Adaptive focusing inversion method and system based on multi-dimensional evaluation driving

Abstract

The invention belongs to the technical field of gravity exploration, and discloses a self-adaptive focusing inversion method and system based on multi-dimensional evaluation driving, and the method comprises the steps: collecting the surface observation gravity data of a to-be-detected region, and carrying out the inversion of the surface observation gravity data of the to-be-detected region in the inversion process of the surface observation gravity data of the to-be-detected region; in each iteration, smooth inversion and focusing inversion based on minimum support are respectively carried out on the result of the last iteration; a mean square error method, a marginal definition operator method, an overall gradient evaluation operator method and a variance evaluation operator method are used for comprehensively evaluating a smooth inversion result and a focusing inversion result respectively, an optimal iteration path is selected, a focusing interval is adaptively selected, and high-resolution density interface imaging is achieved. According to the method, the focusing interval can be adaptively adjusted, and high-resolution identification of the target abnormal body can be realized.

Description

Technical Field

[0001] The present invention belongs to the field of gravity exploration technology, and in particular relates to an adaptive focusing inversion method and system based on multi-dimensional evaluation drive. Background Art

[0002] Gravity exploration has long played a vital role in basic geological surveys and resource and energy exploration. However, due to the volumetric and stacking effects of the gravity field, achieving high vertical resolution has long been a technical bottleneck. While boundary resolution can be effectively improved by applying appropriate mathematical transformations to the density model of the regularized stabilization functional and constructing a model focusing reweighting matrix, traditional focusing algorithms rely heavily on subjective judgment when setting the upper and lower limits of the focusing interval and whether segmented focusing is necessary, lacking quantitative analysis. Summary of the Invention

[0003] To address the challenges of existing technologies, the present invention provides an adaptive focusing inversion method and system driven by multidimensional evaluation. This innovative approach proposes a minimum support-based adaptive focusing iteration method. This method quantitatively evaluates the inversion iteration results using a combination of four methods: mean square error, edge clarity operator, global gradient evaluation operator, and variance evaluation operator. This method enables adaptive judgment of the focusing method. This method not only adaptively adjusts the focusing interval but also achieves high-resolution identification of target anomalies.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] The adaptive focusing inversion method based on multi-dimensional evaluation drive includes the following steps:

[0006] Collecting surface observation gravity data of the area to be measured. During the inversion process of the surface observation gravity data of the area to be measured, smoothing inversion and focusing inversion based on minimum support are performed on the results of the previous iteration in each iteration.

[0007] Four methods, namely mean square error, edge clarity operator, overall gradient evaluation operator and variance evaluation operator, are used to comprehensively evaluate the smooth inversion results and focused inversion results respectively, select the optimal iterative path, and adaptively select the focusing space to achieve high-resolution density interface imaging.

[0008] Preferably, using the mean square error to comprehensively evaluate the smoothing inversion result and the focused inversion result includes:

[0009] The smooth inversion results and the focused inversion results are compared with the previous inversion results. Calculate the mean square error using the following formula:

[0010] ;

[0011] ;

[0012] in, is the number of model divisions; is the grid label; is the result of the previous iteration; To smooth the inversion result; To focus the inversion results;

[0013] Smooth inversion prediction data Focusing on inversion prediction data Respectively with observation data The data fitting error is calculated using the root mean square error, and the calculation formula is:

[0014] ;

[0015] .

[0016] Preferably, the use of an edge clarity operator to comprehensively evaluate the smooth inversion results and the focused inversion results includes:

[0017] Preset inversion results For one Matrix of

[0018] The inversion results Perform gradient conversion and pre-set the horizontal and vertical templates as and , expressed as:

[0019] ;

[0020] ;

[0021] The weights in the horizontal and vertical directions are predefined as 2, and the weights in other directions are predefined as 1;

[0022] Scheduled inversion results The gradient value of the convolution on the horizontal and vertical template is and , expressed as:

[0023] ;

[0024] The inversion result The gradient magnitude at each point and gradient direction Expressed as:

[0025] ;

[0026] ;

[0027] for The gradient magnitude of each position is compared with the gradient magnitudes of the two adjacent pixels along the gradient direction. If the gradient magnitude of the point is not a local maximum, the point is preset to 0, otherwise it is set to 1, resulting in a dimension of The edge matrix ;

