Sample augmentation method and system for data quality category imbalance

Abstract

The invention discloses a sample augmentation method and system for data quality category imbalance, and the method comprises the steps: obtaining original marine physical data, and carrying out the preprocessing of the original marine physical data, and obtaining marine feature data; quality classification processing is carried out on the ocean feature data to obtain ocean classification data, and the ocean classification data comprise ocean samples and sample category labels; according to the sample category labels, forward noise diffusion is carried out on the ocean samples, a noise sample sequence is obtained, the noise sample sequence comprises a plurality of noise samples of different noise adding time steps, and the noise samples are ocean samples in a Gaussian noise state; and inputting the noise sample sequence into a sample denoising model for reverse denoising to obtain an augmented sample corresponding to the noise sample sequence. According to the method, the sample augmentation effect of the marine physical data can be effectively improved, and the quality of the generated data sample is improved. The invention relates to the technical field of data processing.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a sample augmentation method and system for imbalanced data quality categories. Background Art

[0002] In recent years, with the increasing demand for exploration and development of seabed resources (such as oil, gas, methane hydrate, etc.), marine physical data (such as temperature, pressure, salinity, marine sediments, current data, etc.) have become increasingly important in practical applications. However, due to factors such as the complexity of the marine environment and the limitations of detection equipment, marine physical data are usually derived measurements obtained by indirect means, which makes the data quality categories of marine physical data extremely unbalanced. Specifically, there are fewer high-quality marine physical data and more medium and low-quality marine physical data, which makes subsequent models tend to learn more about categories with more data when processing category-skewed data.

[0003] At present, the traditional methods of sample augmentation for ocean physical data generally include undersampling, oversampling and cost-sensitive learning. These methods can alleviate the problem of data quality category imbalance of ocean physical data to a certain extent, but these methods are relatively limited in processing complex ocean physical data, and the quality of the generated data samples is poor.

[0004] Therefore, the problems existing in the prior art still need to be solved and optimized. Summary of the invention

[0005] In order to solve at least one of the above-mentioned technical problems, the present application provides a sample augmentation method and system for unbalanced data quality categories, wherein the method can effectively improve the sample augmentation effect of marine physical data and improve the quality of generated data samples.

[0006] According to a first aspect of the present application, a sample augmentation method for imbalanced data quality categories is provided, the method comprising: Acquiring original ocean physical data, and preprocessing the original ocean physical data to obtain ocean characteristic data; Performing quality classification processing on the ocean characteristic data to obtain ocean classification data, wherein the ocean classification data includes ocean samples and sample category labels, and the sample category labels are used to characterize the data quality category of the corresponding ocean samples; According to the sample category label, forward noise diffusion is performed on the ocean sample to obtain a noise sample sequence, wherein the noise sample sequence includes noise samples at a plurality of different noise adding time steps, and the noise sample is an ocean sample in a Gaussian noise state; The noise sample sequence is input into a sample denoising model for reverse denoising to obtain augmented samples corresponding to the noise sample sequence.

[0007] Furthermore, in an embodiment of the present application, the preprocessing of the original ocean physical data to obtain ocean characteristic data includes: Performing data cleaning on the original ocean physical data to obtain first intermediate data; Performing quartile anomaly detection on the first intermediate data to obtain second intermediate data; The second intermediate data is normalized to obtain the ocean characteristic data.

[0008] Furthermore, in an embodiment of the present application, the quality classification processing of the ocean characteristic data to obtain ocean classification data includes: Obtaining the data standard deviation and data mean of the ocean characteristic data, and a preset ocean data quality classification table; Performing data quality analysis on the ocean characteristic data according to the data standard deviation and the data mean to obtain a data quality analysis result, wherein the data quality analysis result is used to characterize the degree of dispersion of the ocean characteristic data; According to the data quality analysis result, the ocean data quality classification table is matched with quality categories to obtain the ocean samples and the sample category labels corresponding to the ocean samples.

[0009] Further, in the embodiment of the present application, the forward noise diffusion is performed on the ocean sample according to the sample category label to obtain a noise sample sequence, including: Get the preset sample threshold; According to the sample category labels, all the ocean samples are screened to obtain a plurality of intermediate sample sets, where the intermediate sample sets are used to characterize a collection of ocean samples with the same data quality category; According to the sample threshold, all the intermediate sample sets are screened for sample quantity to obtain a plurality of target sample matrices, wherein the target sample matrices are used to characterize a matrix set of target samples, wherein the target samples are ocean samples in the intermediate sample set whose sample quantity is less than the sample threshold, and each target sample matrix corresponds to an intermediate sample set whose sample quantity is less than the sample threshold; Forward noise is added to the target sample matrix to obtain the noise sample sequence.

[0010] Furthermore, in an embodiment of the present application, the step of adding forward noise to the target sample matrix to obtain the noise sample sequence includes: Get the preset noise adding time step threshold and Gaussian noise; Taking the current noise adding time step as the first time step, obtaining the cumulative product parameter corresponding to the first time step; Performing sample noise addition on the target sample matrix according to the cumulative product parameter and the Gaussian noise to obtain a noisy sample matrix; Comparing the first time step with the noise adding time step threshold to obtain a first comparison result; If the first comparison result is that the first time step is less than the noise adding time step threshold, the noise adding sample matrix is ​​determined as the noise sample corresponding to the first time step, and the step of taking the current noise adding time step as the first time step and obtaining the cumulative product parameter corresponding to the first time step is returned to be executed; or, if the first comparison result is that the first time step is equal to the noise adding time step threshold, the noise adding sample matrix is ​​determined as the noise sample corresponding to the first time step, and the noise sample sequence is generated according to the noise samples of all different noise adding time steps.

[0011] Furthermore, in the embodiment of the present application, the method further includes: Get the preset sample augmentation threshold; Get the current augmented sample and sample augmented sequence; Comparing the current sample augmentation sequence with the sample augmentation threshold to obtain a second comparison result; If the second comparison result is that the total number of augmented samples in the sample augmented sequence is less than the sample augmented threshold, the sample augmented sequence is updated according to the augmented samples, and then the step of obtaining the current augmented samples and the sample augmented sequence is returned to be executed; or, if the second comparison result is that the total number of augmented samples in the sample augmented sequence is equal to the sample augmented threshold, a sample augmented set is obtained according to the sample augmented sequence.

