Data augmentation method and related device based on the combination of sample space and feature space
By combining data augmentation in sample space and feature space, the long-term visual feature data augmentation problem is solved, the data set quality and detection performance are improved, and the problem of sample imbalance and model overfitting is avoided.
Patent Information
- Application Number
- CN202410666705.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-28
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-05-28
AI Technical Summary
The existing data augmentation method is difficult to apply to long-term visual feature data, and it is only augmented in the sample space, resulting in limited improvement in detection performance and may even cause the problems of sample imbalance and model overfitting.
Using a data augmentation method based on the combination of sample space and feature space, by obtaining video data sets from different periods, constructing long-term visual feature data sets, and exchanging data in the sample space. The twin neural network trains the model to obtain feature space expression Carry out weighted fusion to complete data augmentation.
Without affecting the difference in data characteristics, the quality of the data set is enhanced, the detection performance of complex data is improved, and the problems of sample imbalance and model overfitting are avoided.
Smart Images

Figure CN118537678B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of deep learning sample augmentation, and in particular to a data augmentation method based on the combination of sample space and feature space and related devices. Background Art
[0002] Data augmentation is a method of increasing the amount of data by adding minor changes to existing data or creating new synthetic data from existing data. In deep learning, data augmentation technology is widely used to alleviate scenarios with insufficient data, thereby improving the generalization ability of the model. At present, the method of obtaining high-quality data has become an important topic in the field of artificial intelligence. Existing data augmentation methods are divided into transformation-based methods, cropping-based methods, and aliasing-based methods. Transformation-based methods target the inherent properties of transformed data, such as replacing the color channels of an image, changing brightness, and changing resolution. Cropping-based methods add new samples by changing part of the data in the sample, such as setting part of the data to 0, adding Gaussian noise, and random erasing. Aliasing-based methods perform data augmentation by splicing, mixing, and superimposing different samples. If samples with different labels are aliased, new labels can be added to the new samples.
[0003] However, in deep learning, some data features are not obvious in a short period of time, such as long-term visual feature data, which often requires long-term tracking and collection. The technology of using the differences in data features in different periods to increase the accuracy of neural network detection is often used in fields such as chronic disease detection. Such data collection is difficult and it is difficult to obtain sufficiently high-quality data for training models. Existing data augmentation methods are difficult to apply to such data, and current data augmentation methods only consider augmenting data in the sample space, which can bring good detection performance for data with obvious features in the sample space, but has limited improvement in detection performance for more complex data, and may even cause problems such as sample imbalance and model overfitting due to data augmentation. Therefore, it is urgent to develop a data augmentation method for long-term visual feature data. Summary of the invention
[0004] The purpose of the present invention is to provide a data augmentation method and related devices based on the combination of sample space and feature space, which can be applicable to data augmentation of long-term visual feature data. By combining the data augmentation methods of sample space and feature space, the quality of the data set obtained by data augmentation is enhanced.
[0005] To achieve the above object, the present invention provides the following solutions:
[0006] In one aspect, the present invention provides a data augmentation method based on the combination of sample space and feature space, comprising the following steps:
[0007] Video datasets of different periods are obtained; each video dataset of a period includes several videos of human activities at different time periods.
[0008] Based on video data sets of different periods, a long-term visual feature data set is constructed; the long-term visual feature data set includes long-term visual feature data of several different time periods; the long-term visual feature data includes visual features of human activity videos of the same time period in different periods.
[0009] For any two long-term visual feature data, the visual features of human activity videos of the same period are exchanged to obtain a long-term visual feature dataset after sample space data augmentation.
[0010] Based on the long-term visual feature dataset after sample space data augmentation, a twin neural network dataset is constructed; the twin neural network dataset includes several sample pairs.
[0011] A neural network model is trained using the twin neural network data set to obtain a trained neural network model; the trained neural network model can accurately predict the label of the sample.
[0012] For any long-term visual feature data in the long-term visual feature data set after the sample space data is augmented, the long-term visual feature data is input into the trained neural network model to obtain the feature space expression of the long-term visual feature data.
[0013] The feature space expressions of any two long-term visual feature data with different labels are weightedly fused to obtain a long-term visual feature dataset after feature space data augmentation.
[0014] Optionally, based on video datasets of different periods, a long-term visual feature dataset is constructed, which specifically includes the following steps:
[0015] For any human activity video in video data sets of different periods, visual features are extracted; the visual features include vector expressions of coordinate matrices of key points of the human body in the human activity video.