[0028] select Corresponding coordinates Calculate the slopes of the three straight lines for the three points within the preset distance, and finally find the average value, that is:

[0029] ;

[0030] ;

[0031] ;

[0032] in, is the grid center density value; , It is defined as the slope of the corresponding straight line in the horizontal direction; The edge slope value in the horizontal direction calculated after the edge of the abnormal body is determined;

[0033] Sum all edge grayscale values ​​of the entire inversion result:

[0034] ;

[0035] Use the same method to calculate the sum of the slope values ​​of the anomaly edge in the vertical direction :

[0036] ;

[0037]

[0038] ;

[0039]

[0040] in, is the grid center density value; , It is defined as the slope of the corresponding straight line in the vertical direction; The vertical slope value of the edge of the anomaly is calculated after it is determined to be the edge of the anomaly. The average value of the slope value of the anomaly edge is calculated over the entire inversion result. This is the evaluation value of the edge clarity of the abnormal body, and the formula is:

[0041] .

[0042] Preferably, using the overall gradient evaluation operator to comprehensively evaluate the smooth inversion results and the focused inversion results includes:

[0043] Preset inversion mesh division , distance weighting is performed on the eight grids around the central grid, where and Adjacent mesh weights is 1, and Adjacent mesh weights for ;

[0044] For the density value of each grid in the inversion result, subtract the density values ​​of the adjacent 8 domain grids and calculate the weighted sum of the 8 differences. The calculation formula is:

[0045] ;

[0046] in, is the gradient value corresponding to the grid, is the weight of different adjacent grids; is the density value of the central grid, is the density value of the adjacent grid, where and The value of is 1 or 2; then add the values ​​obtained from all grids and divide by the total number of grids to get the overall gradient value E, which is calculated as follows:

[0047] .

[0048] Preferably, using the variance evaluation operator to comprehensively evaluate the smoothing inversion results and the focusing inversion results includes:

[0049] Preset existence threshold The inversion results Divided into background With abnormal body ,So and The average values ​​of and , the relationship that exists is:

[0050] ;

[0051] in, for Divided into The probability of for Divided into The probability of ; for The global mean of

[0052] Inversion results The global variance Expressed as:

[0053] ( - ;

[0054] There must be a value in make Take the maximum value, that is:

[0055] ;

[0056] when , then the grid is located inside the anomaly, where is the density value corresponding to the grid;

[0057] The variance of the density values ​​of all grids inside the anomaly is calculated using the following formula:

[0058] ;

[0059] in, is the total number of grids located inside the anomaly; is the average density difference of all grids in the anomaly.

[0060] Preferably, selecting the optimal iterative path includes:

[0061] The weights of the four evaluation mechanisms, namely, mean square error, edge clarity, overall gradient value, and internal variance of anomalies, are uniformly set to one. If a certain evaluation mechanism considers that the result of focused inversion is Better than the result of smooth inversion , then output the evaluation value , otherwise output ; Accumulate the evaluation values:

[0062] ;

[0063] in, is the evaluation value output by the four evaluation mechanisms; then when When , the result of the focused inversion is considered to be The result of the smooth inversion , then the output value of this iteration is ;when When , the result of smooth inversion is considered to be Results of optimal focusing inversion , then the output value of this iteration is ;when When , the fitting error of the output data is A small one.

[0064] The present invention also discloses an adaptive focusing inversion system based on multi-dimensional evaluation drive, wherein the system is used in any one of the methods described above, comprising: an inversion module and a multi-dimensional evaluation module;

[0065] The inversion module is used to collect gravity data of the area to be measured. In the process of inverting the surface observation gravity data of the area to be measured, smooth inversion and focusing inversion based on minimum support are performed on the results of the previous iteration in each iteration;

[0066] The multidimensional evaluation module is used to use four methods, namely, mean square error, edge clarity operator, overall gradient evaluation operator, and variance evaluation operator, to comprehensively evaluate the smooth inversion results and the focused inversion results, select the optimal iterative path, and adaptively select the focusing space to achieve high-resolution density interface imaging.