[0012] Further, in the embodiment of the present application, the step of inputting the noise sample sequence into a sample denoising model for reverse denoising to obtain augmented samples corresponding to the noise sample sequence includes: Get the preset denoising time step threshold; Taking the current denoising time step as the second time step, obtaining a first denoising sample corresponding to the second time step; Inputting the first denoising sample into the sample denoising model to perform sample denoising prediction to obtain predicted noise; According to the first target noise sample corresponding to the first denoised sample, a posterior probability distribution calculation is performed on the second target noise sample to obtain a posterior and a posterior variance corresponding to the predicted noise, wherein the second target noise sample is a noise sample of a noise adding time step before the first target noise sample; Performing sample denoising on the first denoised sample according to the posterior mean and the predicted noise to obtain a second denoised sample; Comparing the second time step with the denoising time step threshold to obtain a third comparison result; If the third comparison result is that the second time step is greater than the denoising time step threshold, the first denoised sample is updated according to the second denoised sample, and then the step of returning to execute the step of taking the current denoised time step as the second time step and acquiring the first denoised sample corresponding to the second time step; or, if the third comparison result is that the second time step is equal to the denoising time step threshold, the second denoised sample is determined as the augmented sample.

[0013] According to a second aspect of the present application, a sample augmentation system for imbalanced data quality categories is provided, the system comprising: A first processing unit is used to obtain original ocean physical data and pre-process the original ocean physical data to obtain ocean characteristic data; A second processing unit is used to perform quality classification processing on the ocean characteristic data to obtain ocean classification data, wherein the ocean classification data includes ocean samples and sample category labels, and the sample category labels are used to characterize the data quality category of the corresponding ocean samples; a third processing unit, configured to perform forward noise diffusion on the ocean sample according to the sample category label to obtain a noise sample sequence, wherein the noise sample sequence includes noise samples at a plurality of different noise adding time steps, and the noise sample is an ocean sample in a Gaussian noise state; The fourth processing unit is used to input the noise sample sequence into a sample denoising model to perform reverse denoising to obtain augmented samples corresponding to the noise sample sequence.

[0014] According to a third aspect of the present application, a computer device is provided, comprising: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method as described above.

[0015] According to a fourth aspect of the present application, a computer-readable storage medium is provided, in which a program executable by a processor is stored. When the program executable by the processor is executed by the processor, it is used to implement the method described in the above aspects.

[0016] The beneficial effects of the technical solution provided by the embodiment of the present application are: The present application provides a sample augmentation method and system for data quality category imbalance, wherein the method obtains original ocean physical data, and preprocesses the original ocean physical data to obtain ocean characteristic data; performs quality classification processing on the ocean characteristic data to obtain ocean classification data, the ocean classification data includes ocean samples and sample category labels, and the sample category labels are used to characterize the data quality category of the corresponding ocean samples; according to the sample category labels, forward noise diffusion is performed on the ocean samples to obtain a noise sample sequence, the noise sample sequence includes noise samples of several different noise addition time steps, and the noise samples are ocean samples in a Gaussian noise state; the noise sample sequence is input into a sample denoising model for reverse denoising to obtain augmented samples corresponding to the noise sample sequence. By performing forward noise diffusion and reverse denoising on the ocean samples, the method can make the distribution characteristics of the generated augmented samples in each feature dimension consistent with the distribution characteristics of the ocean samples, which can effectively improve the sample augmentation effect of the ocean physical data and improve the quality of the generated data samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flow chart of a sample augmentation method for imbalanced data quality categories provided in an embodiment of the present application; Figure 2 A detailed flow chart of step S101 provided in an embodiment of the present application; Figure 3 A detailed flow chart of step S102 provided in an embodiment of the present application; Figure 4 A detailed flow chart of step S103 provided in an embodiment of the present application; Figure 5 A detailed flow chart of step C4 provided in an embodiment of the present application; Figure 6 A detailed flow chart of step S104 provided in an embodiment of the present application; Figure 7 A network framework diagram of a U-Net neural network model provided in an embodiment of the present application; Figure 8 A schematic diagram of an optional flow chart of a sample augmentation method for imbalanced data quality categories provided in an embodiment of the present application; Fig. 9 A schematic diagram of a framework of a sample augmentation system for imbalanced data quality categories provided in an embodiment of the present application; Fig.10 A structural block diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] The present application is further described below in conjunction with the accompanying drawings and specific embodiments. The described embodiments should not be regarded as limiting the present application, and all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0019] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0021] The following is an introduction to several terms involved in this application: Heat flow: Heat flow can also be called geothermal heat flow or heat flow density. It refers to the heat transmitted from the earth's interior to the surface per unit area and per unit time, and then dissipated into space. It is a comprehensive quantitative representation of the thermal state of the deep crust or lithosphere on the surface.

[0022] Ocean heat flow: Ocean heat flow refers to a heat flow that radiates from the earth's interior to the seafloor surface. In practical applications, through the measurement and analysis of ocean heat flow, we can infer the thermal state of the earth's interior, reveal the thermodynamic characteristics of the crust and mantle, and provide basic data for research on global climate change, seabed resource development, geological disaster prediction (such as earthquakes and volcanic activities), etc.

[0023] At present, the traditional methods of sample augmentation for marine physical data are generally divided into three categories: undersampling, oversampling, and cost-sensitive learning. Among them, the undersampling method reduces the majority class samples to balance the number of minority class samples. Although this method can solve the problem of class imbalance, it may cause information loss of majority class data, thereby affecting the overall performance of the subsequent model. In addition, the oversampling method increases the number of minority class samples by copying minority class samples or generating new samples. When the generated new samples are too simple or similar to minority class samples, this method is prone to overfitting. In addition, the cost-sensitive mode sets different misclassification costs for different categories of samples, so that the model pays more attention to minority class samples. This method requires reasonable adjustment of the cost parameters and may have poor results when processing complex data. That is, these methods can alleviate the problem of data quality class imbalance in the number of marine physical data to a certain extent, but these methods are relatively limited in their effect on processing complex marine physical data, and the quality of the generated data samples is poor.

[0024] In view of this, the embodiment of the present application provides a sample augmentation method and system for imbalanced data quality categories, wherein the method can restore the joint distribution of multidimensional data layer by layer during the denoising process by performing forward noise diffusion and reverse denoising on ocean samples, effectively capturing complex multidimensional data distribution, so that the generated new samples can maintain the distribution characteristics consistent with the original data in each feature dimension, effectively suppressing the overfitting phenomenon, thereby improving the quality of the generated data samples. In addition, the method corrects the noise layer by layer during the denoising process to ensure that the generated samples not only ensure the similarity of data distribution, but also introduce diversity due to noise disturbance, which can improve the generalization of minority class samples; in addition, the method introduces physical prior information (such as the second denoised sample) in each layer of denoising process, which can ensure the physical consistency of the generated samples and improve the prediction performance of the subsequent model on the minority class.