[0016] For any time period, the visual features of human activity videos in the time period in video datasets of different periods are stacked to obtain the long-term visual feature data of the time period.
[0017] Based on the long-term visual feature data of each time period, a long-term visual feature dataset is constructed.
[0018] Optionally, by X t-1 ,X t ,X t+1 Represents video datasets of different periods, subscript t represents period t; the video dataset of period t is represented by the following formula:
[0019]
[0020] The superscript i represents the time period i, and the value range of i is [1, l]. represents the visual features of the i-th human activity video at time t.
[0021] For any period of time, the visual features of human activity videos in different periods of video data are stacked according to the following formula:
[0022]
[0023] Among them, X i Represents the long-term visual feature data of time period i.
[0024] Optionally, for any two long-term visual feature data, the visual features of human activity videos of the same period in the two long-term visual feature data are exchanged according to the following formula:
[0025]
[0026] Among them, X i’ and X i+1’ It is a long-term visual feature data sample obtained by data augmentation in the sample space.
[0027] Optionally, a neural network model is trained using the twin neural network data set to obtain a trained neural network model, specifically comprising the following steps:
[0028] Based on the long short-term memory neural network, a neural network model is constructed.
[0029] For any sample pair in the twin neural network dataset, the two sample data of the sample pair are input into the neural network model respectively to obtain a label pair.
[0030] The Euclidean distance of the label pairs is calculated based on the label pairs, and the loss function value of the neural network model is calculated based on the Euclidean distance of the label pairs and the comparison labels of the sample pairs.
[0031] According to the loss function value, back propagation optimizes the parameters of the neural network model.
[0032] If the current number of iterations does not reach the preset number of iterations, jump to step: For any sample pair in the twin neural network dataset, input the two sample data of the sample pair into the neural network model respectively to obtain a label pair.
[0033] If the current number of iterations reaches the preset number of iterations, the neural network model after parameter optimization is used as the trained neural network model.
[0034] Optionally, the loss function value of the neural network model is calculated according to the following formula:
[0035]
[0036] Among them, L is the loss function value of the neural network model; y is the comparison label of the sample pair. If the labels of the two sample data of the sample pair are the same, the comparison label y is 1, otherwise the comparison label y is 0; d is the Euclidean distance of the label pair. For a sample pair (X j ,X k ), d=||y j -y k ||2,y j For sample data X j The label, y k For sample data X k The label is , and margin is the distance threshold.
[0037] Optionally, the feature space expressions of any two long-term visual feature data with different labels are weighted fused according to the following formula:
[0038] X h,label =(1-label)X h,0 +labelX h,1
[0039] Among them, X h,label is a long-term visual feature data sample obtained by data augmentation in the feature space, label is X h,label The label of X, label∈[0.1,0.3]∪[0.7,0.9], h,0 and X h,1 They are the feature space expressions of any two long-term visual feature data with different labels.
[0040] On the other hand, the present invention provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the data augmentation method based on the union of sample space and feature space as described above.
[0041] On the other hand, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the data augmentation method based on the combination of sample space and feature space as described above.
[0042] On the other hand, the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the data augmentation method based on the combination of sample space and feature space as described above.
[0043] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:
[0044] The present invention provides a data augmentation method and related device based on the combination of sample space and feature space, the method includes the following steps: obtaining video data sets of different periods; constructing a long-term visual feature data set based on the video data sets of different periods; and performing data augmentation in the sample space for any two long-term visual feature data; constructing a twin neural network data set based on the data set after the sample space data augmentation, and training a neural network model, inputting the data in the data set after the sample space data augmentation into the trained neural network model to obtain its feature space expression, and finally, in the feature space, weightedly fusion the feature space expressions of any two long-term visual feature data with different labels to complete the data augmentation of the feature space. Based on the long-term visual feature data, the present invention adds newly generated data on the basis of the existing data without affecting the difference in the features of the data in different periods. Compared with other general data augmentation schemes, the data augmentation method combining the sample space and the feature space enhances the data quality of complex data. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0046] Figure 1 A flow chart of a data augmentation method based on the combination of sample space and feature space provided in Example 1 of the present invention;
[0047] Figure 2 A specific flow chart of step S5 in a data augmentation method based on the combination of sample space and feature space provided in Embodiment 1 of the present invention;
[0048] Figure 3 This is a diagram of the internal structure of a computer device provided in Example 4 of the present invention. DETAILED DESCRIPTION
[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0050] The purpose of the present invention is to provide a data augmentation method and related devices based on the combination of sample space and feature space, aiming to perform data augmentation based on the characteristics of long-term visual feature data, and to improve the quality of complex data sets by combining the data augmentation methods of sample space and feature space.
