Data augmentation method and apparatus

Abstract

The present disclosure relates to a data enhancement method and device, belonging to the technical field of data enhancement. The method comprises: obtaining a to-be-enhanced data set, performing matrix decomposition processing on the to-be-enhanced data set to obtain a multi-channel information set of the to-be-enhanced data set and a weight of the multi-channel information set, performing full permutation and combination processing on the weight to obtain an updated weight, and generating an enhanced data set of the to-be-enhanced data set according to the multi-channel information set and the updated weight. Therefore, the present disclosure obtains a multi-channel information set and a weight of the multi-channel information set by performing matrix decomposition processing on the to-be-enhanced data set, generates an enhanced data set according to the multi-channel information set and the updated weight after performing full permutation and combination processing on the weight to obtain an updated weight, realizes the maximization of the data set, reduces the probability of overfitting in the data processing process, and thus improves the accuracy of subsequent data processing.

Description

Technical Field

[0001] This disclosure relates to the field of data augmentation technology, and in particular to a data augmentation method and apparatus. Background Technology

[0002] Currently, artificial intelligence trains models by processing data from datasets. When the dataset is small, the size of the dataset is usually increased by transforming the scale or orientation of the original dataset. However, this approach is limited by the content of the original dataset itself, preventing the maximum dataset size from being achieved. Furthermore, the probability of overfitting during data processing is relatively high, resulting in lower accuracy. Summary of the Invention

[0003] This disclosure provides a data augmentation method, apparatus, terminal device, electronic device, and storage medium to at least solve the problems in related technologies where the amount of dataset cannot be maximized and the probability of overfitting during data processing is relatively high, resulting in low accuracy of data processing.

[0004] The technical solution disclosed herein is as follows:

[0005] According to a first aspect of the present disclosure, a data augmentation method is provided, comprising: acquiring a dataset to be augmented; performing matrix factorization on the dataset to be augmented to obtain a multi-channel information set of the dataset to be augmented and weights of the multi-channel information set; performing full permutation and combination processing on the weights to obtain updated weights; and generating an augmented dataset of the dataset to be augmented based on the multi-channel information set and the updated weights.

[0006] In one embodiment of this disclosure, the step of performing matrix factorization on the dataset to be enhanced to obtain a multi-channel information set and weights of the multi-channel information set includes: performing matrix factorization on the dataset to be enhanced according to a non-negative matrix factorization method to obtain the multi-channel information set and the weights.

[0007] In one embodiment of this disclosure, the step of performing matrix factorization processing on the dataset to be augmented according to the non-negative matrix factorization method to obtain the multi-channel information set and the weights includes: performing matrix factorization processing on the dataset to be augmented according to the non-negative matrix factorization method to decompose the dataset to be augmented into a convolution of an initial multi-channel information set and initial weights; and adjusting the initial multi-channel information set and the initial weights using the Lagrange method to obtain the multi-channel information set and the weights.

[0008] In one embodiment of this disclosure, adjusting the initial multi-channel information set and the initial weights using the Lagrange method to obtain the multi-channel information set and the weights includes: calculating a first convolution result of the initial multi-channel information set and the initial weights; and adjusting the initial multi-channel information set and the initial weights using the Lagrange method based on a multiplication update rule, with the minimum distance between the first convolution result and the norm of the dataset to be augmented as the adjustment objective, to obtain the multi-channel information set and the weights.

[0009] In one embodiment of this disclosure, the step of performing full permutation and combination processing on the weights to obtain updated weights includes: setting a minimum and maximum allowed value for each value in the weights; randomly selecting a value for each value between the minimum and maximum allowed values ​​to generate a combination of values ​​different from the weights, thereby obtaining the updated weights.

[0010] In one embodiment of this disclosure, the number of channels in the information set is adjustable.

[0011] In one embodiment of this disclosure, obtaining the dataset to be augmented includes: obtaining datasets under multiple single dimensions; calculating the second convolution results of the multiple datasets to obtain the dataset to be augmented.

[0012] According to a second aspect of the present disclosure, a data augmentation apparatus is provided, comprising: an acquisition module configured to acquire a dataset to be augmented; a first processing module configured to perform matrix factorization processing on the dataset to be augmented to obtain a multi-channel information set and weights of the multi-channel information set; a second processing module configured to perform full permutation and combination processing on the weights to obtain updated weights; and a generation module configured to generate an augmented dataset of the dataset to be augmented based on the multi-channel information set and the updated weights.

[0013] In one embodiment of this disclosure, the first processing module is further configured to perform matrix factorization processing on the dataset to be enhanced according to a non-negative matrix factorization method to obtain the multi-channel information set and the weights.

