Mineral identification method based on full-spectrum hyperspectral remote sensing data

Abstract

The invention discloses a mineral identification method based on full-spectrum hyperspectral remote sensing data, which belongs to the technical field of spectrum detection and specifically comprises the following steps: acquiring full-spectrum hyperspectral remote sensing data of a to-be-detected area; dividing the target area into a plurality of grids, and generating a hyperspectral curve according to the reflectivity of a ground object; any grid is used as an undetermined grid, the spectrum difference degree of the undetermined grid and the adjacent grids is calculated, and all the adjacent grids are traversed when the difference degree is lower than a threshold value; obtaining an undetermined grid set, and calculating a mean value to determine a central grid; calculating the product of the spectral difference degree between the central grids and the Euclidean distance, screening the closest central grid, marking a corresponding set as a target area, setting a new grid, dividing the target area again, calculating the spectral difference degree with a peripheral grid by taking a new undetermined grid as a center, marking the spectral difference degree with the peripheral grid being lower than a threshold value as a new target area, and repeating until the nth grid, obtaining a final target area; and extracting spectrum and texture characteristic parameters, comparing with a mineral spectrum curve, and outputting a mineral identification result.

Description

Technical Field

[0001] The present invention relates to the technical field of spectral detection, and in particular to a mineral identification method based on full-spectrum hyperspectral remote sensing data. Background Art

[0002] With the continuous advancement of geoscience research and resource exploration, the demand for accurate mineral identification and classification is increasing. Traditional mineral identification methods rely primarily on field sampling and laboratory analysis. While these methods offer high accuracy, they are complex, time-consuming, costly, and have limited coverage. For example, in large-scale geological surveys, field sampling requires significant manpower and material resources, and sampling is often difficult in hard-to-reach areas. The emergence of hyperspectral remote sensing technology offers a new and promising approach to mineral identification.

[0003] Hyperspectral remote sensing can capture continuous spectral information across nearly the entire visible light spectrum, extending to near-infrared, shortwave infrared, and even thermal infrared. Its extremely narrow band width allows for detailed spectral characterization of ground objects. Different minerals, due to differences in chemical composition and crystal structure, exhibit unique spectral reflectance and absorption characteristics, providing a theoretical basis for mineral identification based on hyperspectral remote sensing data. However, full-spectrum hyperspectral remote sensing data is characterized by large data volumes, high dimensionality, and high data redundancy, which pose significant challenges to mineral identification.

[0004] Existing hyperspectral remote sensing data processing and mineral identification methods directly use raw full-spectrum hyperspectral data for mineral identification, which is computationally intensive and inefficient. Therefore, how to effectively process full-spectrum hyperspectral remote sensing data, extract key information, reduce data dimensionality, and accurately identify minerals using appropriate algorithms has become a pressing technical challenge in this field. Summary of the Invention

[0005] The purpose of the present invention is to provide a mineral identification method based on full-spectrum hyperspectral remote sensing data to solve the following technical problems: Directly using raw full-spectrum hyperspectral data for mineral identification is computationally intensive and inefficient.

[0006] The purpose of the present invention can be achieved through the following technical solutions: A mineral identification method based on full-spectrum hyperspectral remote sensing data includes the following steps: S1. Obtain full-spectrum hyperspectral remote sensing data of the area to be measured, preprocess the remote sensing data, and reduce the dimension of the remote sensing data based on principal component analysis; S2. Divide the target area into several grids W i , i∈[1,n], n is a natural number, grid Wi The side length is L i , L i For a preset length, a hyperspectral curve is generated according to the change of the ground object reflectance in each grid area in the target spectral range; S3. Mark any grid as a pending grid, calculate the spectral difference S between the pending grid and adjacent grids, merge adjacent grids whose spectral difference S is lower than a preset threshold into the pending grid, and calculate the spectral difference for all adjacent grids in sequence; S4. Obtain several pending grid sets whose final number of grids is greater than the set number. For any pending grid set, calculate the mean of the hyperspectral curves of all grids in the pending grid set, and use the grid closest to the mean as the center grid of the pending grid set. S5. Calculate the product of the spectral difference and the Euclidean distance between each central grid and the other central grids, select the closest central grids based on the product data, and mark the corresponding undetermined grid set as the target area; S6. Set grid W i+1 , the side length of the grid is L i+1 , L i+1 =mL i , m∈(0,1), divide the target area into several grids W i+1 , will be combined with the center grid W i The inner grid W i+1 As the undetermined grid W i+1 , with the grid to be determined W i+1 Calculate the center and all the peripheral grids W in turn i+1 The spectral difference between the two is smaller than the preset threshold, and the outer grid W i+1 With the undetermined grid W i+1 Mark them together as the new target area; repeat the grid division until the grid W is reached n , get the grid W n Corresponding to the final target area, the spectral characteristic parameters and texture characteristic parameters of the remote sensing data of the final target area are extracted, and compared with the mineral spectral curve in the spectral library to output the mineral identification results.

