A method for gridding and noise removal of geophysical potential field data

Abstract

The present invention provides a method for gridding and removing noise from geophysical potential field data, and relates to the technical field of gravity and magnetic exploration data processing. The method first establishes a set of original data sets including geophysical potential field data and noisy geophysical potential field data sets; inputs the specified original data set into a diffusion model for training to obtain a diffusion model; secondly, obtains the number of denoising steps corresponding to the noisy data set, and trains a noise level selection network with it and the noisy data set; then, according to the coordinates and outliers of the geophysical potential field data to be processed, determines the position of the scattered data, and assigns values ​​to the anomalies at the corresponding positions to generate scattered data; finally, uses the diffusion model to realize gridding of the scattered data, and uses the noise selection network model combined with the diffusion model to realize gridding data denoising. The method of the present invention can improve the accuracy of gridding of potential field data, more accurately restore the distribution characteristics of the potential field, and remove noise interference.

Description

Technical Field

[0001] The invention relates to the technical field of gravity and magnetic exploration data processing, and in particular to a geophysical potential field data gridding and noise removal method. Background Art

[0002] Due to the influence of factors such as the terrain of the data collection area, the division of measuring points and lines, the spatial distribution of the collected geophysical potential field data is often irregular. At the same time, due to the influence of factors such as measuring instruments and geological environment, the collected data often contains certain noise. In the process of data processing, the collected data is generally gridded; and noise removal helps to improve the accuracy of data interpretation. Therefore, gridding and noise removal of potential field data are one of the indispensable steps in data processing. Through gridding and noise removal, the foundation is laid for the interpretation of high-precision potential field data.

[0003] At present, both the commonly used numerical calculation methods and the equivalent source method have the problem of low gridding accuracy. In recent years, the rapid development of artificial intelligence technology has provided new ideas for solving the above problems. Many scholars have carried out a lot of research on the gridding method of geophysical potential field data based on artificial intelligence: Li An et al. used a neural network based on the Bagging algorithm to interpolate the background field of ocean gravity anomaly based on satellite altimetry, which improved the adaptive accuracy of the matching algorithm and the navigation accuracy; Liu Huan et al. proposed a recursive neural network reconstruction algorithm based on long short-term memory network on the basis of three machine learning models: support vector machine, random forest algorithm and gradient enhancement, and achieved good reconstruction effect compared with the traditional linear method. Similarly, in the denoising of geophysical potential field data, Lin Fanqiang et al. used a deep learning method to solve the problem of poor denoising effect when there is a huge difference between the noise distribution of transient electromagnetic data signals and the noise distribution of simulation data sets; Li Jin et al. used a method combining long short-term memory network and convolutional neural network to denoise data with strong noise, proving that its denoising effect is better than the characteristic parameter classification method.

[0004] The above research shows that artificial intelligence has great potential in geophysical potential field data gridding and denoising, and has achieved good results. Compared with traditional methods, intelligent methods can more completely preserve and restore geophysical data characteristics. Therefore, the research on intelligent methods for potential field data gridding and noise removal can provide technical support for future high-precision data processing and interpretation. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method for gridding and noise removal of geophysical potential field data in view of the above-mentioned deficiencies in the prior art, so as to realize gridding and noise removal of geophysical potential field data.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for gridding and removing noise of geophysical potential field data, comprising the following steps:

[0007] Step 1: Generate geophysical potential field data using the random model forward modeling method, and add noise to generate a geophysical potential field noisy data set;

[0008] Step 1.1: Generate the original geophysical potential field dataset using the stochastic model forward modeling method;

[0009] The random model forward modeling method is used to randomly generate rectangular models of different scales, shapes, positions, densities and magnetic susceptibilities in the determined survey area space, and the potential field anomaly data of the model is calculated to form the original geophysical potential field data set;

[0010] The potential field anomaly data includes gravity anomaly data, gravity gradient data, magnetic anomaly and magnetic gradient type data; the original geophysical potential field data set is composed of various types of potential field anomaly data, wherein each set of data represents an anomaly generated by a certain random model;

[0011] Step 1.2: Extract a certain proportion of geophysical potential field data and add different levels of noise to generate a geophysical potential field noisy data set;

[0012] The proportion of the extracted data should be at least 1% of the original geophysical potential field data set, and the number of each type of data should be the same, that is, the number of different component data and different anomaly data used should be consistent, and the added noise should be Gaussian noise;