[0025] A sample augmentation method and system for imbalanced data quality categories provided in an embodiment of the present application can be specifically illustrated by the following embodiments. First, a sample augmentation method for imbalanced data quality categories in an embodiment of the present application is described.

[0026] The sample augmentation method for unbalanced daily data quality categories provided in the embodiment of the present application can be applied to marine physical data processing application scenarios. In the marine physical data processing application scenarios, the original marine physical data can be augmented by the method provided in the embodiment of the present application, which can improve the effect of processing complex marine physical data and provide high-quality marine physical data for subsequent research on global climate change, seabed resource development, geological disaster prediction (such as earthquakes and volcanic activities), etc.

[0027] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0028] Reference Figure 1 , Figure 1is an optional flowchart of a sample augmentation method for imbalanced data quality categories provided in an embodiment of the present application. Figure 1 The sample augmentation method for imbalanced data quality categories may include but is not limited to steps S101 to S104.

[0029] Step S101, obtaining original ocean physical data, and preprocessing the original ocean physical data to obtain ocean characteristic data; In the embodiments of the present application, the original ocean physical data may be ocean physical data disclosed on the Internet, for example, the disclosed ocean physical data may be obtained from a public ocean geophysical data platform (such as the International Ocean Drilling Program IODP), or it may be ocean physical data measured at different sampling points around the world. The present application does not limit the method for obtaining ocean physical data. In addition, the original ocean physical data in the embodiments of the present application may be any one of temperature, pressure, salinity, marine sediments, ocean current data, etc. The embodiments of the present application take the original ocean physical data as ocean heat flow data as an example for illustration, and the same is true for other types of ocean physical data, which can be simply deduced by analogy.

[0030] It can be understood that for ocean heat flow data, it usually includes characteristic values ​​of multiple dimensions such as sampling point number, longitude and latitude coordinates, sampling water depth, sampling altitude, collection time, geothermal gradient, thermal conductivity, heat flow value, etc. The preprocessing in step S101 can be data cleaning, data conversion and data standardization of the characteristic values ​​of multiple dimensions in the ocean heat flow data, so as to obtain ocean characteristic data.

[0031] Reference Figure 2 In some embodiments, the step S101, preprocessing the original ocean physical data to obtain ocean characteristic data, includes: A1. Cleaning the original ocean physical data to obtain first intermediate data; A2. Perform quartile anomaly detection on the first intermediate data to obtain second intermediate data; A3. Normalize the second intermediate data to obtain the ocean characteristic data.

[0032] In an embodiment of the present application, step A1 may be to clean the characteristic values ​​of each dimension of the original ocean physical data. If a characteristic value of a certain dimension is missing in the original ocean physical data, the original ocean physical data corresponding to the missing dimensional characteristic value may be deleted to obtain the first intermediate data.

[0033] It is understandable that the quartile anomaly detection in step A2 can be based on the interquartile range (IQR) method to identify and remove outliers in the first intermediate data. Specifically, the embodiment of the present application can perform outlier detection on the dimensional characteristic values ​​of the geothermal gradient, thermal conductivity and heat flow value in the first intermediate data one by one, and mark the dimensional characteristic values ​​of the geothermal gradient, thermal conductivity and heat flow value that are less than the lower limit value, or the dimensional characteristic values ​​of the geothermal gradient, thermal conductivity and heat flow value that are greater than the upper limit value as abnormal notes and delete them, so as to obtain the second intermediate data.

[0034] It should be noted that the data normalization in step A3 can be Z-score normalization of the dimensional feature values ​​such as sampling water depth, sampling altitude, acquisition time, geothermal gradient, thermal conductivity, and heat flow value in the second intermediate data, so that each dimensional feature value has the same scale. For a certain second intermediate data, the equivalent expression of its Z-score normalization can be:

[0035] in, is the j-th normalized dimensional feature value of the i-th second intermediate data; is the j-th dimension eigenvalue of the i-th second intermediate data; is the mean corresponding to all the eigenvalues ​​of the jth dimension; is the standard deviation corresponding to all j-th dimension eigenvalues.

[0036] It is worth mentioning that since the dimensional feature values ​​corresponding to the longitude and latitude coordinates of the second intermediate data are not suitable for Z-score normalization, the data normalization in step A3 also includes performing Min-Max Normalization on the longitude and latitude coordinates of the second intermediate data to obtain the dimensional feature values ​​of the normalized longitude and latitude coordinates; then, the ocean characteristic data is obtained by integrating all normalized dimensional feature values.

[0037] Step S102: performing quality classification processing on the ocean characteristic data to obtain ocean classification data, wherein the ocean classification data includes ocean samples and sample category labels, and the sample category labels are used to characterize the data quality category of the corresponding ocean samples; In an embodiment of the present application, after all ocean characteristic data are obtained, the data quality of each ocean characteristic data can be evaluated and classified. While the classified ocean characteristic data is determined as an ocean sample, a sample category label is added to each ocean sample. The sample category label is used to characterize the data quality category of the corresponding ocean sample. Among them, the data quality category in the embodiment of the present application can be the quality level of the ocean sample, which can be specifically divided into excellent, good, qualified, poor, etc. The examples in this application are for illustration only and are not intended to limit the present application.

[0038] Reference Figure 3 In some embodiments, the step S102 of performing quality classification processing on the ocean characteristic data to obtain ocean classification data includes: B1. Obtaining the data standard deviation and data mean of the ocean characteristic data, and a preset ocean data quality classification table; B2. performing data quality analysis on the ocean characteristic data according to the data standard deviation and the data mean to obtain a data quality analysis result, wherein the data quality analysis result is used to characterize the degree of dispersion of the ocean characteristic data; B3. According to the data quality analysis result, the ocean data quality classification table is matched with the quality category to obtain the ocean sample and the sample category label corresponding to the ocean sample.

[0039] In an embodiment of the present application, the ocean data quality table in step B1 can be an ocean data quality classification table based on the relative coefficient of variation. In the table, the relative coefficient of variation in each range corresponds to a data quality category. For example, when the relative coefficient of variation is less than or equal to 10%, the corresponding data quality category can be excellent, when the relative coefficient of variation is greater than 10% and less than or equal to 20%, the corresponding data quality category can be good, when the relative coefficient of variation is greater than 20% and equal to 30%, the corresponding data quality category can be qualified, and when the relative coefficient of variation is greater than 30%, the corresponding data quality category can be poor, etc. The present application only illustrates the range example of the relative coefficient of variation. The specific range of the relative coefficient of variation can be flexibly set according to actual conditions, and the present application does not limit it here.