[0014] In one embodiment of this disclosure, the first processing module is further configured to perform: matrix factorization processing on the dataset to be augmented according to the non-negative matrix factorization method, so as to decompose the dataset to be augmented into a convolution of an initial multi-channel information set and initial weights; and adjust the initial multi-channel information set and the initial weights using the Lagrange method to obtain the multi-channel information set and the weights.

[0015] In one embodiment of this disclosure, the first processing module is further configured to perform: calculating the second convolution result of the initial multi-channel information set and the initial weights; adjusting the initial multi-channel information set and the initial weights using the Lagrange method based on the multiplication update rule, with the minimum norm distance between the second convolution result and the first convolution result as the adjustment target, to obtain the multi-channel information set.

[0016] In one embodiment of this disclosure, the second processing module is further configured to perform: setting a minimum and a maximum allowed value for each value in the weights; randomly selecting a value for each value between the minimum and the maximum allowed value to generate a combination of values ​​different from the weights, thereby obtaining the updated weights.

[0017] In one embodiment of this disclosure, the number of channels in the information set is adjustable.

[0018] In one embodiment of this disclosure, the acquisition module is further configured to perform: acquiring multiple datasets under a single dimension; calculating the second convolution result of the multiple datasets to obtain the dataset to be augmented.

[0019] According to a third aspect of the present disclosure, a terminal device is provided, including: a data enhancement device as described in the second aspect of the present disclosure.

[0020] According to a fourth aspect of the present disclosure, an electronic device is provided, comprising: a processor; a memory for storing executable instructions of the processor; wherein the processor is configured to execute the instructions to implement the method as described in the first aspect of the present disclosure.

[0021] According to a fifth aspect of the present disclosure, a computer-readable storage medium is provided that, when instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to perform the method described in the first aspect of the present disclosure.

[0022] The technical solution provided by the embodiments of this disclosure brings at least the following beneficial effects: by performing matrix factorization on the dataset to be augmented to obtain the multi-channel information set and the weights of the multi-channel information set, and then performing full permutation and combination processing on the weights to obtain the updated weights, the augmented dataset is generated based on the multi-channel information set and the updated weights, thereby maximizing the dataset, reducing the probability of overfitting during data processing, and thus improving the accuracy of subsequent data processing.

[0023] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0024] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0025] Figure 1 This is a flowchart illustrating a data augmentation method according to an exemplary embodiment.

[0026] Figure 2 This is a flowchart illustrating another data augmentation method according to an exemplary embodiment.

[0027] Figure 3 This is a flowchart illustrating a data augmentation method for obtaining a multi-channel information set and weights according to an exemplary embodiment.

[0028] Figure 4 This is a schematic diagram illustrating the W value obtained by nonnegative matrix decomposition of the spectral energy distribution according to an exemplary embodiment.

[0029] Figure 5 This is a schematic diagram illustrating an arbitrary spectral energy distribution according to an exemplary embodiment.

[0030] Figure 6 This is a block diagram illustrating a data enhancement device according to an exemplary embodiment.

[0031] Figure 7 This is a block diagram illustrating a terminal device according to an exemplary embodiment.

[0032] Figure 8 This is a block diagram illustrating an electronic device according to an exemplary embodiment. Detailed Implementation

[0033] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0034] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and devices consistent with some aspects of this disclosure as detailed in the appended claims.

[0035] In related technologies, the enhanced performance of computer hardware brought about by technological advancements, along with faster processing speeds and increased storage capacity, allows for the collection and processing of larger datasets. The more extensive and numerous the datasets, the lower the probability of overfitting in the final trained model. Currently, common data augmentation methods in the industry include, but are not limited to, one or more combinations of transformations such as shearing, rotation, reflection, flipping, scaling, translation, scale adjustment, contrast adjustment, noise perturbation, and color transformation. However, currently, the collection and labeling of each dataset requires significant manpower. If the number of samples in the dataset is too small, overfitting can occur, reducing the accuracy of data processing.

[0036] The problem of overfitting can be described as follows: For example, design a model to determine whether an object is a leaf. The dataset used to feed this model contains several leaves with sharp edges. The model designer hopes that the model can satisfy every training data point, so the model includes the sharp edges in the parameters. However, when testing the generalization performance of this model, it is found that the effect is very poor. This is because the trained model believes that leaves must have sharp edges, so it excludes all leaves without sharp edges. But in fact, leaves do not necessarily have sharp edges. This is because the model designer made the model overfit the entire training set during training, resulting in fitting noise points as well, thus causing the overfitting problem.