[0067] Compared with the prior art, the present invention has the following beneficial effects:

[0068] This paper innovatively proposes an adaptive focusing iteration method based on minimum support. This method quantitatively evaluates the inversion iteration results using a combination of four methods: mean square error, edge clarity operator, global gradient evaluation operator, and variance evaluation operator. This method not only enables adaptive adjustment of the focusing interval but also enables high-resolution identification of target anomalies. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0070] Figure 1 Schematic diagram of the process of an adaptive focusing inversion method based on multi-dimensional evaluation drive according to an embodiment of the present invention;

[0071] Figure 2 Schematic diagram of a conventional focusing algorithm according to an embodiment of the present invention;

[0072] Figure 3 Schematic diagram of an adaptive focusing algorithm under the constraints of a comprehensive evaluation mechanism according to an embodiment of the present invention;

[0073] Figure 4 2D Dahongshan anomaly density model and gravity response according to an embodiment of the present invention, wherein (a) is a vertical gravity component fitting diagram; (b) is a 2D Dahongshan anomaly density inversion model. DETAILED DESCRIPTION

[0074] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0075] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0076] First, some technical terms used in this invention are explained:

[0077] Mean Squared Error (MSE) is a metric that measures the difference between predicted and actual values ​​and is used to evaluate the performance of regression models. Specifically, MSE calculates the average of the squares of the differences between all predicted values ​​and the actual observed values.

[0078] Edge sharpness operators generally refer to operators used to detect image edges. These operators can identify places in an image where pixel values ​​suddenly change, i.e., edges. Edge detection is an important step in image processing, which helps segment images, identify object contours, and more. Edge detection operators locate edges by calculating the gradient of an image. The magnitude and direction of the gradient can indicate the speed and direction of brightness changes in the image. Common edge detection operators include: Sobel operator: Calculates the gradient of an image by combining Gaussian smoothing and differential extremes. Prewitt operator: Similar to the Sobel operator, but uses a different convolution kernel to calculate the gradient. Canny operator: A combination of a series of steps, including Gaussian smoothing, gradient calculation, non-maximum suppression, and double thresholding, to detect edges in an image. These operators can be used in different application scenarios, such as computer vision and image analysis.

[0079] The global gradient evaluation operator refers to a vector operator used to calculate the rate of change of a multivariate function in each variable direction. It is mainly used in image processing to detect pixel intensity changes and determine edge positions.

[0080] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0081] Example 1

[0082] like Figure 1 As shown, an embodiment of the present invention provides an adaptive focusing inversion method based on multi-dimensional evaluation drive, comprising the following steps:

[0083] Collect gravity data of the area to be measured. In the process of inverting the surface observation gravity data of the area to be measured, perform smooth inversion and minimum support-based focused inversion on the results of the previous iteration in each iteration.

[0084] Four methods, namely mean square error, edge clarity operator, overall gradient evaluation operator and variance evaluation operator, are used to comprehensively evaluate the smooth inversion results and focused inversion results respectively, select the optimal iterative path, and adaptively select the focusing space to achieve high-resolution density interface imaging.

[0085] In this embodiment, to further improve the stability of the inversion, reduce its multiplicity, and increase the imaging resolution, Zhdanov (2002) proposed and constructed a minimum support stable focusing functional within the framework of regularized inversion (Tikhonov and Arsenin, 1977). The specific formula is as follows: (1)

[0086] in, is the objective function, Fitting functionals to the data domain; is the regularized model domain stable focusing functional, where the coefficient is the regularization factor.

[0087] The data domain fitting functional in formula (1) It can also be written as follows:

[0088] (2)

[0089] in, In order to seek norm; is the gravity forward operator; is the density model; For observational gravity data. Model domain stable focusing functional It can be written as:

[0090] (3)

[0091] in, is the prior model; It is a fusion of traditional sensitivity-based depth weighting and minimum support focused weighted The reweighted matrix of .in It is based on the diagonal focusing matrix, which is expressed as follows:

[0092] (4)

[0093] in is the focusing factor, which is a decimal not equal to 0.