[0040] It can be understood that since the geothermal gradient and thermal conductivity of the ocean characteristic data have their own uncertainties (i.e., standard deviations), and the heat flow value in the ocean characteristic data can be calculated by the geothermal gradient and thermal conductivity, the equivalent expression of the specific heat flow value can be:

[0041] Where HF is the heat flow value of ocean characteristic data; k is the geothermal gradient; is the thermal conductivity.

[0042] In the embodiment of the present application, the uncertainty of the ocean characteristic data can be obtained through the Error Propagation Formula, so as to obtain the data standard deviation for measuring the uncertainty of the ocean characteristic data. The data standard deviation can also characterize the degree of discreteness of the ocean characteristic data.

[0043] It should be noted that after obtaining the data standard deviation and the data mean, the data quality analysis in step B2 can be based on the relative coefficient of variation to quantitatively characterize the data quality of the ocean characteristic data, and obtain the relative coefficient of variation corresponding to the ocean characteristic data (i.e., the data quality analysis result). The equivalent expression of the data quality analysis result can be:

[0044] Among them, COV is the data quality analysis result corresponding to the ocean characteristic data; is the data standard deviation; is the data mean.

[0045] It is worth mentioning that the quality category matching in step B3 can be based on the relative coefficient of variation of each range in the ocean data quality table, and the data quality analysis results can be range matched to obtain the relative coefficient of variation and data quality category of the target range, and then the data quality category is assigned as a label to the corresponding ocean feature data, thereby obtaining ocean samples and corresponding sample category labels.

[0046] It is worth noting that, in an optional implementation, the ocean samples in the embodiment of the present application can also be obtained by screening out characteristic channel values ​​such as longitude and latitude, geothermal gradient, thermal conductivity, heat flow value, etc. in the ocean characteristic data after assigning the data quality category as a label to the corresponding ocean characteristic data, and determining the corresponding ocean samples based on the screened characteristic channel values.

[0047] Step S103: According to the sample category label, forward noise diffusion is performed on the ocean sample to obtain a noise sample sequence, wherein the noise sample sequence includes noise samples of several different noise adding time steps, and the noise sample is an ocean sample in a Gaussian noise state; In an embodiment of the present application, step S103 can be based on the sample category label, adding Gaussian noise to the ocean sample one by one at a noisy time step until the ocean sample completely loses information, thereby obtaining noise samples at different noisy time steps, each noise sample is an ocean sample after adding a number of Gaussian noises, wherein the number of Gaussian noises added to the noise sample is the same as the noisy time step, and the Gaussian noise can be a type of noise that obeys a normal distribution.

[0048] It should be noted that the noise sample sequence in the embodiment of the present application can be a noise sample sequence constructed from one ocean sample, or a noise sample sequence constructed from multiple ocean samples. Specifically, the embodiment of the present application can be based on an ocean sample, by gradually adding Gaussian noise of different noise adding time steps to the ocean sample, and determining the noise sample of each noise adding time step as a noise sample sequence, and each noise sample in the noise sample sequence corresponds to an ocean sample; or, the embodiment of the present application can construct an ocean sample matrix based on multiple ocean samples, each matrix element is used to characterize an ocean sample, and then gradually add Gaussian noise of different noise adding time steps to the ocean sample matrix, so as to obtain a noise sample sequence, each noise sample in the noise sample sequence contains a number of noise sub-samples, and each noise sub-sample corresponds to a matrix element in the ocean sample matrix. The examples of this application are for illustration only and are not intended to limit this application.

[0049] Reference Figure 4 In some embodiments, the step S103, performing forward noise diffusion on the ocean sample according to the sample category label to obtain a noise sample sequence, includes: C1. Obtain the preset sample threshold; C2. Screening all the ocean samples according to the sample category labels to obtain a number of intermediate sample sets, where the intermediate sample sets are used to represent a collection of ocean samples with the same data quality category; C3. According to the sample threshold, all the intermediate sample sets are screened for sample quantity to obtain a plurality of target sample matrices, wherein the target sample matrices are used to characterize a matrix set of target samples, wherein the target samples are ocean samples in the intermediate sample set whose sample quantity is less than the sample threshold, and each target sample matrix corresponds to an intermediate sample set whose sample quantity is less than the sample threshold; In an embodiment of the present application, the sample threshold in step C1 is used to characterize the total target sample number of marine samples of each data quality category. The specific value of the sample threshold can be set according to the actual situation, such as any one of 100, 1000, 5000, etc., or the specific value of the sample threshold can be determined according to the relative relationship of marine samples of each data quality category. There are many specific determination methods, which will not be repeated in this application. In addition, the sample screening in step C2 can be based on the sample category label corresponding to each marine sample, and the marine samples with the same sample category label are divided into the same data set, so as to obtain several intermediate sample sets representing marine samples of the same data quality category, and each intermediate sample set corresponds to a different data quality category.

[0050] It can be understood that, for a certain intermediate sample set, the sample quantity screening in step C3 can first obtain the total number of ocean samples in the intermediate sample set (i.e., the sample number), and then compare the size relationship between the sample threshold and the sample number in the intermediate sample set. If the number of samples in the intermediate sample set is greater than or equal to the sample threshold, the intermediate sample set can be excluded; or, if the number of samples in the intermediate sample set is less than the sample threshold, then the corresponding target sample matrix can be generated based on each ocean sample in the intermediate sample set, and the same applies to the remaining intermediate sample sets.

[0051] For example, for a target sample matrix of a certain data quality category, if each ocean sample in the target sample matrix includes characteristic channel values ​​such as longitude and latitude, geothermal gradient, thermal conductivity, and heat flow value, the matrix expression of the target sample matrix can be:

[0052] in, is the nth target sample matrix; Lon is the longitude in the longitude and latitude feature channel value; Lat is the latitude in the longitude and latitude feature channel value; is the characteristic channel value of geothermal gradient; k is the characteristic channel value of thermal conductivity; HF is the characteristic channel value of heat flow; y is the data quality category; m is the total number of ocean samples in the target sample matrix.

[0053] C4. Add forward noise to the target sample matrix to obtain the noise sample sequence.