[0007] As a further solution of the present invention: in S1, the pretreatment includes: Band removal, radiation calibration and atmospheric correction are performed on the full-spectrum hyperspectral remote sensing data. The band removal is to check the radiation DN value of each band and remove the bands with DN value of 0 or signal-to-noise ratio below the standard; the radiation calibration is to convert the radiation DN value output by the sensor into surface radiation brightness value through the radiation calibration coefficient; the atmospheric correction is to correct the data based on the MODTRAN correction model through statistical correction parameters.

[0008] As a further solution of the present invention: in S3, the calculation of the spectral difference S is specifically as follows: ; Where A and B represent the interval values ​​of the target spectral interval, x represents the spectral band, f(x) represents the hyperspectral curve of the undetermined grid, and g(x) represents the hyperspectral curve of the adjacent grid.

[0009] As a further solution of the present invention: in S4, when the number of grids closest to the mean is greater than one, the grid at which the geometric center of the grid closest to the mean is located is marked as the center grid.

[0010] As a further solution of the present invention: In S5, the specific process of selecting the closest plurality of center grids according to the product data is as follows: For any central grid, obtain the product data of the central grid and other central grids, and sort the product data from small to large, establish a rectangular coordinate system, use the serial number as the horizontal coordinate and the product value as the vertical coordinate, draw a scatter plot, and connect the scatter plots in sequence, mark the obtained curve as curve q(t), t represents the serial number, and connect the first coordinate point and the last coordinate point, mark the obtained straight line as h(t), calculate the integral difference of h(t)-q(t), mark the one with the largest difference as the target sequence, and perform the target sequence; in the scatter plot where the target sequence is located, connect the first coordinate point as the starting point and the other coordinate points as the end points in sequence to obtain several line segments, mark the end point corresponding to the first line segment with a slope greater than the preset slope as a node, and the central grids corresponding to the product data coordinate points before the node are the closest several central grids.

[0011] As a further solution of the present invention: in said S6, the grid is divided repeatedly until the grid W is reached. n , get the grid W n The specific process of the corresponding final target area is: For the i-th target area, use the grid W i+1 Divide it and calculate the undetermined grid W in the central grid i+1 With the outer grid W i+1 The spectral difference between the two, the periphery includes adjacent and non-adjacent, and the peripheral grid W with a difference less than a preset threshold is selected. i+1 With the undetermined grid W i+1 They are collectively marked as the i+1th target area, until i=n, ​​and the final target area is obtained.

[0012] As a further solution of the present invention: in S6, extracting the spectral characteristic parameters and texture characteristic parameters of the remote sensing data of the final target area includes: The spectral feature parameters include spectral curve features, spectral transformation features and spectral measurement features, and the texture feature parameters include gray level co-occurrence matrix, gray level run length, local binary pattern and Gabor filter.

[0013] Beneficial effects of the present invention: This invention removes noise and redundant information from the data through full-spectrum hyperspectral remote sensing data preprocessing and dimensionality reduction, highlighting the spectral characteristics of minerals and laying the foundation for subsequent accurate mineral identification. The target area is divided into multiple levels of grids, and similar grids are automatically merged based on spectral differences, making the determination of the target area more precise, reducing human intervention and errors, and improving the rationality and accuracy of target area determination. During the grid division process, by gradually reducing the grid side length, it can be adaptively adjusted according to the distribution characteristics of the ground objects, better capturing the boundaries and distribution details of minerals, and improving the ability to identify complex geological structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The present invention will be further described below with reference to the accompanying drawings.

[0015] Figure 1 It is a schematic flow diagram of the present invention. DETAILED DESCRIPTION

[0016] 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 any creative efforts shall fall within the scope of protection of the present invention.

[0017] See also Figure 1 As shown, the present invention is a mineral identification method based on full-spectrum hyperspectral remote sensing data, comprising the following steps: S1. Obtain full-spectrum hyperspectral remote sensing data of the area to be measured and preprocess it: Full spectrum hyperspectral remote sensing data acquisition: Hyperspectral sensors carried by aviation or aerospace platforms are used to capture images of the target area. These sensors can acquire electromagnetic wave information reflected by ground objects with extremely high spectral resolution, covering a wide range of wavelengths from visible light to near-infrared, short-wave infrared, and even thermal infrared, thereby obtaining a nearly continuous spectral curve.