[0094] In this embodiment, in the traditional focusing algorithm, when selecting the focus segment, a smooth inversion is usually performed, and then the focus starting point is determined based on the result of the smooth inversion, and the focus inversion end point is selected based on personal experience. If the inversion imaging effect is not good, it is necessary to continuously adjust the focus interval. The specific process is as follows Figure 2 As shown. Therefore, the traditional focusing algorithm has the problems of strong subjectivity and low efficiency. In view of this, the present invention proposes an adaptive focusing algorithm under the constraints of a comprehensive evaluation mechanism. The algorithm collects gravity data of the area to be measured. In the process of inverting the surface observation gravity data of the area to be measured, the smooth inversion and the focusing inversion based on the minimum support are performed on the results of the previous iteration in each iteration; the conventional smooth inversion and the focusing inversion based on the minimum support are performed respectively, and the results are recorded as and , as shown below:

[0095] ;

[0096] ;

[0097] in, is the result of the previous iteration; is the optimal step size for smooth inversion; is the fastest rising direction of the smooth inversion; The optimal step size for focused inversion; is the direction of the steepest rise in the focused inversion;

[0098] The four methods of mean square error, edge clarity operator, overall gradient evaluation operator and variance evaluation operator are used to comprehensively evaluate the smooth inversion results and the focused inversion results, select the optimal iterative path, and adaptively select the focus space to achieve high-resolution density interface imaging. The specific process is as follows: Figure 3 shown.

[0099] In this embodiment, the present invention uses a mean square error evaluation algorithm to evaluate the similarity between the previous iteration result and the current iteration result, so as to understand the progress of the current iteration in the search space and select the best iteration direction, making it more likely that the algorithm will move towards the optimal solution.

[0100] First, the previous inversion results The conventional smooth inversion and the focusing inversion based on minimum support are performed respectively, and the results are recorded as and Then, the two inversion results were compared with the previous inversion results. The mean square error is calculated as follows:

[0101] (5)

[0102] (6)

[0103] in, is the number of model divisions; is the grid label; is the result of the previous iteration; To smooth the inversion result; To focus on the inversion result; the higher the similarity between the current inversion result and the previous iterative inversion result, the smaller the mean square error value.

[0104] Smooth inversion prediction data Focusing on inversion prediction data Respectively with observation data The data fitting error is calculated using the root mean square error, and the calculation formula is as follows:

[0105]

[0106]

[0107] The smaller the data fitting error, the higher the degree of consistency between the predicted data and the actual data.

[0108] According to the above principle analysis, we can know that: and When , the data fitting errors of the two inversion results are reduced, and both inversion results develop in a more positive direction. In order to avoid falling into the local optimum, a more radical iteration direction is selected at this time, so it is believed that the inversion result with a larger mean square error is better. and , which shows that the two inversion results fluctuate in the process of finding the global optimum. In order to ensure the stability of the algorithm, a more conservative iteration direction is selected this time, and the one with smaller mean square error is considered to be better; in other cases, in order to ensure the stability of the algorithm, the one with smaller data fitting error is considered to be better.

[0109] In this embodiment, the clarity of the anomaly boundary is an important aspect of evaluating the inversion imaging effect, so the present invention introduces an edge clarity evaluation algorithm.

[0110] Assume the inversion result For one The inversion results include smooth inversion results and focused inversion results. The edge strength and direction at any position are first converted into Gradient conversion is performed. In order to highlight the changes in the horizontal and vertical directions of the abnormal body obtained in each iteration for the convenience of edge recognition, the horizontal and vertical templates are defined as and , as shown below:

[0111] (9)

[0112] (10)

[0113] In order to reduce the influence of some outliers in the inversion results on the identification of the anomaly boundary, the weights in the horizontal and vertical directions are defined as 2, and the weights in other directions are defined as 1. The gradient value of the convolution on the horizontal and vertical template is and . Its definition is as follows:

[0114] (11)

[0115] The inversion result The gradient magnitude at each point and gradient direction As shown in formula (12) and formula (13):

[0116] )

[0117]

[0118] In order to ensure the accuracy of the edge, non-maximum suppression is required: The gradient magnitude at each position is compared with the gradient magnitudes of the two adjacent pixels along the gradient direction. If the gradient magnitude of the point is not a local maximum, the point is set to 0, otherwise it is set to 1, resulting in a dimension of The edge matrix .

[0119] The part is the inversion boundary, then for The next step can be processed at this time.