[0054] Reference Figure 5 Further, the step C4, adding forward noise to the target sample matrix to obtain the noise sample sequence, comprises: C41, obtain the preset noise adding time step threshold and Gaussian noise; C42, taking the current noise addition time step as the first time step, and obtaining the cumulative product parameter corresponding to the first time step; C43, performing sample noise addition on the target sample matrix according to the cumulative product parameter and the Gaussian noise to obtain a noisy sample matrix; C44, comparing the first time step and the noise adding time step threshold to obtain a first comparison result; C45. If the first comparison result is that the first time step is less than the noise adding time step threshold, the noise adding sample matrix is ​​determined as the noise sample corresponding to the first time step, and the step of taking the current noise adding time step as the first time step and obtaining the cumulative product parameter corresponding to the first time step is returned to be executed; Alternatively, C46, ​​if the first comparison result is that the first time step is equal to the noise addition time step threshold, the noise addition sample matrix is ​​determined as the noise sample corresponding to the first time step, and the noise sample sequence is generated according to the noise samples of all different noise addition time steps.

[0055] In an embodiment of the present application, step C4 can be to gradually add Gaussian noise with a variance in the range of [0, 1] and obeying a normal distribution to the target sample matrix of the minority class. The variance of the Gaussian noise corresponding to each noise-adding time step can be the same or different. Specifically, when adding Gaussian noise, a unit matrix with all elements on the main diagonal being 1 can be used to ensure that the Gaussian noise added at each noise-adding time step is independent and identically distributed isotropic noise, so that the noise samples in the noise sample sequence present a Gaussian noise state. Specifically, the noise-adding time step threshold in step C41 is the total number of steps for adding Gaussian noise to the target sample matrix of the minority class, and its specific data can be set according to actual conditions, such as 500, 1000, 5000, etc.

[0056] It can be understood that the cumulative product parameter in step C42 is the product of the intermediate variables, and the intermediate variables can be determined based on the variance of the Gaussian noise corresponding to the corresponding noise adding time step. Specifically, if the first time step is t, the equivalent expression of the intermediate variables and the cumulative product parameter in the embodiment of the present application can be:

[0057] in, is the intermediate variable when the first time step is t; is the variance of Gaussian noise; is the cumulative product parameter corresponding to the first time step t; is the intermediate variable when the first time step is q.

[0058] It should be noted that step C43 may be based on the cumulative product parameter, adding Gaussian noise to the target sample matrix, so as to obtain a noisy sample matrix corresponding to the first time step, and the equivalent expression of the noisy sample matrix may be:

[0059] in, is the noise sample matrix corresponding to the first time step; is the target sample matrix; ϵ is Gaussian noise.

[0060] It is worth mentioning that step C44 can be to compare the size relationship between the first time step and the noise time step threshold. If the first comparison result is that the first time step is less than the noise time step threshold, it means that there is at least one noise time step of Gaussian noise that has not been added to the target sample matrix. At this time, the currently obtained noise sample matrix can be determined as the noise sample corresponding to the current noise time step, and the current noise time step can be updated. Specifically, the current noise time step can be added by 1, and then return to execute step C42; or, if the first comparison result is that the first time step is equal to the noise time step threshold, it means that there is no Gaussian noise of any noise time step that has not been added to the target sample matrix. At this time, the currently obtained noise sample matrix can be determined as the noise sample corresponding to the current noise time step, and based on the size relationship of the noise time step, the noise samples of all different noise time steps are assembled into a noise sample sequence.

[0061] Step S104: input the noise sample sequence into a sample denoising model for reverse denoising to obtain augmented samples corresponding to the noise sample sequence.

[0062] In an embodiment of the present application, step S104 may be to input the noise sample sequence into a sample denoising model for reverse denoising, learn the transition probability distribution between noise samples of adjacent noisy time steps in the noise sample sequence through the sample denoising model, and based on all learned transition probability distributions, gradually denoise the noise sample of the last noisy time step to obtain an augmented sample.

[0063] Reference Figure 6 and Figure 7 In some embodiments, the step S104, inputting the noise sample sequence into a sample denoising model for reverse denoising to obtain augmented samples corresponding to the noise sample sequence, includes: D1. Get the preset denoising time step threshold; D2, taking the current denoising time step as the second time step, and obtaining a first denoising sample corresponding to the second time step; D3, inputting the first denoising sample into the sample denoising model to perform sample denoising prediction to obtain predicted noise; D4. Calculate the posterior probability distribution of a second target noise sample according to the first target noise sample corresponding to the first denoised sample to obtain a posterior mean corresponding to the predicted noise, where the second target noise sample is a noise sample of a noise adding time step before the first target noise sample; D5. Perform sample denoising on the first denoised sample according to the posterior mean and the predicted noise to obtain a second denoised sample; D6. Compare the second time step with the denoising time step threshold to obtain a third comparison result; D7. If the third comparison result is that the second time step is greater than the denoising time step threshold, the first denoising sample is updated according to the second denoising sample, and then the step of returning to the current denoising time step as the second time step and obtaining the first denoising sample corresponding to the second time step is executed; Alternatively, D8: if the third comparison result is that the second time step is equal to the denoising time step threshold, the second denoised sample is determined as the augmented sample.

[0064] In the embodiment of the present application, the denoising time step threshold in step D1 is similar to the aforementioned denoising time step threshold, and can be simply derived by analogy. The first denoising sample in step D2 can be the noise sample of the last denoising time step in the noise sample sequence, or it can be the second denoising sample obtained by the denoising process of the previous second time step. Specifically, if the second time step is the first denoising time step in the denoising time step sequence, it means that the noise sample has not undergone the denoising process. At this time, the first denoising sample corresponding to the first time step can be the noise sample of the last denoising time step in the noise sample sequence; or, if the second time step is the second or later denoising time step in the denoising time step sequence, then the first denoising sample corresponding to the second time step can be the second denoising sample obtained by the denoising process of the previous second time step. For example, when the second time step is the third denoising time step in the denoising time step sequence, the corresponding first denoising sample can be the second denoising sample corresponding to the second denoising time step. The example of this application is for illustration only and is not intended to limit this application.

[0065] In addition, the denoising time step sequence in the embodiment of the present application can be a reverse order of the denoising time step sequence. For example, the denoising time step sequence can be composed of 10 time steps sorted in ascending order. At this time, the corresponding denoising time step sequence can be composed of 10 or more time steps sorted in descending order, that is, the first time step in the denoising time step sequence is the tenth time step in the denoising time step sequence, the second time step in the denoising time step sequence is the ninth time step in the denoising time step sequence, and the eleventh time step in the denoising time step sequence can be newly obtained on the basis of reverse order of the denoising time step sequence. The same applies to the remaining time steps. The example of this application is only for illustration and will not be elaborated in this application.