[0013] Step 2: Use the original geophysical potential field dataset to train the diffusion model until a diffusion model that meets the requirements is obtained;

[0014] Step 2.1: Normalize the original geophysical potential field dataset;

[0015] Step 2.2: Use the normalized original geophysical potential field dataset to train the diffusion model;

[0016] The calculation process of the diffusion model includes two parts: a forward process and a reverse process;

[0017] Step 3: Calculate the number of denoising steps corresponding to each set of noisy data, and train the noise level selection network together with the noisy data set to obtain a noise level selection network model that meets the requirements;

[0018] Step 3.1: Calculate the mean square error (MSE) between the denoised data and the original geophysical potential field data without noise;

[0019] Step 3.2: Using the mean square error (MSE) value between the denoised data in step 3.1 and the original geophysical potential field data without noise, and using the diffusion model trained in step 2, obtain the number of denoising steps corresponding to each group of noisy data;

[0020] Let the number of iterations be k , k = 0, 1, 2, …, k max , k max Indicates the maximum number of iterations, calculates and records the noisy data k The mean square error (MSE) value at the iteration. The MSE value decreases as the number of iterations increases. If the MSE value increases, the number of reverse calculations is the current number of iterations. k Minus 1, that is, the number of denoising steps is k - 1;

[0021] Step 3.3: Input the number of denoising steps obtained in step 3.2 and the corresponding noisy data into the noise level selection network for training;

[0022] Normalize the noisy data set and use the classification network model as the noise level selection network; the classification network model uses cross entropy as the loss function;

[0023] Step 4: According to the coordinates and outliers of the geophysical potential field data to be processed, the location of the scattered data is determined, and the anomalies at the corresponding locations are assigned values ​​to generate scattered data;

[0024] Step 4.1: Determine the measuring points and measuring line distances of the scattered data according to the geophysical potential field data to be processed, and determine the spatial distribution range of the scattered data;

[0025] Step 4.2: Calculate the Euclidean distance between the position of the geophysical potential field data point to be processed and the position of the scattered data point, and assign a value to the designated scattered point position through the Euclidean distance;

[0026] Traversing the coordinates of the geophysical potential field data points to be processed, assigning the data to be processed to the corresponding position in the scattered data according to the corresponding minimum Euclidean distance, and obtaining the scattered data, which includes the data coordinates and the corresponding potential field data;

[0027] Step 5: Input the scattered data into the diffusion model and grid the scattered data;

[0028] The trained diffusion model is used to calculate the reverse process of the scattered data to obtain the missing data in the grid data. , which obeys the mathematical expectation The variance is The normal distribution of is expressed as follows:

[0029] in, and are the Gaussian model distribution parameters of the diffusion model, is the mathematical expectation of the Gaussian model distribution, is the variance of the Gaussian model distribution;

[0030] Finally, the gridded data of the geophysical potential field data to be processed is obtained The formula is expressed as:

[0031] in, For scattered data, is the gridded data;

[0032] Step 6: Input the gridded data into the noise level selection network, calculate the number of steps required for denoising, and combine the diffusion model for denoising;

[0033] Step 6.1: Calculate the number of denoising steps through the noise level selection network;

[0034] The gridded data generated in step 5 Input the noise level selection network, classify the noise contained in the gridded data through the noise level selection network, and calculate the number of steps required for denoising k ;

[0035] Step 6.2: Use the diffusion model combined with the number of denoising steps in step 6.1 to denoise the gridded data;

[0036] Gridding the data And the corresponding number of steps required for denoising k Input diffusion model to remove noise. The calculation process of denoising is as follows: first calculate the number of iterations required step value, whose value is k , k - 1, …,1, in the grid data according to different iteration times step The values ​​are carried out in sequence step , step - 1, …,1,0 times of reverse process to obtain the denoised data.