[0051] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0052] Example 1
[0053] like Figure 1 As shown, a data augmentation method based on the combination of sample space and feature space in this embodiment includes the following steps:
[0054] S1. Obtain video datasets of different periods; each period of video dataset includes several human activity videos of different periods. t-1 ,X t ,X t+1 It represents a video dataset of different periods, and the subscript t represents period t. In period t, l videos of human activities are taken at different periods, and each video has n frames. If period t is one day and each period is one hour, then there are 24 periods in period t.
[0055] S2, based on video data sets of different periods, construct a long-term visual feature data set; the long-term visual feature data set includes long-term visual feature data of several different periods; the long-term visual feature data includes visual features of human activity videos of the same period of different periods. Step S2 specifically includes the following steps:
[0056] S21. Extract visual features from any human activity video in the video data set of different periods; the visual features include the vector expression of the human body key point coordinate matrix in the human activity video. For each frame, extract 17 human body key point coordinate matrices, each coordinate is 2-dimensional, and then obtain n 17×2-dimensional human body key point coordinate matrices, flatten the matrix into a 34-dimensional vector, and stack the n vectors into a 34×n matrix, denoted as Video dataset X at time t t It can be further expressed by the following formula:
[0057]
[0058] The superscript i represents the time period i, and the value range of i is [1, l]. represents the visual features of the i-th human activity video at time t.
[0059] S22, for any time period, stack the visual features of the human activity videos in the time period in the video data sets of different time periods to obtain the long-term visual feature data of the time period. Specifically, the visual features of the human activity videos in the time period in the video data sets of different time periods are stacked according to the following formula:
[0060]
[0061] Among them, "+" represents the stacking operation of the matrix, X i Represents the long-term visual feature data of time period i, which is a matrix with a dimension of 34×3n.
[0062] S23. Construct a long-term visual feature dataset based on the long-term visual feature data of each time period.
[0063] S3, for any two long-term visual feature data, the visual features of the human activity videos of the same period are exchanged to obtain a long-term visual feature data set after the sample space data is augmented. i and X i+1 , the visual features of human activity videos of the same period in two long-term visual feature data are exchanged according to the following formula:
[0064]
[0065] Among them, X i’ and X i+1’ It is a long-term visual feature data sample obtained by data augmentation in the sample space.
[0066] S4. Based on the long-term visual feature dataset after sample space data augmentation, a twin neural network dataset is constructed; the twin neural network dataset includes several sample pairs. Two sample data are randomly selected from the augmented dataset to construct a sample pair (X j ,X k ), either one of them may be a new sample obtained by augmentation, or both may be new samples obtained by augmentation. j and X k When the labels of the samples are the same, the sample pair (X j ,X k ) is 1, otherwise it is 0. When the method provided in this embodiment is used to distinguish patients with Alzheimer's disease, the label of the sample data is 1 or 0, 1 represents patients with Alzheimer's disease, and 0 represents normal people.
[0067] S5. Use the twin neural network data set to train a neural network model to obtain a trained neural network model; the trained neural network model can accurately predict the label of the sample. Figure 2As shown in the flowchart, in this embodiment, step S5 specifically includes the following steps:
[0068] S51. Construct a neural network model based on the long short-term memory neural network LSTM.
[0069] S52. For any sample pair (X j ,X k ), input the two sample data of the sample pair into the neural network model respectively, and obtain the label pair (y j ,y k ).
[0070] S53, based on the label pairs, calculate the Euclidean distance of the label pairs, and based on the Euclidean distance of the label pairs and the comparison labels of the sample pairs, calculate the loss function value of the neural network model. In this embodiment, the loss function value of the neural network model is calculated according to the following formula:
[0071]
[0072] Among them, L is the loss function value of the neural network model; y is the comparison label of the sample pair. If the labels of the two sample data of the sample pair are the same, the comparison label y is 1, otherwise the comparison label y is 0; d is the Euclidean distance of the label pair. For a sample pair (X j ,X k ), d=||y j -y k ||2,y j For sample data X j The label, y k For sample data X k The label is, margin is the distance threshold, which is set to 1 in the experiment.