[0066] It can be understood that step D3 can be to input the first denoised sample into the sample denoising model for sample denoising prediction, and determine the predicted noise corresponding to the first denoised sample through the transition probability distribution previously learned by the sample denoising model. Exemplarily, the embodiment of the present application takes the second time step as the first denoised time step in the denoising time step sequence, and its corresponding first target noise sample is , taking the U-Net neural network model as an example, the transition probability distribution learned by the sample denoising model can be expressed as:

[0067] in, is the noise sample from time step t Go back to the noise sample at time step t-1 The transition probability distribution of . is a parameterized Gaussian distribution. is the mean value estimated for each time step in the denoising process by the U-Net neural network model, which is used to characterize the noise sample at time step t The estimated noise sample The most likely value of . is the covariance matrix estimated by the U-Net neural network model for each time step during the denoising process, which is used to characterize the denoising uncertainty at time step t. θ is the trained U-Net neural network model parameter.

[0068] It should be noted that after obtaining the predicted noise corresponding to the first denoised sample, step D4 may be to calculate the posterior probability distribution between the first target noise sample and the second target noise sample. Specifically, if the first denoised sample is the noise sample of the last noise adding time step, then the first target noise sample corresponding to it may also be the noise sample of the last noise adding time step, and the second target noise sample is the noise sample of the second to last noise adding time step. For another example, if the second time step corresponding to the first denoised sample is the second denoised time step, then the first target noise sample corresponding to the first denoised sample at this time is the noise sample of the second to last noise adding time step, and the second target noise sample may be the noise sample of the third to last noise adding time step, and then the posterior mean corresponding to the predicted noise is obtained.

[0069] Exemplarily, if the first target noise sample is , the second target noise sample is When , the equivalent expression of its posterior mean can be:

[0070] in, is the posterior mean; is the intermediate variable at time step t; is the Gaussian noise at time step t; is the cumulative product parameter at time step t.

[0071] It is worth mentioning that step D5 can be based on the posterior mean and the predicted noise, the first denoised sample is randomly sampled to obtain the second denoised sample, the equivalent expression of the second denoised sample can be:

[0072] in, is the second denoised sample; is the sample denoising model; z is the additional noise that obeys the normal distribution; is the standard deviation of the noise term at time step t.

[0073] It is worth noting that step D6 can be to compare the size relationship between the second time step and the denoising time step threshold to obtain a third comparison result. Specifically, if the third comparison result is that the second time step is greater than the denoising time step threshold, the second time step can be subtracted by 1, and the current second denoising sample can be used as the first denoising sample in the next denoising process, and then return to execute step D2; or, if the third comparison result is that the second time step is equal to the denoising time step threshold, the final second denoising sample can be determined as the augmented sample obtained after data augmentation.

[0074] It should be added that, in the embodiment of the present application, the quality evaluation of the closeness of the sample distribution of the augmented samples to the data distribution of the original ocean physical data can be performed based on evaluation indicators such as KL divergence, Wasserstein distance and root mean square error (RMSE), so as to obtain the augmented sample quality evaluation result. If the augmented sample quality evaluation result is significantly different from the data distribution of the original ocean physical data, the parameters of the sample denoising model can be updated by the back propagation algorithm. Specifically, the noise attenuation factor, diffusion steps, optimizer, learning rate, U-net structure and other parameters can be adjusted to achieve the update of the sample denoising model parameters. There are many specific implementation methods, which will not be repeated in this application.

[0075] Reference Figure 8 In some embodiments, the method further comprises: E1. Obtain a preset sample augmentation threshold; E2. Get the current augmented sample and sample augmented sequence; E3. Compare the current sample augmentation sequence with the sample augmentation threshold to obtain a second comparison result; E4. If the second comparison result is that the total number of augmented samples in the sample augmented sequence is less than the sample augmented threshold, the sample augmented sequence is updated according to the augmented samples, and then the step of obtaining the current augmented samples and sample augmented sequence is returned to be executed; Alternatively, E5: if the second comparison result is that the total number of augmented samples in the sample augmented sequence is equal to the sample augmented threshold, obtaining a sample augmented set according to the sample augmented sequence.

[0076] In the embodiment of the present application, the sample augmentation threshold is used to characterize the target sample data quantity of a certain data quality category of the ocean data, which can be set according to the actual situation. For ocean data of a certain data quality category, the current augmented sample can be an augmented sample obtained at the current latest time step and not added to the sample augmented sequence, and the augmented sample can be obtained based on step S103 and step S104. The current sample augmented sequence can be a sequence set of historical augmented samples added to the sample augmented sequence at the previous time step.

[0077] It can be understood that step E3 can be to compare the total number of augmented samples in the sample augmented sequence with the sample augmented threshold value to obtain a second comparison result. Specifically, if the second comparison result is that the total number of augmented samples in the sample augmented sequence is less than the sample augmented threshold value, the current augmented sample can be added to the sample augmented sequence, and then the process returns to step E2; or if the second comparison result is that the total number of augmented samples in the sample augmented sequence is equal to the sample augmented threshold value, the current sample augmented sequence can be determined as a sample augmented set and spliced ​​with the original ocean physical data to obtain ocean data with balanced data quality categories.

[0078] Fig. 9 A schematic diagram of a sample augmentation system for imbalanced data quality categories provided in an embodiment of the present application, the system comprising: The first processing unit 810 is used to obtain original ocean physical data and pre-process the original ocean physical data to obtain ocean characteristic data; A second processing unit 820 is used to perform quality classification processing on the ocean characteristic data to obtain ocean classification data, wherein the ocean classification data includes ocean samples and sample category labels, and the sample category labels are used to characterize the data quality category of the corresponding ocean samples; A third processing unit 830 is used to perform forward noise diffusion on the ocean sample according to the sample category label to obtain a noise sample sequence, wherein the noise sample sequence includes noise samples of several different noise adding time steps, and the noise sample is an ocean sample in a Gaussian noise state; The fourth processing unit 840 is used to input the noise sample sequence into a sample denoising model to perform reverse denoising to obtain augmented samples corresponding to the noise sample sequence.

[0079] It can be understood that the contents of the above method embodiments are all applicable to the present system embodiments, the functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0080] Fig.10 A schematic diagram of the structure of a computer device provided in an embodiment of the present application includes: at least one processor 980; At least one memory 920, used to store at least one program; When the at least one program is executed by the at least one processor 980, the at least one processor 980 implements the methods described in the foregoing embodiments.

[0081] An embodiment of the present application also provides a computer-readable storage medium, which stores a program executable by a processor. When the program executable by the processor is executed by the processor 980, it is used to implement the methods described in the above embodiments.

[0082] Specifically, the computer device may be a user terminal or a server.