[0037] The beneficial effects of adopting the above technical solution are: a method for gridding and removing noise from geophysical potential field data, which uses Euclidean distance to redefine and generate scattered data, can maximize the use of known data and retain the characteristics of the data to be processed; the use of a diffusion model can achieve accurate reconstruction of missing data; the use of a noise level network can achieve a better denoising effect. It has been verified that the MSE value of the method of the present invention after gridding and denoising of geophysical potential field data generated by a theoretical model is as low as 0.1457. It can solve the problem of low accuracy of noisy potential field data after gridding. Compared with other methods, it has the advantages of high accuracy and high robustness, and is suitable for gridding and denoising of measured potential field data. The method of the present invention is of great significance for improving the processing and interpretation accuracy of geophysical potential field data. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A flow chart of a method for gridding and removing noise from geophysical potential field data provided by an embodiment of the present invention;

[0039] Figure 2 Theoretical model position and gravity gradient provided for embodiments of the present invention V zz Original data, where (a) is the theoretical model position and (b) is the gravity gradient V zz Data images;

[0040] Figure 3 The invention provides a method for generating scattered point data and a final gridded image for the geophysical potential field data to be processed, wherein (a) is the calculated scattered point data, and (b) is the gridded image. V zz Data images;

[0041] Figure 4 The measured scattered point data of the Vinton salt dome area provided by the embodiment of the present invention, wherein (a) represents the gravity gradient data V xx , (b) shows the gravity gradient data V xy , (c) represents the gravity gradient data V xz , (d) represents gravity gradient data V yy , (e) represents gravity gradient data V yz , (f) is the gravity anomaly data V z , (g) is the gravity gradient data V zz ;

[0042] Figure 5The measured data of the Vinton salt dome area after denoising provided by the embodiment of the present invention, wherein (a) is the gravity gradient data after denoising V xx , (b) is the denoised gravity gradient data V xy , (c) is the denoised gravity gradient data V xz , (d) is the denoised gravity gradient data V yy , (e) is the denoised gravity gradient data V yz , (f) is the denoised gravity anomaly data V z , (g) is the gravity gradient data V zz . DETAILED DESCRIPTION

[0043] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0044] In this embodiment, a method for gridding and removing noise from geophysical potential field data is provided. Figure 1 As shown, the following steps are included:

[0045] Step 1: Generate geophysical potential field data using the random model forward modeling method, and add noise to generate a geophysical potential field noisy data set;

[0046] Step 1.1: Generate the original geophysical potential field dataset using the stochastic model forward modeling method;

[0047] The random model forward modeling method is used to randomly generate rectangular models of different scales, shapes, positions, densities and magnetic susceptibilities in the determined survey area space, and the potential field anomaly data of the model is calculated to form the original geophysical potential field data set;

[0048] The potential field anomaly data includes gravity anomaly data, gravity gradient data, magnetic anomaly and magnetic gradient type data; the original geophysical potential field data set is composed of various types of potential field anomaly data, wherein each set of data represents an anomaly generated by a certain random model;

[0049] In this embodiment, the forward modeling used is calculated in the Cartesian coordinate system, and the forward modeling method used is the random distribution cuboid forward modeling, that is, the cuboids with random positions, random sizes, random residual densities and random magnetization intensity are generated and distributed in the survey area. The distribution ranges of the generated data on the x, y, and z axes are 0m to 6400m, 0m to 6400m, and 0m to 1000m, respectively. Among them, the distance between the measuring point and the survey line is 50m. The gravity anomaly data are calculated for the established cuboid theoretical model V z , Gravity gradient data V xx , V xy , V xz , V yy , V yz , V zz and the total magnetic field anomaly Δ T The data set used contains 10 5 Group data.

[0050] Step 1.2: Extract a certain proportion of geophysical potential field data and add different levels of noise to generate a noisy geophysical potential field data set;

[0051] In terms of the number of extracted data, the proportion of the extracted data should be at least 1% of the original data set, and the number of each type of data should be the same, that is, the number of different component data and different abnormal data used is consistent. The noise added is Gaussian noise, and the formula for adding noise is as follows:

[0052] (1)

[0053] Among them, X n is the noise-added data matrix, X is the original data matrix, X max is the maximum value of the original data matrix, It means that it obeys the normal distribution with mathematical expectation of 0 and variance of 1. Noiselevel Indicates the noise level as a percentage. Noiselevel It refers to the percentage of the maximum value of different original data matrices;

[0054] In this embodiment, in order to cover as many noise types as possible, 1000 data of each type are extracted and Gaussian noise of 1%, 3%, 5%, and 7% of the maximum value of the corresponding data is added respectively.