[0037] Based on this, this disclosure proposes a data augmentation method. By performing matrix factorization on the dataset to be augmented to obtain a multi-channel information set and its weights, and then performing full permutation and combination of the weights to obtain updated weights, an augmented dataset is generated based on the multi-channel information set and the updated weights. This maximizes the dataset size, reduces the probability of overfitting during data processing, and thus improves the accuracy of subsequent data processing.

[0038] Those skilled in the art will understand that, similar to previous data processing procedures, data is first collected based on the final results required for data processing. Data processing includes computational methods that require processing through a large number of samples, such as mathematical matrix operations, linear integral transformations, multivariate statistical analysis, and artificial intelligence (AI) processing. The final result is the indicative result obtained through the above data processing methods, such as mapping matrix indicators, indoor and outdoor indicators, social indicators, or species classification.

[0039] Figure 1 This is a flowchart illustrating a data augmentation method according to an exemplary embodiment, such as... Figure 1As shown, the data augmentation method of this disclosure embodiment may include the following steps S101-S104.

[0040] S101, Obtain the dataset to be enhanced.

[0041] It should be noted that the execution entity of the data enhancement method in this embodiment can be a terminal device or a server. Terminal devices include, but are not limited to, mobile phones, tablets, laptops, desktop computers, vehicle terminals, and smart home appliances. The data enhancement method in this embodiment can be executed by the data enhancement device, which can be configured in any terminal device to execute the data enhancement method. The data enhancement device can also be configured in a server; there are no limitations on this.

[0042] It should be noted that the data augmentation method of this disclosure is applicable to various types of datasets with features based on image processing, such as datasets with color, texture, and contour. This disclosure uses a spectral energy dataset as an example for description and does not constitute a limitation on the type of dataset.

[0043] In the embodiments of this disclosure, the dataset is the original dataset to be enhanced. The dataset to be enhanced is acquired for subsequent processing.

[0044] S102, perform matrix factorization on the dataset to be augmented to obtain the multi-channel information set and the weights of the multi-channel information set of the dataset to be augmented.

[0045] In the embodiments of this disclosure, the multi-channel information set is a collection of all important information from multiple channels in the process of convolving the new dataset under all single dimensions to obtain the dataset to be enhanced. The weights are the weights required to perfectly reproduce the dataset to be enhanced using the multi-channel information set. Matrix factorization is performed on the dataset to be enhanced obtained in step S101 to obtain the corresponding multi-channel information set and its weights. It should be noted that the number of channels in the multi-channel information set is adjustable. Specifically, the number of channels can be appropriately adjusted according to the final actual requirements. For example, taking a spectral energy dataset as an example, the number of channels can be selected based on the magnitude of the dominant wavelength and half-width in the spectrum. Typically, data with non-repeating dominant wavelengths and small half-widths are selected to adjust the number of channels. The smaller the half-width, the higher the importance. By selecting data with smaller half-widths, some less important data within the half-width range can be eliminated.

[0046] S103, perform full permutation and combination processing on the weights to obtain the updated weights.

[0047] In the embodiments of this disclosure, the weights of the multi-channel information set obtained in step S102 are subjected to a full permutation combination process to update the weights, thereby obtaining the updated weights. Here, the full permutation refers to the full permutation method in the recursive algorithm. By combining the weights using the full permutation method, all other weights besides the original weight can be obtained, i.e., the updated weights. It should be noted that the weights are those required to perfectly reproduce the dataset to be augmented using the multi-channel information set. At this point, the weights represent only a specific solution to the dataset to be augmented. In this case, if different weights are applied to the multi-channel information set, data that is similar to but different from the dataset to be augmented can be obtained. Therefore, the weights need to be processed to obtain the updated weights.

[0048] S104, Based on the multi-channel information set and the updated weights, generate an augmented dataset of the dataset to be augmented.

[0049] In the embodiments of this disclosure, the augmented dataset is a dataset obtained based on the dataset to be augmented, according to the multi-channel information set and the updated weights. It should be noted that by applying different weights to the multi-channel information set, some data that is similar to but different from the dataset to be augmented can be obtained, i.e., an augmented dataset of the dataset to be augmented. In the augmented dataset, some additional data that is also quite important in addition to the content of the dataset to be augmented can definitely be found.

[0050] The data augmentation method provided in this disclosure involves obtaining a dataset to be augmented, performing matrix factorization on the dataset to obtain a multi-channel information set and weights for the multi-channel information set, performing full permutation and combination processing on the weights to obtain updated weights, and generating an augmented dataset based on the multi-channel information set and the updated weights. Therefore, this disclosure maximizes the dataset size by performing matrix factorization on the dataset to be augmented to obtain a multi-channel information set and weights, and then generating an augmented dataset based on the updated weights after performing full permutation and combination processing on the weights. This reduces the probability of overfitting during data processing and improves the accuracy of subsequent data processing.