[0073] S54, according to the loss function value, back propagation optimizes the parameters of the neural network model. If the two samples in the sample pair are the same samples, the distance is reduced, and if they are different samples, the distance is increased.
[0074] If the current number of iterations does not reach the preset number of iterations, jump to step S52: for any sample pair in the twin neural network data set, input the two sample data of the sample pair into the neural network model respectively to obtain a label pair.
[0075] If the current number of iterations reaches the preset number of iterations, step S55 is executed: the neural network model after parameter optimization is used as the trained neural network model.
[0076] S6. Use the trained neural network model to map the data in the long-term visual feature data set after the sample space data is augmented to the feature space. Specifically, for any long-term visual feature data in the long-term visual feature data set after the sample space data is augmented, the long-term visual feature data is input into the trained neural network model to obtain the feature space expression of the long-term visual feature data. The specific method is as follows: Input the sample data in the long-term visual feature data set after the sample space data is augmented into the trained neural network model to obtain the expression of the sample in the feature space, which is recorded as As shown below:
[0077]
[0078] S7, weighted fusion of feature space expressions of any two long-term visual feature data with different labels to obtain a long-term visual feature data set after feature space data augmentation. In this embodiment, weighted fusion of feature space expressions of any two long-term visual feature data with different labels is performed according to the following formula:
[0079] X h,label =(1-label)X h,0 +labelX h,1
[0080] Among them, X h,label is the long-term visual feature data sample obtained by data augmentation in the feature space, X h,0 is the feature space expression of a long-term visual feature data with label 0 in the feature space, X h,1 is the feature space expression of a long-term visual feature data with label 1 in the feature space, label is X h,label The label is label∈[0.1,0.3]∪[0.7,0.9]. At this time, since the sample data has entered the feature space, the augmented data may have no corresponding data in the sample space, so the label is not 0 or 1, but the binary classification task is turned into a regression task. Correspondingly, the classification layer of the regression model needs to be modified, such as using the sigmoid function as the activation function of the output layer (or any continuous function with an output between 0 and 1), so that the augmented data can be used for the training of the regression model.
[0081] The present embodiment provides a data augmentation method based on the combination of sample space and feature space, which obtains video data sets from different periods and constructs a long-term visual feature data set based on it; for any two data, first perform data augmentation in the sample space; based on the data set after data augmentation in the sample space, construct a twin neural network data set, and train a neural network model, and then use the trained neural network model to obtain the feature space expression of the data, and finally in the feature space, perform weighted fusion of the feature space expressions of any two data with different labels to complete the data augmentation in the feature space. Based on the long-term visual feature data, without affecting the difference in data features in different periods, newly generated data is added to the existing data. Compared with other general data augmentation schemes, the data augmentation method combining sample space and feature space improves the data quality of complex data, and uses the interpolation function in the feature space to add new labels to the samples, converting the classification task into a regression task, which can increase the detection capability of the model.
[0082] Example 2
[0083] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a data augmentation method based on the combination of a sample space and a feature space in embodiment 1.
[0084] Example 3
[0085] A computer program product includes a computer program, which, when executed by a processor, implements the steps of a data augmentation method based on the combination of a sample space and a feature space in embodiment 1.
[0086] Example 4
[0087] A computer device, which may be a database, may have an internal structure as shown in Figure 3As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store pending transactions. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a data augmentation method based on the combination of sample space and feature space in Example 1 is implemented.
[0088] It should be noted that the object information (including but not limited to object device information, object personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the present invention are all information and data authorized by the object or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.
[0089] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided by the present invention may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited thereto. The processor involved in each embodiment provided by the present invention may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited thereto.
[0090] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0091] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of the present invention. At the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.