[0055] Step 2: Use the original geophysical potential field dataset to train the diffusion model until a diffusion model that meets the requirements is obtained;

[0056] Step 2.1: Normalize the original geophysical potential field data set. The data normalization formula is as follows:

[0057] (2)

[0058] in, x max and x min Respectively represent the maximum and minimum values ​​of the data to be normalized, x represents the value before normalization, x scale Represents the normalized value;

[0059] Step 2.2: Use the normalized original geophysical potential field dataset to train the diffusion model;

[0060] The calculation process of the diffusion model includes two parts: the forward process and the reverse process. The forward process formula is as follows:

[0061] (3)

[0062] in, Indicated in Under the conditions The probability distribution of is the unit matrix, t is the number of steps, x t Indicates that the input data passes through t The value after adding noise, is the variance used at each step, T is the total number of training steps, in the forward process β t With the number of steps t Positive correlation, Indicates that the input is x t The data obeys the mathematical expectation , the variance is The formula for the reverse process is:

[0063] (4)

[0064] in, Indicated in Under the conditions The probability distribution of and The parameters obtained from network training (training times ≥ 50,000 times), is the mathematical expectation of the Gaussian model distribution, is the variance of the Gaussian model distribution, Represents the input data The mathematical expectation is The variance is Normal distribution of

[0065] In this embodiment, in order to ensure that the diffusion model used is sufficiently trained, 60,000 trainings are performed on the normalized original geophysical potential field dataset.

[0066] Step 3: Calculate the number of denoising steps corresponding to each set of noisy data, and train the noise level selection network together with the noisy data set to obtain a noise level selection network model that meets the requirements;

[0067] Step 3.1: Calculate the mean square error (MSE) between the denoised data and the original geophysical potential field data without noise. The formula is:

[0068] (5)

[0069] in, m represents the amount of noisy data, x i represents the original data without noise, g i represents the preliminary denoising result obtained by the reverse process of the diffusion model;

[0070] Step 3.2: Using the mean square error (MSE) value between the denoised data in step 3.1 and the original geophysical potential field data without noise, and using the diffusion model trained in step 2, obtain the number of denoising steps corresponding to each group of noisy data;

[0071] Let the number of iterations be k , k = 0, 1, 2, …, k max , k max Indicates the maximum number of iterations, calculates and records the noisy data k The mean square error (MSE) value at the iteration. The MSE value decreases as the number of iterations increases. If the MSE value increases, the number of reverse calculations is the current number of iterations. k Minus 1, that is, the number of denoising steps is k - 1;

[0072] Step 3.3: Input the number of denoising steps obtained in step 3.2 and the corresponding noisy data into the noise level selection network for training;

[0073] The noisy data set is normalized and the classification network model is used as the noise level selection network; the classification network model uses cross entropy as the loss function, and its formula is as follows:

[0074] (6)

[0075] in, is the actual target category, To predict the target category, select the model with the minimum cross entropy and a cross entropy less than 0.1;

[0076] In this embodiment, the noise level selection network used is ResNet 18, which is trained at least 200 times, and the minimum cross entropy value of the final model is 0.088.

[0077] Step 4: According to the coordinates and outliers of the geophysical potential field data to be processed, the location of the scattered data is determined, and the anomalies at the corresponding locations are assigned values ​​to generate scattered data;

[0078] Step 4.1: Determine the measuring points and measuring line distances of the scattered data according to the geophysical potential field data to be processed, and determine the spatial distribution range of the scattered data;

[0079] The data to be processed include the measured abnormal values ​​and their corresponding coordinates. In order to ensure the accuracy of data reconstruction, it is necessary to determine the spatial distribution range of the scattered data, so as to determine the range of the data after gridding, that is, the range of the data to be processed should be included in the measurement area of ​​the scattered data, and the distance between the measurement points and measurement lines of the scattered data should be determined.

[0080] In this embodiment, when the theoretical model test is adopted, the theoretical model position used is as follows: Figure 2 (a) includes two equal-sized and same residual density (1 g / cm 3 ) of the cuboid, the distribution range of the cuboid positions on the x-axis is 1325m~2825m, 3575m~5075m, the distribution range on the y-axis is 1200m~4600m, and the distribution range on the z-axis is 400m~500m. The corresponding model V zz Data image such as Figure 2 As shown in (b), it is drawn when the spatial distribution ranges of the x, y and z axes are 0 to 6400m, 0 to 6400m and 0 to 1000m, and the measuring points and measuring lines are both determined to be 50m.