[0051] Figure 2 This is a flowchart illustrating another data augmentation method according to an exemplary embodiment, such as... Figure 2 As shown, the data augmentation method of this disclosure embodiment may include the following steps.

[0052] S201, obtain datasets under multiple single dimensions.

[0053] In the embodiments of this disclosure, datasets of the dataset to be augmented in multiple single dimensions can be acquired for subsequent processing. It should be noted that the dataset to be augmented is a multi-dimensional dataset. In the data collection process for the dataset to be augmented, the datasets in multiple single dimensions can be considered as a combination of high-dimensionality (hd), low-dimensionality (ld), and noise (w). This is obtained through principal component analysis (PCA), a statistical method that recombines the original variables of the data into a set of mutually independent linear combinations through orthogonal transformation.

[0054] In one embodiment, taking a spectral energy dataset as an example, the spectral energy dataset can be recombined into a set of mutually independent linear combination variables representing high-dimensional, low-dimensional, and noise components. Multiple new datasets can be generated based on multiple single dimensions of the spectral energy dataset, with each single dimension corresponding to a new dataset.

[0055] For example, the spectral energy dataset to be enhanced can include single dimensions such as brightness and chromaticity. That is, the datasets under single dimensions such as brightness and chromaticity can be represented as a combination of high-dimensional, low-dimensional and noise. Therefore, the expression of the spectral energy dataset to be enhanced is shown in the following formula (1), where i is the i-th single dimension of the spectral energy dataset, i(x) is the number of datasets under the i-th single dimension, x is the number of single dimensions in the spectral energy dataset, and di(x) is the dataset under the i-th single dimension of the spectral energy dataset to be enhanced.

[0056] di(x) = hd + ld + w, where i = 1, ..., n (1)

[0057] S202, calculate the second convolution result of multiple datasets to obtain the dataset to be augmented.

[0058] In the embodiments of this disclosure, convolution calculations are performed on the datasets under multiple single dimensions obtained in step S201 to obtain the second convolution results of the corresponding multiple datasets, which are the datasets to be enhanced.

[0059] Convolving the datasets with multiple single dimensions in equation (1) yields the dataset to be enhanced, which can be represented by equation (2), where n represents the number of single dimensions in the dataset to be enhanced, and d n (x) represents the dataset under the nth single dimension, that is, the dataset has the dimensional features of the spectral energy dataset to be enhanced, and dt(x) represents the dataset to be enhanced obtained after convolution calculation of the dataset under the n single dimensions.

[0060] dt(x) = d1(x) × d2(x) × … d n (x) (2)

[0061] The step S102 in the above embodiment, "perform matrix factorization on the dataset to be enhanced to obtain the multi-channel information set and the weights of the multi-channel information set of the dataset to be enhanced", may specifically include the following step S203.

[0062] S203, perform matrix factorization on the dataset to be augmented using the nonnegative matrix factorization method to obtain a multi-channel information set and weights.

[0063] In the embodiments of this disclosure, the dataset to be augmented is matrix decomposed according to the non-negative matrix factorization (NMF) method, wherein the non-negative matrix factorization method is a method of decomposing a large non-negative matrix into the product of two small non-negative matrices.

[0064] The step S103 in the above embodiment, "perform full permutation and combination processing on the weights to obtain the updated weights", may specifically include the following steps S204-S205.

[0065] S204, set the minimum and maximum allowed values ​​for each value in the weight.

[0066] In the embodiments of this disclosure, each value in the weights of the multi-channel information set obtained in step S203 is set to obtain the minimum and maximum allowed values ​​for each value. For example, the weights of the multi-channel information set include 5 values, and it is stipulated that the minimum allowed value for each of the 5 values ​​is 1 and the maximum allowed value is 10.

[0067] S205: For each value, randomly select a value between the minimum and maximum allowed values ​​to generate a combination of values ​​different from the weights, and obtain the updated weights.

[0068] In the embodiments of this disclosure, the generated different numerical combinations correspond one-to-one with the updated weights. For each value in the weights, a value is randomly selected between the minimum and maximum allowed values ​​set in step S204. By arranging these values ​​in a certain order, all numerical combinations different from the weights can be generated, resulting in the updated weights. For example, as mentioned above, the weights of the multi-channel information set include five values ​​(2, 4, 5, 7, 3), which are selected between 1 and 10. One possible numerical combination is (3, 2, 5, 9, 6), which is the updated weight.

[0069] S206, Based on the multi-channel information set and the updated weights, generate an augmented dataset for the dataset to be augmented.