Claims
1. A data augmentation method based on the combination of sample space and feature space, It is characterized in that include: Obtain video data sets from different periods; each video data set includes several human activity videos from different periods; Based on video data sets of different periods, a long-term visual feature data set is constructed; the long-term visual feature data set includes long-term visual feature data of several different periods; the long-term visual feature data includes visual features of human activity videos of the same period of different periods; For any two long-term visual feature data, the visual features of human activity videos of the same period are exchanged to obtain a long-term visual feature dataset after sample space data augmentation; Based on the long-term visual feature dataset after sample space data augmentation, a twin neural network dataset is constructed; the twin neural network dataset includes a plurality of sample pairs; Using the twin neural network data set to train a neural network model to obtain a trained neural network model; the trained neural network model can predict the label of the sample; For any long-term visual feature data in the long-term visual feature data set after the sample space data is augmented, the long-term visual feature data is input into the trained neural network model to obtain a feature space expression of the long-term visual feature data; The feature space expressions of any two long-term visual feature data with different labels are weightedly fused to obtain a long-term visual feature dataset after feature space data augmentation; Based on video datasets from different periods, a long-term visual feature dataset is constructed, including: Extracting visual features from any human activity video in video data sets of different periods; the visual features include vector expressions of coordinate matrices of key points of the human body in the human activity video; For any time period, visual features of human activity videos in the time period in video data sets of different periods are stacked to obtain long-term visual feature data of the time period; Based on the long-term visual feature data of each time period, a long-term visual feature dataset is constructed.
2. According to claim 1, a data augmentation method based on the combination of sample space and feature space, It is characterized in that By X t-1 ,X t ,X t+1 Represents video datasets of different periods, subscript t represents period t; the video dataset of period t is represented by the following formula: The superscript i represents the time period i, and the value range of i is [1, l]. represents the visual features of the i-th human activity video at time t; For any period of time, the visual features of human activity videos of the period in the video datasets of different periods are stacked according to the following formula: Among them, X i Represents the long-term visual feature data of time period i.
3. According to claim 2, a data augmentation method based on the combination of sample space and feature space, It is characterized in that For any two long-term visual feature data, the visual features of human activity videos of the same period in the two long-term visual feature data are exchanged according to the following formula: Among them, X i’ and X i+1’ It is a long-term visual feature data sample obtained by data augmentation in the sample space.
4. The data augmentation method based on the combination of sample space and feature space according to claim 1, It is characterized in that Using the twin neural network data set to train a neural network model to obtain a trained neural network model specifically includes: Based on the long short-term memory neural network, a neural network model is constructed; For any sample pair in the twin neural network data set, two sample data of the sample pair are respectively input into the neural network model to obtain a label pair; Calculating the Euclidean distance of the label pair based on the label pair, and calculating the loss function value of the neural network model based on the Euclidean distance of the label pair and the comparison label of the sample pair; According to the loss function value, back propagation optimizes the parameters of the neural network model; If the current number of iterations does not reach the preset number of iterations, jump to step: for any sample pair in the twin neural network data set, input the two sample data of the sample pair into the neural network model respectively to obtain a label pair; If the current number of iterations reaches the preset number of iterations, the neural network model after parameter optimization is used as the trained neural network model.
5. According to claim 4, a data augmentation method based on the combination of sample space and feature space, It is characterized in that The loss function value of the neural network model is calculated according to the following formula: Among them, L is the loss function value of the neural network model; y is the comparison label of the sample pair. If the labels of the two sample data of the sample pair are the same, the comparison label y is 1, otherwise the comparison label y is 0; d is the Euclidean distance of the label pair. For a sample pair (X j ,X k ), d=||y j -y k || 2 ,y j is the label of sample data Xj, y k For sample data X k The label is , and margin is the distance threshold.
6. The data augmentation method based on the combination of sample space and feature space according to claim 1, It is characterized in that The feature space expressions of any two long-term visual feature data with different labels are weighted fused according to the following formula: X h,label =(1-label)X h,0 +labelX h,1 Among them, X h,label is a long-term visual feature data sample obtained by data augmentation in the feature space, label is X h,label The label of X, label∈[0.1,0.3]∪[0.7,0.9], h,0 and X h,1 They are the feature space expressions of any two long-term visual feature data with different labels.
7. A computer device, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of a data augmentation method based on the union of sample space and feature space as described in any one of claims 1 to 6.
8. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the steps of a data augmentation method based on the combination of sample space and feature space described in any one of claims 1 to 6 are implemented.
9. A computer program product comprising a computer program, It is characterized in that When the computer program is executed by a processor, the steps of a data augmentation method based on the combination of sample space and feature space described in any one of claims 1 to 6 are implemented.
Citation Information
Patent Citations
Image change detection method and device integrating residual network and U-Net network, storage medium and equipment
CN111915571A
Three-dimensional model sketch retrieval method based on heterogeneous twin neural network
CN113656616A
Video processing method and apparatus, and storage medium and device
WO2022171067A1