[0081] Step 4.2: Calculate the Euclidean distance between the position of the geophysical potential field data point to be processed and the position of the scattered data point, and assign a value to the designated scattered point position through the Euclidean distance;

[0082] The Euclidean distance calculation formula between the position of the data point to be processed and the position of the scattered data point is:

[0083] (7)

[0084] in, Dis the Euclidean distance between the geophysical potential field data point to be processed and the scattered data point, d 1, l 1 is the coordinate of the data point to be processed, d 2, l 2 is the coordinate of the scattered data point;

[0085] Traversing the coordinates of the geophysical potential field data points to be processed, assigning the data to be processed to the corresponding position in the scattered data according to the corresponding minimum Euclidean distance, and obtaining the scattered data, which includes the data coordinates and the corresponding potential field data;

[0086] In this embodiment, during the theoretical model test, the scattered data obtained after calculation are as follows: Figure 3 (a) shown.

[0087] Step 5: Input the scattered data into the diffusion model and grid the scattered data;

[0088] The trained diffusion model is used to calculate the reverse process of the scattered data to obtain the missing data in the grid data. , which obeys the mathematical expectation The variance is The normal distribution of is expressed as follows:

[0089] (8)

[0090] in, and are the Gaussian model distribution parameters of the diffusion model in step 2, is the mathematical expectation of the Gaussian model distribution, is the variance of the Gaussian model distribution;

[0091] Finally, the gridded data of the geophysical potential field data to be processed is obtained The formula is expressed as:

[0092] (9)

[0093] in, For scattered data, is the gridded data;

[0094] In this embodiment, during the theoretical model test, after the scattered point data in step 4.2 is input into the diffusion model, the following can be obtained: Figure 3 (a) Gridded data with noise.

[0095] Step 6: Input the gridded data into the noise level selection network, calculate the number of steps required for denoising, and combine the diffusion model for denoising;

[0096] Step 6.1: Calculate the number of denoising steps through the noise level selection network;

[0097] The gridded data generated in step 5 Input the noise level selection network, classify the noise contained in the gridded data through the noise level selection network, and calculate the number of steps required for denoising k ;

[0098] Step 6.2: Use the diffusion model combined with the number of denoising steps in step 6.1 to denoise the gridded data;

[0099] Gridding the data And the corresponding number of steps required for denoising k Input diffusion model to remove noise. The calculation process of denoising is as follows: first calculate the number of iterations required step value, whose value is k , k - 1, …,1, in the grid data according to different iteration times step The values ​​are carried out in sequence step , step - 1, …, 1, 0 times of reverse process to obtain the denoised data; this process is calculated by formula (4) and the reverse process is similar to formula (8), represents the denoised data, and at the same time, It also satisfies the mathematical expectation The variance is The normal distribution is:

[0100] (10)

[0101] In this embodiment, in the theoretical model test, the final result of using the diffusion model to denoise the gridded data is as follows: Figure 3 (b) as shown.

[0102] In order to verify the practicality of the method of the present invention, this embodiment also uses the measured geophysical potential field data of the Vinton salt dome area in the United States for testing. The results are as follows: Figure 4 , Figure 5 As shown, Figure 4 is the calculated scattered data, Figure 5The measured geophysical potential field data after final gridding and denoising, the full tensor airborne gravity gradient data used here ranges from x: 438.504km-450.104km, y: 3335.584km-3345.18km, and the data scale is 128×128. The ground gravity anomaly range used is x: 438km-450km, y: 3330km-3342.565km, and the data scale is 80×80. The method proposed in the present invention realizes the gridding and denoising of the data in the Vinton salt dome area, and the final grid data obtained has high accuracy, and the noise is suppressed, which is beneficial to carry out high-precision data processing and interpretation research.

[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.