[0070] Specifically, step S206 in this embodiment is the same as step S104 in the above embodiment, and will not be repeated here.

[0071] The data augmentation method provided in this disclosure acquires multiple datasets under a single dimension, calculates the second convolution results of the multiple datasets to obtain a dataset to be augmented, performs matrix factorization on the dataset to be augmented using a non-negative matrix factorization method to obtain a multi-channel information set and weights, sets a minimum and maximum allowable value for each value in the weights, and randomly selects a value between the minimum and maximum allowable values ​​for each value to generate a combination of values ​​different from the weights, obtaining updated weights. Based on the multi-channel information set and the updated weights, an augmented dataset of the dataset to be augmented is generated. Therefore, this disclosure maximizes the dataset size by performing matrix factorization on the dataset to be augmented to obtain a multi-channel information set and weights for the multi-channel information set, and then generating an augmented dataset based on the multi-channel information set and the updated weights after performing full permutation and combination processing on the weights. This reduces the probability of overfitting during data processing, thereby improving the accuracy of subsequent data processing. Simultaneously, by randomly selecting a value between the minimum and maximum allowable values ​​for each weight, the richness of the augmented dataset is improved, further enhancing the accuracy of subsequent data processing.

[0072] Furthermore, step S203 in the above embodiment, "performing matrix factorization on the dataset to be augmented using the nonnegative matrix factorization method to obtain a multi-channel information set and weights," may include the following steps:

[0073] S301, perform matrix factorization on the dataset to be augmented according to the non-negative matrix factorization method, so as to decompose the dataset to be augmented into an initial multi-channel information set and a convolution of initial weights.

[0074] In the embodiments of this disclosure, the dataset to be augmented is subjected to matrix factorization processing according to the non-negative matrix factorization method, so as to decompose the dataset to be augmented into an initial multi-channel information set and a convolution of initial weights.

[0075] For example, the spectral energy dataset to be enhanced obtained in the above example is subjected to matrix decomposition using the non-negative matrix factorization method. For the dataset to be enhanced, the non-negative matrix factorization method can find a relationship between the dataset to be enhanced and the convolution of the initial multi-channel information set and the initial weights, and make it satisfy the relationship in the following equation (3), where W is the initial multi-channel information set, H is the initial weight of the initial multi-channel information set, and dt(x) represents the dataset to be enhanced.

[0076]

[0077] S302, the initial multi-channel information set and initial weights are adjusted using the Lagrange method to obtain the multi-channel information set and weights.

[0078] In the embodiments of this disclosure, the initial multi-channel information set and initial weights are adjusted using the Lagrange method to obtain the corresponding multi-channel information set and weights. It should be noted that this disclosure does not impose excessive limitations on the specific adjustment method, and it can be set according to the actual situation.

[0079] In some implementations, the first convolution result of the initial multi-channel information set and initial weights is calculated. Based on the multiplicative update rules, the initial multi-channel information set and initial weights are adjusted using the Lagrange method with the minimum norm distance between the first convolution result and the dataset to be augmented as the adjustment objective, to obtain the multi-channel information set and weights.

[0080] For example, as described in the example above, under the condition that the norm distance between the first convolution result and the dataset to be enhanced is minimized, i.e., ‖dt(x)-WH‖ is minimized, the initial multi-channel information set W and the initial weight H in equation (3) are adjusted using the Lagrange method based on the multiplication update rule. Then the updated multi-channel information set and weights can be represented by the following equations (4) and (5), respectively. When the multi-channel information set and weights finally become a set of stable values ​​with very small changes, it means that the final result after the adjustment of these values ​​is the final multi-channel information set and weights. In equation (4), H aμ For the updated weights, α and μ are the number of rows and columns of the initial weight H matrix, and α*μ is the size of the initial weight H. In equation (5), W ia For the updated multi-channel information set, i and α are the number of rows and columns of the initial multi-channel information set W as a matrix, and i*α is the size of the initial multi-channel information set W. T H is the transpose of the initial multi-channel information set W after row and column interchange. T This is the transpose of the initial weights H after row and column swapping; other meanings are the same as in the above embodiment.

[0081]

[0082]

[0083] Therefore, by adjusting the initial multi-channel information set and initial weights obtained from the dataset to be augmented through non-negative matrix factorization using the Lagrange method, the accuracy of the weights in the multi-channel information set is increased, thereby further improving the accuracy of subsequent data processing.