[0120] Finally, according to the inversion clear imaging edge density difference change should be abrupt, while the blurred imaging boundary is flat, the present invention proposes a method for judging the clarity of the abnormal body boundary, the density value corresponding to each grid is regarded as a straight line, and the flatness of the straight line is measured by the slope of the straight line, that is, Corresponding coordinates Calculate the slopes of the three lines at the three nearby points and find the average value, which is:

[0121]

[0122]

[0123]

[0124] in, is the grid center density value; , It is defined as the slope of the corresponding straight line in the horizontal direction; is the edge slope value in the horizontal direction calculated after the edge of the abnormal body is determined.

[0125] Then sum up all edge gray values ​​of the entire inversion result:

[0126]

[0127] Use the same method to calculate the sum of the slope values ​​of the anomaly edge in the vertical direction :

[0128]

[0129]

[0130]

[0131]

[0132] in, is the grid center density value; , It is defined as the slope of the corresponding straight line in the vertical direction; The edge slope value in the vertical direction is calculated after the edge of the anomaly is determined. Finally, the edge slope value of the anomaly is averaged over the entire inversion result. This is the evaluation value of the edge clarity of the abnormal body, and the formula is:

[0133] According to the above principle analysis, when the edge gradient of the abnormal body is large, that is, the evaluation value of the edge clarity of the abnormal body The larger it is, the more obvious the boundary of the abnormal body is.

[0134] In this embodiment, to further evaluate the clarity of the inversion imaging, the present invention proposes a global gradient algorithm to assess the overall quality of the inversion imaging. A larger overall gradient value indicates a more prominent anomaly, which in turn indicates a clearer inversion image and a better imaging effect.

[0135] Assume inversion grid division , because the inversion grid is anisotropic, the eight grids around the central grid are distance-weighted, where and Adjacent mesh weights is 1, and Adjacent mesh weights for .

[0136] For the density value of each grid in the inversion result, the density values ​​of the eight adjacent grids are subtracted from it, and the weighted sum of the eight differences is calculated. The calculation formula is as follows:

[0137] , )

[0138] in, is the gradient value corresponding to the grid, is the weight of different adjacent grids; is the density value of the central grid, is the density value of the adjacent grid, where and The value of is 1 or 2; then add the values ​​obtained from all grids and divide by the total number of grids to get the overall gradient value E, which is calculated as follows:

[0139]

[0140] According to the above principle analysis, when the overall gradient of the inversion result is larger, that is, the overall gradient value E is larger, the imaging is clearer and the imaging effect is better.

[0141] In this embodiment, the clarity of the anomaly imaging is of great significance for guiding mining, so the present invention proposes a variance evaluation algorithm to evaluate the clarity of the anomaly imaging of the inversion iteration result.

[0142] Assume there is a threshold The inversion results Divided into background With abnormal body ,So and The average values ​​of and The following relationship exists

[0143]

[0144] in, for Divided into The probability of for Divided into The probability of ; for The global mean of .

[0145] Then the inversion result The global variance It can be expressed as

[0146] ( -

[0147] The clearer the inversion imaging result, The larger the overall gradient value, that is, the global variance The bigger. Then There must be a value in make Take the maximum value, that is

[0148]

[0149] when , If it is the density value corresponding to the grid, it is considered that the grid is located inside the abnormal body and can be processed in the next step.

[0150] The density values ​​of all grids inside the abnormal body of the present invention are calculated for variance, and the calculation formula is as follows;

[0151]

[0152] in, is the total number of grids located inside the anomaly; is the average density difference of all grids in the anomaly.

[0153] Then when the variance The larger the value, the clearer the inversion result is in depicting the abnormal body and the better the imaging effect.

[0154] In this embodiment, in conventional imaging processes based on minimum support techniques, the number of continuously focused segments is adjusted entirely by personal experience. To improve this, the present invention proposes a comprehensive inversion result evaluation mechanism that utilizes mean square error, edge clarity evaluation value, global gradient algorithm, and variance to constrain the minimum support technique.

[0155] The specific process of the algorithm is as follows: Conventional smooth inversion and focused inversion based on minimum support are performed respectively, and the results are recorded as and . And and The mean square error, edge clarity, overall gradient value, and internal variance of the anomaly are calculated. The present invention sets the weights of the four evaluation mechanisms to one. If a certain evaluation mechanism considers the result of the focus inversion to be Optimal smoothing inversion algorithm , then output the evaluation value , otherwise output Then add up the evaluation values:

[0156]

[0157] in, is the evaluation value output by the four evaluation mechanisms. When , the result of the focused inversion is considered to be The result of the smooth inversion , then the output value of this iteration is ;when When , the result of smooth inversion is considered to be Results of optimal focusing inversion , then the output value of this iteration is ;when In order to make the data fitting error converge faster and increase the stability of the algorithm, the output A smaller one.

