition model used for identifying the specific target in the image.

[0063] Figure 3 An example of an entity structure diagram of an electronic device is shown in Figure 3 As shown, the electronic device can include a processor 31, a communications interface 32, a memory 33, and a communications bus 34, wherein the processor 31, the communications interface 32, and the memory 33 communicate with each other through the communications bus 34. The processor 31 can invoke the logical instructions in the memory 33 to execute the image training set generation method based on data expansion, which comprises: obtaining image training data, the image training data comprising multiple frames of training images and specific target features corresponding to each frame of the training image, the image training data comprising positive sample data and negative sample data; according to the specific target features, performing feature labeling on each frame of the training image, and performing data expansion on the feature labeling region by using a first data expansion strategy, and updating the negative sample data; and obtaining an image training set according to the positive sample data and the updated negative sample data.

[0064] In addition, the logic instructions in the memory 33 described above can be implemented in the form of a software function unit and sold or used as an independent product, which can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions for making a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0065] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to execute the data enhancement-based image training set generation method provided by the above-mentioned methods. The method comprises: obtaining image training data, the image training data comprising a plurality of training images and specific target features corresponding to each training image, and the image training data comprising positive sample data and negative sample data; performing feature labeling on each training image according to the specific target features, performing data enhancement on the feature labeling region by using a first data enhancement strategy, and updating the negative sample data; and obtaining an image training set according to the positive sample data and the updated negative sample data.

[0066] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to execute the data enhancement-based image training set generation method provided by the above-mentioned methods. The method comprises: obtaining image training data, the image training data comprising a plurality of training images and specific target features corresponding to each training image, and the image training data comprising positive sample data and negative sample data; performing feature labeling on each training image according to the specific target features, performing data enhancement on the feature labeling region by using a first data enhancement strategy, and updating the negative sample data; and obtaining an image training set according to the positive sample data and the updated negative sample data.

[0067] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0068] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0069] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for generating image training sets based on data augmentation, characterized in that, include: Acquire image training data, which includes multiple training images and specific target features corresponding to each training image, and the image training data includes positive sample data and negative sample data; Based on the specific target features, feature labels are applied to each frame of the training image, and the feature-labeled regions are augmented using a first data augmentation strategy, and the negative sample data is updated. Based on the positive sample data and the updated negative sample data, an image training set is obtained; Obtaining the image training data includes: Acquire image data; Specific target recognition is performed on the image data to obtain image training data; The image training data is divided into positive sample data and negative sample data, wherein the positive sample data and the negative sample data each contain a different number of training images and their corresponding specific target features; After performing feature labeling on each frame of the training images, the method further includes: Based on the feature-labeled regions, a second data augmentation strategy is used to augment the regions outside the feature-labeled regions in each frame of the training image.

2. The image training set generation method based on data augmentation according to claim 1, characterized in that, The first data augmentation strategy includes at least one of occlusion, contrast transformation, noise, blurring, color transformation, erasure, and filling.

3. The image training set generation method based on data augmentation according to claim 1, characterized in that, The step of labeling the training images in each frame according to the specific target features includes: Based on the specific target features, the specific targets in each frame of the training images are labeled using a detection box labeling method; or... Based on the specific target features, the specific targets in each frame of the training image are labeled using a dot-connection labeling method.

4. The image training set generation method based on data augmentation according to claim 1, characterized in that, The second data augmentation strategy includes at least one of flipping, rotating, cropping, deforming, and scaling; and / or, The second data augmentation strategy includes at least one of noise, blur, color transformation, erasure, and fill.

5. An image training set generation device based on data augmentation, characterized in that, include: The data acquisition module acquires image training data, which includes multiple training images and specific target features corresponding to each training image. The image training data includes positive sample data and negative sample data. The first data augmentation module performs feature labeling on each frame of the training image based on the specific target features, applies a first data augmentation strategy to the feature-labeled regions for data augmentation, and updates the negative sample data. The image generation module obtains an image training set based on the positive sample data and the updated negative sample data; The data acquisition module includes: The acquisition unit acquires image data; The recognition unit performs specific target recognition on the image data to obtain image training data; The partitioning unit divides the image training data into positive sample data and negative sample data, wherein the positive sample data and the negative sample data each contain a different number of training images and their corresponding specific target features; The device further includes: The data augmentation module is used to perform data augmentation on the regions outside the feature-labeled regions in each frame of the training images after feature labeling is performed on each frame of the training images, based on the feature-labeled regions and using a second data augmentation strategy.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the data augmentation-based image training set generation method as described in any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the data augmentation-based image training set generation method as described in any one of claims 1 to 4.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the data augmentation-based image training set generation method as described in any one of claims 1 to 4.

Images