[0084] For example, Figure 4This is a schematic diagram of the multi-channel information set W values ​​obtained by nonnegative matrix factorization of the spectral energy distribution, as shown below. Figure 4 As shown, the horizontal axis represents wavelength, with the unit being nm. It can be seen that... Figure 4 If the horizontal axis of the graph ranges from 400nm to 700nm, then the light wave is visible light. The vertical axis represents the radiant power energy per unit wavelength interval, with the unit being W. Those skilled in the art will understand that spectral reflectance is a physical quantity describing the physical properties of an object's surface, the spectral power distribution (SPD) curve is the distribution of light energy at different frequencies, and the spectral energy distribution is the spectral radiant output of an object at various wavelengths under a given light source.

[0085] Where di(x) represents the spectral reflectance and spectral power distribution curves, the high-dimensional part of dt(x) refers to the higher part of the spectral reflectance and spectral power distribution curve data, the low-dimensional part refers to the lower part of the spectral reflectance and spectral power distribution curve data, and the noise refers to the jagged parts at both ends of the spectral reflectance and spectral power distribution curves or the peaks. dt(x) represents the spectral energy distribution. After convolving the spectral reflectance and spectral power distribution curves, i.e., di(x), the spectral energy distribution of the object, i.e., dt(x), is obtained. Then, the non-negative matrix factorization method is used to process dt(x) to obtain the multi-channel information set W value and the weight H value of the multi-channel information set. These multi-channel information set W values ​​can be regarded as the contents of both the spectral reflectance and spectral power distribution, i.e., di(x), and they present the most important dominant wavelength distribution in the spectral energy distribution dt(x). Then, the full permutation method of the recursive algorithm is used to calculate the remaining spectral energy distribution dt(x) through the H value, which is used as the current result. Therefore, the original dt(x) consisted of 1 million sets of spectral energy distribution data. After this processing, the resulting dt(x) can be increased by 9 times to 9 million sets of data. The new spectral energy distribution dt(x) arbitrarily generated using this method is as follows: Figure 5 As shown, Figure 5 The x and y coordinates have the same meaning as Figure 4 This will not be elaborated upon here.

[0086] Figure 6 This is a block diagram illustrating a data enhancement device according to an exemplary embodiment. Figure 6 As shown, the data enhancement device 600 of this embodiment includes: an acquisition module 601, a first processing module 602, a second processing module 603, and a generation module 604.

[0087] Module 601 is configured to retrieve the dataset to be augmented.

[0088] The first processing module 602 is configured to perform matrix factorization processing on the dataset to be augmented, and obtain the multi-channel information set and the weights of the multi-channel information set of the dataset to be augmented.

[0089] The second processing module 603 is configured to perform a full permutation and combination process on the weights to obtain the updated weights.

[0090] The generation module 604 is configured to perform the generation of an augmented dataset of the dataset to be augmented based on the multi-channel information set and the updated weights.

[0091] In one embodiment of this disclosure, the first processing module 602 is further configured to perform matrix factorization processing on the dataset to be augmented according to the non-negative matrix factorization method to obtain a multi-channel information set and weights.

[0092] In one embodiment of this disclosure, the first processing module 602 is further configured to perform: matrix factorization processing on the dataset to be augmented according to the non-negative matrix factorization method, so as to decompose the dataset to be augmented into a convolution of an initial multi-channel information set and initial weights; and adjust the initial multi-channel information set and initial weights using the Lagrange method to obtain a multi-channel information set and weights.

[0093] In one embodiment of this disclosure, the first processing module 602 is further configured to perform: calculating the second convolution result of the initial multi-channel information set and the initial weights; and adjusting the initial multi-channel information set and the initial weights using the Lagrange method based on the multiplication update rule, with the minimum norm distance between the second convolution result and the first convolution result as the adjustment objective, to obtain the multi-channel information set.

[0094] In one embodiment of this disclosure, the second processing module 603 is further configured to perform: setting a minimum and maximum allowable value for each value in the weights; randomly selecting a value for each value between the minimum and maximum allowable values ​​to generate a combination of values ​​different from the weights, thereby obtaining the updated weights.

[0095] In one embodiment of this disclosure, the number of channels in the information set is adjustable.

[0096] In one embodiment of this disclosure, the acquisition module 601 is further configured to perform: acquiring multiple datasets under a single dimension; calculating the second convolution result of the multiple datasets to obtain the dataset to be augmented.

[0097] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments relating to the apparatus, and will not be elaborated upon here.

[0098] The data augmentation apparatus provided in the embodiments of this disclosure acquires a dataset to be augmented, performs matrix factorization on the dataset to be augmented to obtain a multi-channel information set and weights for the multi-channel information set, performs full permutation and combination processing on the weights to obtain updated weights, and generates an augmented dataset based on the multi-channel information set and the updated weights. Therefore, this disclosure maximizes the dataset size by performing matrix factorization on the dataset to be augmented to obtain the multi-channel information set and weights, and then generating an augmented dataset based on the updated weights after performing full permutation and combination processing on the weights. This reduces the probability of overfitting during data processing and improves the accuracy of subsequent data processing.