[0018] During data collection, it is necessary to ensure the accuracy and stability of sensors to guarantee the quality of the acquired data. At the same time, the temporal and spatial distribution of data collection should be rationally planned according to the scope and characteristics of the study area to improve the representativeness and completeness of the data.

[0019] Remote sensing data preprocessing: Band Removal: Check the radiometric DN value for each band of full-spectrum hyperspectral remote sensing data. The radiometric DN value is the raw digital signal intensity recorded by the sensor, and its magnitude reflects the spectral reflectance characteristics of the ground object. In actual applications, some bands may have a radiometric DN value of 0 due to sensor failure, environmental interference, or the characteristics of the ground object itself.

[0020] Bands with a radiance DN value of 0 contain no valid ground feature information and therefore need to be removed. Furthermore, the signal-to-noise ratio is an important indicator of band quality. Bands with a signal-to-noise ratio below the set standard may have useful information buried in noise, negatively impacting subsequent analysis and processing, and therefore also need to be removed.

[0021] Radiometric calibration: In order to convert the DN value output by the sensor into a surface radiance value with actual physical significance, radiometric calibration is required. Radiometric calibration determines the sensor's response to different radiation intensities by establishing a quantitative relationship between the sensor output and a known radiation source.

[0022] The specific implementation method is to use pre-determined radiometric calibration coefficients to perform a linear or nonlinear transformation on the radiometric DN value output by the sensor. Radiometric calibration coefficients are usually obtained by measuring and calibrating a standard radiation source in a laboratory or in the field, ensuring the accuracy and reliability of the calibration results.

[0023] Atmospheric correction: The purpose of atmospheric correction is to eliminate the influence of the atmosphere on hyperspectral remote sensing data, ensuring that the acquired data truly reflects the spectral characteristics of the ground objects. Gas molecules, aerosols, and other components in the atmosphere scatter, absorb, and refract solar radiation, causing attenuation and distortion of the radiation energy received by the sensor.

[0024] Atmospheric correction is performed based on the MODTRAN correction model. MODTRAN is a widely used model for atmospheric simulation and correction. It accounts for multiple atmospheric components and physical processes, such as gas absorption, aerosol scattering, and molecular scattering. The input parameters of the MODTRAN model are determined through statistical analysis and parameter estimation of meteorological parameters, geographic information, and sensor observation angles in the study area.

[0025] The MODTRAN model calculates atmospheric transmittance and path radiation for different wavelengths, and then uses these parameters to correct the remote sensing data. During the correction process, the radiant energy received by the sensor is divided by the atmospheric transmittance and the path radiation is subtracted to obtain the atmospherically corrected surface radiance value.

[0026] S2. Target area grid division and hyperspectral curve generation: Target area mesh division: According to the research requirements and the actual situation of the target area, the entire target area is divided into several grids W according to certain rules. i , where i ranges from [1,n], and the grid W i The side length is L i , there are n grids of different sizes. i The size and shape of the grid can be adjusted according to the specific application scenario, but in order to facilitate subsequent processing and analysis, a regular square or rectangular grid is usually used.

[0027] Hyperspectral curve generation: For each grid W i , analyzing the changes in the reflectivity of the internal objects in the target spectral range. The target spectral range is pre-set based on the research purpose and the spectral characteristics of the objects, and is usually selected to highlight the differences in the objects.

[0028] By performing statistical analysis on the reflectance of each pixel within the grid, such as calculating the mean and median, we can determine the trend of the average reflectance of the grid across the target spectral range. This trend reflects the spectral characteristics of the main features within the grid and provides an important basis for subsequent mineral identification.

[0029] Based on the obtained average reflectance variation trend, a corresponding hyperspectral curve is generated. The hyperspectral curve is an important tool for describing the spectral characteristics of ground objects. It connects the reflectance values ​​in different bands in the order of wavelength to form a continuous curve.

[0030] S3. Grid marking, spectrum difference calculation and grid merging: Pending grid markers: From the divided grids Wi, each grid is marked as a pending grid according to research requirements or specific analysis purposes. The selection of the pending grid can be determined based on factors such as the distribution characteristics of the ground objects and the significance of the spectral characteristics.