[0158] This paper innovatively proposes an adaptive focusing iteration algorithm based on minimum support. This algorithm quantitatively evaluates the inversion iteration results using a combination of four methods: mean square error, edge clarity operator, global gradient evaluation operator, and variance evaluation operator. This method achieves adaptive judgment of the focusing algorithm. Simulation experiments demonstrate that this algorithm not only adaptively adjusts the focusing interval but also achieves high-resolution identification of target anomalies. Test results using actual data from Dahongshan demonstrate that this algorithm can accurately characterize the Dahongshan anomaly.

[0159] In order to verify the effectiveness of the new adaptive focusing inversion algorithm based on mean square error constraint proposed in this paper, the present invention applies the algorithm to the measured data. First, the space is divided into 100 × 100 = 10,000 rectangular units of equal size and closely arranged. The physical properties of the initial model are set to 0g / cm³, and the focusing factor is set to 0. The regularization factor reduction coefficient is set to 10^-9, and the inversion iteration convergence termination condition is that the data fitting error is 0.1%.

[0160] The inversion imaging results are given by Figure 4 As can be seen, the inversion results of the present invention have a high recovery ability for the anomaly at a depth of 2500 meters to 6000 meters. The anomaly is roughly "U"-shaped, and the boundary and horizontal resolution are relatively high. The residual density obtained by inversion is about 0.1 to 0.22 g / cm 3 , the maximum residual density is 0.23g / cm 3 , the density amplitude is well restored, showing the overall morphology and spatial distribution of the intrusive rock mass.

[0161] Example 2

[0162] The present invention also discloses an adaptive focusing inversion system based on multi-dimensional evaluation drive, wherein the system is used in any one of the methods described above, comprising: an inversion module and a multi-dimensional evaluation module;

[0163] The inversion module is used to collect surface observation gravity data of the area to be measured. During the inversion process of the surface observation gravity data of the area to be measured, smooth inversion and minimum support-based focused inversion are performed on the results of the previous iteration in each iteration.

[0164] The multidimensional evaluation module is used to use four methods, namely mean square error, edge clarity operator, overall gradient evaluation operator, and variance evaluation operator, to comprehensively evaluate the smooth inversion results and focused inversion results, select the optimal iterative path, and adaptively select the focusing interval to achieve high-resolution density interface imaging.

[0165] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. An adaptive focusing inversion method driven by multi-dimensional evaluation, characterized in that: The following steps are involved: Collecting surface observation gravity data of the area to be measured. During the inversion process of the surface observation gravity data of the area to be measured, smoothing inversion and focusing inversion based on minimum support are performed on the results of the previous iteration in each iteration. Four methods, namely mean square error, edge clarity operator, overall gradient evaluation operator and variance evaluation operator, are used to comprehensively evaluate the smooth inversion results and focused inversion results respectively, select the optimal iterative path, and adaptively select the focusing interval to achieve high-resolution density interface imaging.