[0099] Figure 7 This is a block diagram illustrating a terminal device 700 according to an exemplary embodiment.

[0100] like Figure 7 As shown, the terminal device 700 of this embodiment includes: the data enhancement device 600 as described in the above embodiment.

[0101] The terminal device provided in the embodiments of this disclosure can execute the data augmentation device as described above, acquire the dataset to be augmented, perform matrix factorization on the dataset to be augmented to obtain the multi-channel information set and the weights of the multi-channel information set, perform full permutation and combination processing on the weights to obtain updated weights, and generate an augmented dataset based on the multi-channel information set and the updated weights. Therefore, this disclosure maximizes the dataset size by performing matrix factorization on the dataset to be augmented to obtain the multi-channel information set and the weights of the multi-channel information set, and then generating an augmented dataset based on the multi-channel information set and the updated weights after performing full permutation and combination processing on the weights. This reduces the probability of overfitting during data processing and improves the accuracy of subsequent data processing.

[0102] Figure 8 This is a block diagram illustrating an electronic device 800 according to an exemplary embodiment.

[0103] like Figure 8 As shown, the above-mentioned electronic device 800 includes:

[0104] The memory 801 and the processor 802 are connected by a bus 803, which connects the different components (including the memory 801 and the processor 802). The memory 801 stores a computer program, and when the processor 802 executes the program, it implements the data enhancement method described in the embodiments of this disclosure.

[0105] Bus 803 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0106] Electronic device 800 typically includes a variety of electronic device readable media. These media can be any available media that can be accessed by electronic device 800, including volatile and non-volatile media, removable and non-removable media.

[0107] Memory 801 may also include computer system readable media in the form of volatile memory, such as random access memory (RAM) 804 and / or cache memory 805. Electronic device 800 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 806 may be used to read and write non-removable, non-volatile magnetic media (… Figure 8 Not shown; usually referred to as a "hard drive"). Although Figure 8 Not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk") and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 803 via one or more data media interfaces. Memory 801 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of this disclosure.

[0108] A program / utility 808 having a set (at least one) of program modules 807 may be stored, for example, in memory 801. Such program modules 807 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 807 typically perform the functions and / or means described in the embodiments of this disclosure.

[0109] Electronic device 800 can also communicate with one or more external devices 809 (e.g., keyboard, pointing device, display 810, etc.), and with one or more devices that enable a user to interact with the electronic device 800, and / or with any device that enables the electronic device 800 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed through input / output (I / O) interface 812. Furthermore, electronic device 800 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 813. Figure 8 As shown, network adapter 813 communicates with other modules of electronic device 800 via bus 803. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with electronic device 800, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0110] The processor 802 performs various functional applications and data processing by running programs stored in the memory 801.

[0111] It should be noted that the implementation process and technical principles of the electronic device in this embodiment are explained in the foregoing description of the data enhancement device in the embodiments of this disclosure, and will not be repeated here.

[0112] The electronic device provided in this disclosure can execute the data augmentation method described above, obtain a dataset to be augmented, perform matrix factorization on the dataset to be augmented to obtain a multi-channel information set and weights of the multi-channel information set, perform full permutation and combination processing on the weights to obtain updated weights, and generate an augmented dataset based on the multi-channel information set and the updated weights. Therefore, this disclosure maximizes the dataset size by performing matrix factorization on the dataset to be augmented to obtain a multi-channel information set and weights, and then generating an augmented dataset based on the updated weights after performing full permutation and combination processing on the weights. This reduces the probability of overfitting during data processing and improves the accuracy of subsequent data processing.

[0113] To implement the above embodiments, this disclosure also proposes a computer-readable storage medium.

[0114] When the instructions in the computer-readable storage medium are executed by the processor of an electronic device, the electronic device is able to perform the data enhancement method as described above. Optionally, the computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc.

[0115] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0116] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A data augmentation method, characterized in that, include: Obtain the dataset to be augmented; The dataset to be augmented is subjected to matrix factorization to obtain a multi-channel information set and weights of the multi-channel information set. The multi-channel information set is the set of all important information of multiple channels in the process of convolving the new dataset under all single dimensions to obtain the dataset to be augmented. The weights are the weights required to perfectly reproduce the dataset to be augmented using the multi-channel information set. The weights are subjected to a full permutation and combination process to obtain the updated weights; Based on the multi-channel information set and the updated weights, an enhanced dataset is generated for the dataset to be enhanced. The step of performing a full permutation and combination process on the weights to obtain the updated weights includes: Set the minimum and maximum allowed values ​​for each value in the weights; For each value, a value is randomly selected between the minimum and maximum allowed values ​​to generate a combination of values ​​different from the weights, thus obtaining the updated weights.