[0083] The present application embodiment takes the computer device as a user terminal as an example, and the details are as follows: like Fig.10 As shown, the computer device 900 may include an RF (Radio Frequency) circuit 910, a memory 920 including one or more computer-readable storage media, an input unit 930, a display unit 940, a sensor 950, an audio circuit 960, a WiFi module 970, a processor 980 including one or more processing cores, and a power supply 990. Those skilled in the art will appreciate that Fig.10 The device structure shown in the figure does not constitute a limitation on the electronic device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently. The RF circuit 910 can be used for receiving and sending signals during information transmission or calls. In particular, after receiving the downlink information of the base station, it is handed over to one or more processors 980 for processing; in addition, the data related to the uplink is sent to the base station. Generally, the RF circuit 910 includes but is not limited to an antenna, at least one amplifier, a tuner, one or more oscillators, a user identity module (SIM) card, a transceiver, a coupler, an LNA (Low Noise Amplifier), a duplexer, etc. In addition, the RF circuit 910 can also communicate with the network and other devices through wireless communication. Wireless communication can use any communication standard or protocol, including but not limited to GSM (Global System of Mobile communication), GPRS (General Packet Radio Service), CDMA (Code Division Multiple Access), WCDMA (Wideband Code Division Multiple Access), LTE (Long Term Evolution), email, SMS (Short Messaging Service), etc. The memory 920 can be used to store software programs and modules. The processor 980 executes various functional applications and data processing by running the software programs and modules stored in the memory 920. The memory 920 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the device 900 (such as audio data, a phone book, etc.), etc. In addition, the memory 920 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices. Accordingly, the memory 920 may also include a memory controller to provide the processor 980 and the input unit 930 with access to the memory 920. Although Fig.10 The RF circuit 910 is shown, but it is understandable that it is not an essential component of the device 900 and can be omitted as required without changing the essence of the invention.

[0084] The input unit 930 can be used to receive input digital or character information, and generate keyboard, mouse, joystick, optical or trackball signal input related to user settings and function control. Specifically, the input unit 930 may include a touch-sensitive surface 932 and other input devices 931. The touch-sensitive surface 932, also known as a touch display screen or a touch pad, can collect user touch operations on or near it (such as operations performed by users using fingers, styluses, or any other suitable objects or accessories on or near the touch-sensitive surface 932), and drive corresponding connection devices according to a pre-set program. Optionally, the touch-sensitive surface 932 may include a touch detection device and a touch controller. Among them, the touch detection device detects the user's touch orientation, detects the signal brought by the touch operation, and transmits the signal to the touch controller; the touch controller receives the touch information from the touch detection device, converts it into touch point coordinates, and then sends it to the processor 980, and can receive and execute commands sent by the processor 980. In addition, the touch-sensitive surface 932 can be implemented in various types such as resistive, capacitive, infrared, and surface acoustic waves. In addition to the touch-sensitive surface 932, the input unit 930 may further include other input devices 931. Specifically, the other input devices 931 may include, but are not limited to, one or more of a physical keyboard, function keys (such as volume control keys, switch keys, etc.), a trackball, a mouse, a joystick, and the like. The display unit 940 can be used to display information input by the user or information provided to the user and various graphical user interfaces of the device 900, which can be composed of graphics, text, icons, videos and any combination thereof. The display unit 940 may include a display panel 941. Optionally, the display panel 941 may be configured in the form of LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode), etc. Further, the touch-sensitive surface 932 may be covered on the display panel 941. When the touch-sensitive surface 932 detects a touch operation on or near it, it is transmitted to the processor 980 to determine the type of touch event. Subsequently, the processor 980 provides a corresponding visual output on the display panel 941 according to the type of touch event. Although in Fig.10 In the embodiment, the touch-sensitive surface 932 and the display panel 941 are implemented as two independent components to implement input and output functions, but in some embodiments, the touch-sensitive surface 932 and the display panel 941 can be integrated to implement input and output functions.

[0085] The computer device 900 may also include at least one sensor 950, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor, wherein the ambient light sensor may adjust the brightness of the display panel 941 according to the brightness of the ambient light, and the proximity sensor may turn off the display panel 941 and / or the backlight when the device 900 is moved to the ear. As a type of motion sensor, the gravity acceleration sensor can detect the magnitude of acceleration in each direction (generally three axes), and can detect the magnitude and direction of gravity when stationary, which can be used for applications that identify the posture of the mobile phone (such as horizontal and vertical screen switching, related games, magnetometer posture calibration), vibration recognition related functions (such as pedometer, tapping), etc.; as for other sensors such as gyroscopes, barometers, hygrometers, thermometers, infrared sensors, etc. that can also be configured in the device 900, they will not be repeated here.

[0086] The audio circuit 960, the speaker 961, and the microphone 962 can provide an audio interface between the user and the device 900. The audio circuit 960 can transmit the electrical signal converted from the received audio data to the speaker 961, which is converted into a sound signal for output; on the other hand, the microphone 962 converts the collected sound signal into an electrical signal, which is received by the audio circuit 960 and converted into audio data, and then the audio data is processed by the output processor 980 and sent to another control device through the RF circuit 910, or the audio data is output to the memory 920 for further processing. The audio circuit 960 may also include an earplug jack to provide communication between an external headset and the device 900.

[0087] The device 900 can transmit information with the wireless transmission module set on the fighting device through the WiFi module 970.

[0088] The processor 980 is the control center of the device 900. It uses various interfaces and lines to connect various parts of the entire control device. It executes various functions of the device 900 and processes data by running or executing software programs and / or modules stored in the memory 920, and calling data stored in the memory 920, so as to monitor the control device as a whole. Optionally, the processor 980 may include one or more processing cores; optionally, the processor 980 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the processor 980.

[0089] The device 900 also includes a power supply 990 (such as a battery) for supplying power to each component. Preferably, the power supply can be logically connected to the processor 980 through a power management system, so that the power management system can manage charging, discharging, and power consumption management. The power supply 990 can also include any components such as one or more DC or AC power supplies, recharging systems, power failure detection circuits, power converters or inverters, and power status indicators. Although not shown, the device 900 may also include a camera, a Bluetooth module, etc., which will not be described in detail here.

[0090] An embodiment of the present application further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the method described in the above embodiments.

[0091] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein, for example. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0092] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0093] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0094] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0095] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0096] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store program codes.

[0097] The step numbers in the above method embodiment are only provided for the convenience of explanation and description, and no limitation is imposed on the order of the steps. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0098] The above is a specific description of the preferred implementation of the present application, but the present application is not limited to the described embodiments. Technical personnel familiar with the field may make various equivalent modifications or substitutions without violating the spirit of the present application. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present application.