2. The adaptive focusing inversion method based on multi-dimensional evaluation drive according to claim 1, characterized in that: The comprehensive evaluation of the smooth inversion results and the focused inversion results using mean square error includes: The smooth inversion results and the focused inversion results are compared with the previous inversion results. Calculate the mean square error using the following formula: ； ； in, is the number of model divisions; is the grid label; is the result of the previous iteration; To smooth the inversion result; To focus the inversion results; Smooth inversion prediction data Focusing on inversion prediction data Respectively with observation data The data fitting error is calculated using the root mean square error, and the calculation formula is: ； 。 3. The adaptive focusing inversion method based on multi-dimensional evaluation drive according to claim 1, characterized in that: The edge clarity operator is used to comprehensively evaluate the smooth inversion results and the focused inversion results, including: Preset inversion results For one Matrix of The inversion results Perform gradient conversion and pre-set the horizontal and vertical templates as and , expressed as: ； ； The weights in the horizontal and vertical directions are predefined as 2, and the weights in other directions are predefined as 1; Scheduled inversion results The gradient value of the convolution on the horizontal and vertical template is and , expressed as: ； The inversion result The gradient magnitude at each point and gradient direction Expressed as: ； ； for The gradient magnitude of each position is compared with the gradient magnitudes of the two adjacent pixels along the gradient direction. If the gradient magnitude of the point is not a local maximum, the point is preset to 0, otherwise it is set to 1, resulting in a dimension of The edge matrix ; select Corresponding coordinates Calculate the slopes of the three straight lines for the three points within the preset distance, and finally find the average value, that is: ； ； ； in, is the grid center density value; , It is defined as the slope of the corresponding straight line in the horizontal direction; The edge slope value in the horizontal direction calculated after the edge of the abnormal body is determined; Sum all edge grayscale values ​​of the entire inversion result: ； Use the same method to calculate the sum of the slope values ​​of the anomaly edge in the vertical direction : ； ； in, is the grid center density value; , It is defined as the slope of the corresponding straight line in the vertical direction; The edge slope value in the vertical direction calculated after the anomaly edge is determined to be the edge of the anomaly is averaged over the entire inversion result. This is the evaluation value of the edge clarity of the abnormal body, and the formula is: 。 4. The adaptive focusing inversion method based on multi-dimensional evaluation drive according to claim 3 is characterized in that: The overall gradient evaluation operator is used to comprehensively evaluate the smooth inversion results and the focused inversion results, including: Preset inversion mesh division , distance weighting is performed on the eight grids around the central grid, where and Adjacent mesh weights is 1, and Adjacent mesh weights for ; For the density value of each grid in the inversion result, subtract the density values ​​of the adjacent 8 domain grids and calculate the weighted sum of the 8 differences. The calculation formula is: ； in, is the gradient value corresponding to the grid, is the weight of different adjacent grids; is the density value of the central grid, is the density value of the adjacent grid, where and The value of is 1 or 2; then add the values ​​obtained from all grids and divide by the total number of grids to get the overall gradient value E, which is calculated as follows: 。 5. The adaptive focusing inversion method based on multi-dimensional evaluation drive according to claim 4, characterized in that: The use of variance evaluation operators to comprehensively evaluate the smoothing inversion results and the focused inversion results includes: Preset existence threshold The inversion results Divided into background With abnormal body ,So and The average values ​​of and , the relationship that exists is: ； in, for Divided into The probability of for Divided into The probability of ; for The global mean of Inversion results The global variance Expressed as: （ - ； There must be a value in make Take the maximum value, that is ； when , then the grid is located inside the anomaly, where is the density value corresponding to the grid; The variance of the density values ​​of all grids inside the anomaly is calculated using the following formula: ； in, is the total number of grids located inside the anomaly; is the average density difference of all grids in the anomaly.

6. The adaptive focusing inversion method based on multi-dimensional evaluation drive according to claim 4, characterized in that: Selecting the optimal iterative path includes: The weights of the four evaluation mechanisms, namely, mean square error, edge clarity, overall gradient value, and internal variance of anomalies, are uniformly set to one. If a certain evaluation mechanism considers that the result of focused inversion is Better than the result of smooth inversion , then output the evaluation value , otherwise output ; Accumulate the evaluation values: ； in, is the evaluation value output by the four evaluation mechanisms; then when When , the result of the focused inversion is considered to be The result of the smooth inversion , then the output value of this iteration is ;when When , the result of smooth inversion is considered to be Results of optimal focusing inversion , then the output value of this iteration is ;when When , the fitting error of the output data is A small one.

7. An adaptive focusing inversion system driven by multi-dimensional evaluation, the system being used to implement the method according to any one of claims 1 to 6, characterized in that: include: Inversion module and multi-dimensional evaluation module; The inversion module is used to collect gravity data of the area to be measured. In the process of inverting the surface observation gravity data of the area to be measured, smooth inversion and minimum support-based focused inversion are performed on the results of the previous iteration in each iteration. The multidimensional evaluation module is used to use four methods, namely, mean square error, edge clarity operator, overall gradient evaluation operator, and variance evaluation operator, to comprehensively evaluate the smooth inversion results and the focused inversion results, select the optimal iterative path, and adaptively select the focusing interval to achieve high-resolution density interface imaging.

Images