[0031] Spectral difference calculation: For the marked undetermined grid and its adjacent grids, calculate the spectral difference S between them. The specific formula for calculating the spectral difference S is: ; Where A and B represent the target spectral interval [A, B], x represents the spectral interval, f(x) represents the hyperspectral curve of the undetermined grid, and g(x) represents the hyperspectral curve of the adjacent grid. This formula accumulates and quantifies the difference between the two hyperspectral curves over the target spectral interval through integration.

[0032] During the calculation process, the reflectance value at each spectral segment x must be accurately acquired and processed. Hyperspectral data, due to its numerous and continuous distribution, requires an appropriate numerical integration method to solve this integral. Furthermore, to ensure the accuracy and reliability of the calculation results, data preprocessing and quality control are also required.

[0033] Mesh Merge: The calculated spectral difference S is compared with a preset threshold. If the spectral difference S is lower than the preset threshold, the proposed grid is considered to have a high degree of similarity in spectral characteristics with the adjacent grid, and the adjacent grid can be merged into the proposed grid.

[0034] Spectral differences are calculated and compared for all adjacent grids. During the merging process, the hyperspectral curves and related parameters of the pending grids need to be updated to reflect the spectral characteristics of the new merged grid. The merging history and results also need to be recorded for subsequent analysis and application.

[0035] S4. Determination of the central grid of the pending grid set: Get the pending grid collection: After processing in the previous steps, multiple grids are generated. Based on this, analysis and comparison are performed to select several candidate grid sets whose final number of grids exceeds the set number. These candidate grid sets are determined based on factors such as the distribution characteristics of the features and their spectral signatures. The grids within each set have high similarity and correlation.

[0036] In the process of obtaining the candidate grid set, many factors need to be considered, such as the spatial position of the grid, the changing trend of the spectral curve, etc. At the same time, the grids need to be further screened and verified to ensure that the obtained candidate grid set is representative and reliable.

[0037] Calculate the mean of the hyperspectral curve: For any set of undetermined grids, calculate the mean hyperspectral curve of all grids in it. The method of calculating the mean hyperspectral curve is to add the reflectance values ​​of all grids in the set in the same spectral band, and then divide it by the number of grids to obtain the average reflectance value in the spectral band.

[0038] By calculating and analyzing the average reflectance values ​​for each spectral segment, we can obtain the mean hyperspectral curve for the grid set to be determined. This mean curve reflects the overall spectral characteristics of all grids in the set, providing an important basis for the subsequent determination of the central grid.

[0039] Determine the center grid: The grid closest to the mean is used as the center grid of the set of undetermined grids. To determine the grid closest to the mean, the difference between the hyperspectral curve and the mean curve of each grid needs to be calculated. This difference can be calculated using metrics such as Euclidean distance and spectral angle.

[0040] If there are more than one grid cells closest to the mean, the grid with the closest geometric center is marked as the center grid. In this case, the spatial relationship between the grids needs to be further considered. The center grid's position is determined by calculating the geometric center of the closest grid. The geometric center is calculated by adding the coordinates of all the closest grid cells and dividing by the number of grid cells to obtain the coordinate value of the geometric center.

[0041] S5. Center grid screening and target area determination: Calculate the product data: For each central grid, the product of the spectral difference between it and the other central grids and the Euclidean distance is calculated. Spectral difference can be calculated using the formula mentioned above, quantifying the difference between the two hyperspectral curves over the target spectral range through integration. Euclidean distance is calculated based on the spatial coordinates of the central grid, using the distance formula between two points to calculate the straight-line distance between them.

[0042] The calculated spectral difference is multiplied by the Euclidean distance to obtain the product data between each central grid and other central grids. These product data reflect the similarities and differences between the central grids, providing a basis for subsequent screening of the closest central grid.

[0043] Filter the closest center mesh: Based on the product data, the closest central grids are selected. This selection method can be sorted by the size of the product data, selecting the central grid with the smallest product data as the closest central grid. This ensures that the selected central grids have high similarity in both spectral characteristics and spatial position.

[0044] Draw a scatter plot and a curve: For any central grid, obtain the product data of the central grid and other central grids, and sort the product data from small to large. Use the sequence number as the horizontal axis and the product value as the vertical axis to establish a rectangular coordinate system and draw a scatter plot.

[0045] Connect the points in the scatter plot sequentially, creating a curve labeled q(t), where t represents the sequence number. By connecting the points in the scatter plot, we can visually visualize the changing trend of the product data. Also, connect the first and last coordinate points, creating a straight line labeled h(t). This line represents the relationship between the starting and ending values ​​of the product data.

[0046] Compute the difference and determine the target sequence: Calculate the integral difference of h(t)-q(t). The larger the difference, the closer q(t) is to the horizontal axis of the coordinate, that is, the smoother the beginning of the sequence.