Claims

1. A method for gridding and removing noise from geophysical potential field data, characterized in that: The following steps are involved: Step 1: Generate geophysical potential field data using the random model forward modeling method, and add noise to generate a geophysical potential field noisy data set; Step 2: Use the original geophysical potential field dataset to train the diffusion model until a diffusion model that meets the requirements is obtained; Step 3: Calculate the number of denoising steps corresponding to each set of noisy data, and train the noise level selection network together with the noisy data set to obtain a noise level selection network model that meets the requirements; Step 4: According to the coordinates and outliers of the geophysical potential field data to be processed, the location of the scattered data is determined, and the anomalies at the corresponding locations are assigned values ​​to generate scattered data; Step 5: Input the scattered data into the diffusion model and grid the scattered data; The trained diffusion model is used to calculate the reverse process of the scattered data to obtain the missing data in the grid data. Its mathematical expectation is μ θ (x t , t) variance is ∑ θ (x t , t), the normal distribution formula is as follows: Among them, μ θ (x t , t) and ∑ θ (x t , t) are Gaussian model distribution parameters of the diffusion model, μ θ (x t , t) is the mathematical expectation of the Gaussian model distribution, ∑ θ (x t , t) is the variance of the Gaussian model distribution; Finally, the gridded data of the geophysical potential field data to be processed is obtained The formula is expressed as: in, For scattered data, is the gridded data; Step 6: Input the gridded data into the noise level selection network, calculate the number of steps required for denoising, and combine the diffusion model for denoising; Step 6.1: Calculate the number of denoising steps through the noise level selection network; The gridded data generated in step 5 Input the noise level selection network, classify the noise contained in the gridded data through the noise level selection network, and calculate the number of steps k required for denoising; Step 6.2: Use the diffusion model combined with the number of denoising steps in step 6.1 to denoise the gridded data; Gridding the data The number of steps k required for denoising is input into the diffusion model for noise removal. The denoising calculation process is as follows: first calculate the number of iterations step value required to traverse, whose value is k, k-1,...,1, and then perform the reverse process of step, step-1,...,1,0 on the gridded data according to different iterations step values ​​to obtain the denoised data.

2. A method for gridding and removing noise from geophysical potential field data according to claim 1, characterized in that: The specific method of step 1 is: Step 1.1: Generate the original geophysical potential field dataset using the stochastic model forward modeling method; The random model forward modeling method is used to randomly generate rectangular models of different scales, shapes, positions, densities and magnetic susceptibilities in the determined survey area space, and the potential field anomaly data of the model is calculated to form the original geophysical potential field data set; Step 1.2: Extract a certain proportion of geophysical potential field data and add different levels of noise to generate a geophysical potential field noisy data set; The proportion of the extracted data should be at least 1% of the original data set, and the number of each type of data should be the same, that is, the number of different component data and different abnormal data used should be consistent, and the added noise should be Gaussian noise.

3. A method for gridding and removing noise from geophysical potential field data according to claim 2, characterized in that: The potential field anomaly data includes gravity anomaly data, gravity gradient data, magnetic anomaly and magnetic gradient type data; the original geophysical potential field data set is composed of various types of potential field anomaly data, wherein each set of data represents an anomaly generated by a certain random model.

4. A method for gridding and removing noise from geophysical potential field data according to claim 2, characterized in that: The step 2 uses the normalized original geophysical potential field data set to train the diffusion model; the calculation process of the diffusion model includes two parts: a forward process and a reverse process.

5. A method for gridding and removing noise from geophysical potential field data according to claim 4, characterized in that: The specific method of step 3 is: Step 3.1: Calculate the mean square error (MSE) between the denoised data and the original geophysical potential field data without noise; Step 3.2: By using the mean square error between the denoised data in step 3.1 and the original geophysical potential field data without noise, and using the diffusion model trained in step 2, the number of denoising steps corresponding to each group of noisy data is obtained; Step 3.3: Input the number of denoising steps obtained in step 3.2 and the corresponding noisy data into the noise level selection network for training.

6. A method for gridding and removing noise from geophysical potential field data according to claim 5, characterized in that: The specific method of step 3.2 is: Let the number of iterations be k, k = 0, 1, 2, ..., k max , k max Indicates the maximum number of iterations. Calculate and record the mean square error (MSE) value of the noise data at the kth iteration. The MSE value decreases as the number of iterations increases. If the MSE value increases, the number of reverse process calculations is the current number of iterations (k) minus 1, that is, the number of denoising steps is k-1.

7. A method for gridding and removing noise from geophysical potential field data according to claim 6, characterized in that: The step 3.3 normalizes the noisy data set and uses the classification network model as a noise level selection network; The classification network model uses cross entropy as the loss function.

8. A method for gridding and removing noise from geophysical potential field data according to claim 7, characterized in that: The specific method of step 4 is: Step 4.1: Determine the measuring points and measuring line distances of the scattered data according to the geophysical potential field data to be processed, and determine the spatial distribution range of the scattered data; Step 4.2: Calculate the Euclidean distance between the position of the geophysical potential field data point to be processed and the position of the scattered data point, and assign a value to the designated scattered point position through the Euclidean distance; The coordinates of the geophysical potential field data points to be processed are traversed, and the data to be processed are assigned to their corresponding positions in the scattered data according to their corresponding minimum Euclidean distances to obtain scattered data, which include data coordinates and their corresponding potential field data.

Images