Claims

1. A sample augmentation method for imbalanced data quality categories, characterized in that: The method comprises: Acquiring original ocean physical data, and preprocessing the original ocean physical data to obtain ocean characteristic data; Performing quality classification processing on the ocean characteristic data based on the relative coefficient of variation to obtain ocean classification data, wherein the ocean classification data includes ocean samples and sample category labels, and the sample category labels are used to characterize the data quality category of the corresponding ocean samples; According to the sample category label, forward noise diffusion is performed on the ocean sample to obtain a noise sample sequence, wherein the noise sample sequence includes noise samples of several different noise adding time steps, and the noise sample is an ocean sample in a Gaussian noise state; The noise sample sequence is input into a sample denoising model for reverse denoising to obtain augmented samples corresponding to the noise sample sequence.

2. A sample augmentation method for data quality category imbalance according to claim 1, characterized in that: The preprocessing of the original ocean physical data to obtain ocean characteristic data includes: Performing data cleaning on the original ocean physical data to obtain first intermediate data; Performing quartile anomaly detection on the first intermediate data to obtain second intermediate data; The second intermediate data is normalized to obtain the ocean characteristic data.

3. The sample augmentation method for data quality category imbalance according to claim 1, characterized in that: The performing of quality classification processing on the ocean characteristic data based on the relative coefficient of variation to obtain ocean classification data includes: Obtaining the data standard deviation and data mean of the ocean characteristic data, and a preset ocean data quality classification table based on relative coefficient of variation; Performing data quality analysis on the ocean characteristic data according to the data standard deviation and the data mean to obtain a data quality analysis result, wherein the data quality analysis result is used to characterize the degree of dispersion of the ocean characteristic data; According to the data quality analysis result, the ocean data quality classification table is matched with quality categories to obtain the ocean samples and the sample category labels corresponding to the ocean samples.

4. The method for sample augmentation with imbalanced data quality categories according to claim 1, characterized in that: The step of performing forward noise diffusion on the ocean sample according to the sample category label to obtain a noise sample sequence includes: Get the preset sample threshold; According to the sample category labels, all the ocean samples are screened to obtain a plurality of intermediate sample sets, where the intermediate sample sets are used to characterize a collection of ocean samples with the same data quality category; According to the sample threshold, all the intermediate sample sets are screened for sample quantity to obtain a plurality of target sample matrices, wherein the target sample matrices are used to characterize a matrix set of target samples, wherein the target samples are ocean samples in the intermediate sample set whose sample quantity is less than the sample threshold, and each target sample matrix corresponds to an intermediate sample set whose sample quantity is less than the sample threshold; Forward noise is added to the target sample matrix to obtain the noise sample sequence.

5. The method for sample augmentation for data quality category imbalance according to claim 4, characterized in that: The step of adding forward noise to the target sample matrix to obtain the noise sample sequence includes: Get the preset noise adding time step threshold and Gaussian noise; Taking the current noise adding time step as the first time step, obtaining the cumulative product parameter corresponding to the first time step; Performing sample noise addition on the target sample matrix according to the cumulative product parameter and the Gaussian noise to obtain a noisy sample matrix; Comparing the first time step with the noise adding time step threshold to obtain a first comparison result; If the first comparison result is that the first time step is less than the noise adding time step threshold, the noise adding sample matrix is ​​determined as the noise sample corresponding to the first time step, and the step of taking the current noise adding time step as the first time step and obtaining the cumulative product parameter corresponding to the first time step is returned to be executed; or, if the first comparison result is that the first time step is equal to the noise adding time step threshold, the noise adding sample matrix is ​​determined as the noise sample corresponding to the first time step, and the noise sample sequence is generated according to the noise samples of all different noise adding time steps.

6. The method for sample augmentation for data quality category imbalance according to claim 1, characterized in that: The method further comprises: Get the preset sample augmentation threshold; Get the current augmented sample and sample augmented sequence; Comparing the current sample augmentation sequence with the sample augmentation threshold to obtain a second comparison result; If the second comparison result is that the total number of augmented samples in the sample augmented sequence is less than the sample augmented threshold, the sample augmented sequence is updated according to the augmented samples, and then the step of obtaining the current augmented samples and the sample augmented sequence is returned to be executed; or, if the second comparison result is that the total number of augmented samples in the sample augmented sequence is equal to the sample augmented threshold, a sample augmented set is obtained according to the sample augmented sequence.

7. The method for sample augmentation for data quality category imbalance according to claim 1, characterized in that: The step of inputting the noise sample sequence into a sample denoising model for reverse denoising to obtain augmented samples corresponding to the noise sample sequence includes: Get the preset denoising time step threshold; Taking the current denoising time step as the second time step, obtaining a first denoising sample corresponding to the second time step; Inputting the first denoising sample into the sample denoising model to perform sample denoising prediction to obtain predicted noise; According to the first target noise sample corresponding to the first denoised sample, a posterior probability distribution calculation is performed on the second target noise sample to obtain a posterior mean corresponding to the predicted noise, wherein the second target noise sample is a noise sample of a noise adding time step before the first target noise sample; Performing sample denoising on the first denoised sample according to the posterior mean and the predicted noise to obtain a second denoised sample; Comparing the second time step with the denoising time step threshold to obtain a third comparison result; If the third comparison result is that the second time step is greater than the denoising time step threshold, the first denoised sample is updated according to the second denoised sample, and then the step of returning to execute the step of taking the current denoised time step as the second time step and acquiring the first denoised sample corresponding to the second time step; or, if the third comparison result is that the second time step is equal to the denoising time step threshold, the second denoised sample is determined as the augmented sample.

8. A sample augmentation system for imbalanced data quality categories, characterized in that: The system comprises: A first processing unit is used to obtain original ocean physical data and pre-process the original ocean physical data to obtain ocean characteristic data; A second processing unit is used to perform quality classification processing on the ocean characteristic data based on the relative coefficient of variation to obtain ocean classification data, wherein the ocean classification data includes ocean samples and sample category labels, and the sample category labels are used to characterize the data quality category of the corresponding ocean samples; a third processing unit, configured to perform forward noise diffusion on the ocean sample according to the sample category label to obtain a noise sample sequence, wherein the noise sample sequence includes noise samples at a plurality of different noise adding time steps, and the noise sample is an ocean sample in a Gaussian noise state; The fourth processing unit is used to input the noise sample sequence into a sample denoising model to perform reverse denoising to obtain augmented samples corresponding to the noise sample sequence.

9. A computer device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a program executable by a processor, characterized in that: The program executable by the processor is used to implement the method according to any one of claims 1 to 7 when executed by the processor.

Images