2. The data augmentation method according to claim 1, characterized in that, The step of performing matrix factorization on the dataset to be augmented to obtain the multi-channel information set of the dataset to be augmented and the weights of the multi-channel information set includes: The dataset to be augmented is subjected to matrix factorization using a nonnegative matrix factorization method to obtain the multi-channel information set and the weights.

3. The data augmentation method according to claim 2, characterized in that, The step of performing matrix factorization on the dataset to be enhanced using a nonnegative matrix factorization method to obtain the multi-channel information set and the weights includes: The dataset to be augmented is subjected to matrix factorization processing according to the nonnegative matrix factorization method, so as to decompose the dataset to be augmented into an initial multi-channel information set and a convolution of initial weights; The initial multi-channel information set and the initial weights are adjusted using the Lagrange method to obtain the multi-channel information set and the weights.

4. The data augmentation method according to claim 3, characterized in that, The step of adjusting the initial multi-channel information set and the initial weights using the Lagrange method to obtain the multi-channel information set and the weights includes: Calculate the first convolution result of the initial multi-channel information set and the initial weights; Based on the multiplication-based update rule, with the minimum norm distance between the first convolution result and the dataset to be augmented as the adjustment objective, the Lagrange method is used to adjust the initial multi-channel information set and the initial weights to obtain the multi-channel information set and the weights.

5. The data augmentation method according to claim 1, characterized in that, The number of channels in the multi-channel information set is adjustable.

6. The data augmentation method according to claim 1, characterized in that, The process of obtaining the dataset to be augmented includes: Obtain datasets from multiple single dimensions; The second convolution result of multiple datasets is calculated to obtain the dataset to be augmented.

7. A data augmentation device, characterized in that, include: The acquisition module is configured to acquire the dataset to be augmented; The first processing module is configured to perform matrix factorization on the dataset to be augmented to obtain a multi-channel information set and weights of the multi-channel information set. The multi-channel information set is a set of all important information of multiple channels in the process of convolving all new datasets under all single dimensions to obtain the dataset to be augmented. The weights are the weights required to perfectly reproduce the dataset to be augmented using the multi-channel information set. The second processing module is configured to perform a full permutation and combination process on the weights to obtain the updated weights; The generation module is configured to generate an augmented dataset of the dataset to be augmented based on the multi-channel information set and the updated weights. The second processing module is further configured to execute: Set the minimum and maximum allowed values ​​for each value in the weights; For each value, a value is randomly selected between the minimum and maximum allowed values ​​to generate a combination of values ​​different from the weights, thus obtaining the updated weights.

8. The data enhancement apparatus according to claim 7, characterized in that, The first processing module is further configured to execute: The dataset to be augmented is subjected to matrix factorization using a nonnegative matrix factorization method to obtain the multi-channel information set and the weights.

9. The data enhancement apparatus according to claim 8, characterized in that, The first processing module is further configured to execute: The dataset to be augmented is subjected to matrix factorization processing according to the nonnegative matrix factorization method, so as to decompose the dataset to be augmented into an initial multi-channel information set and a convolution of initial weights; The initial multi-channel information set and the initial weights are adjusted using the Lagrange method to obtain the multi-channel information set and the weights.

10. The data enhancement apparatus according to claim 9, characterized in that, The first processing module is further configured to execute: Calculate the first convolution result of the initial multi-channel information set and the initial weights; Based on the multiplication-based update rule, with the minimum norm distance between the first convolution result and the dataset to be augmented as the adjustment objective, the initial multi-channel information set and the initial weights are adjusted using the Lagrange method to obtain the multi-channel information set and the weights.

11. The data enhancement apparatus according to claim 7, characterized in that, The number of channels in the multi-channel information set is adjustable.

12. The data enhancement apparatus according to claim 7, characterized in that, The acquisition module is further configured to execute: Obtain datasets from multiple single dimensions; The second convolution result of multiple datasets is calculated to obtain the dataset to be augmented.

13. A terminal device, characterized in that, include: The data enhancement apparatus as described in any one of claims 7-12.

14. An electronic device, characterized in that, include: processor; Memory for storing the executable instructions of the processor; The processor is configured to execute the instructions to implement the method as described in any one of claims 1-6.

15. A computer-readable storage medium, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is able to perform the method as described in any one of claims 1-6.

Images