[0047] Mark nodes and target areas: In the scatter plot of the target sequence, connect the first coordinate point as the starting point and the other coordinate points as the end points to obtain several line segments. Mark the end point corresponding to the first line segment with a slope greater than the preset slope as a node. That is, the curve is flat before the node, and the slope increases after the node.

[0048] The center grids corresponding to the product data coordinate points before the node are the closest center grids. The set of pending grids corresponding to these center grids is marked as the target region. This way, the final target region is obtained, completing the screening and marking of the set of pending grids.

[0049] S6. Determination of the final target area and comparison of mineral spectrum curves: Set the grid W i+1 : In order to further refine the target area, a new grid W needs to be set i+1 Grid W i+1 The side length is L i+1 , and L i+1 =mL i , where m∈(0,1). This means that the new mesh side length is a ratio m of the original mesh side length. By adjusting the value of m, the degree of mesh subdivision can be controlled.

[0050] When setting the grid W i+1 When designing a data center, it is necessary to ensure that it can fully cover the target area and make reasonable divisions based on actual needs and data characteristics.

[0051] Divide the target area: Divide the target area into several grids W i+1 During the division process, the shape, size, and matching degree of the grid with the target area need to be considered. Various division methods can be used, such as uniform division and adaptive division, to ensure that the divided grid can accurately reflect the characteristics of the target area.

[0052] For each grid W i+1 , to determine whether it is completely contained in the target area. If so, it is marked as part of the target area; if not, its intersection with the target area is further determined to determine its belonging.

[0053] Calculate the spectral difference: will be with the center grid W i The inner grid W i+1 As the undetermined grid W i+1 . With the grid W to be determined i+1 As the center, calculate the W of all the peripheral grids in turn i+1 The spectral difference between them.

[0054] The spectral difference can be calculated using the formula mentioned above, which quantifies the difference between the two hyperspectral curves over the target spectral range by integration. The calculation process takes into account the spectral characteristics of the target grid and the surrounding grids, as well as their spatial relationship.

[0055] Filter the outer mesh: The outer grids W whose spectral difference is less than the preset threshold are i+1 With the undetermined grid W i+1 The preset threshold needs to be adjusted according to actual needs and data characteristics to ensure that the selected peripheral grids have sufficient similarity with the undetermined grids.

[0056] By screening the peripheral grids, the scope of the target area can be further expanded to include more grids with similar spectral characteristics.

[0057] Iterative calculation: For the i-th target area, use the grid W i+1 Divide it and calculate the undetermined grid W in the central grid i+1 With the outer grid W i+1 During the calculation process, adjacent and non-adjacent peripheral grids need to be considered to fully evaluate their similarity with the grid to be determined.

[0058] The peripheral grids W whose difference is less than the preset threshold are i+1 With the undetermined grid W i+1 The target area is marked as the i+1th target area. Through continuous iterative calculation, the scope of the target area can be gradually expanded until i=n, ​​and the final target area is obtained.

[0059] The spectral characteristic parameters and texture characteristic parameters of the remote sensing data of the final target area are extracted. The spectral characteristic parameters include spectral curve characteristics, spectral transformation characteristics and spectral measurement characteristics. The texture characteristic parameters include gray-level co-occurrence matrix, gray-level run length, local binary pattern and Gabor filter. The data are compared with the mineral spectral curves in the spectral library and the mineral identification results are output.

[0060] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0061] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present application should be included within the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

[0062] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A mineral identification method based on full-spectrum hyperspectral remote sensing data, characterized in that: The following steps are involved: S1. Obtain full-spectrum hyperspectral remote sensing data of the area to be measured, pre-process the remote sensing data, and reduce the dimension of the remote sensing data based on principal component analysis; S2. Divide the target area into several grids W i , i∈[1,n], n is a natural number, grid W i The side length is L i , L i For a preset length, a hyperspectral curve is generated according to the change of the reflectance of the ground objects in each grid area in the target spectral range; S3, marking any grid as a pending grid, calculating the spectral difference S between the pending grid and the adjacent grids, merging the adjacent grids whose spectral difference S is lower than a preset threshold into the pending grid, and calculating the spectral difference of all adjacent grids in turn; S4, obtaining a number of pending grid sets whose final number of grids is greater than a set number, and for any pending grid set, calculating the mean of the hyperspectral curves of all grids in the pending grid set, and taking the grid closest to the mean as the central grid of the pending grid set; S5, calculating the product of the spectral difference and the Euclidean distance between each central grid and other central grids, selecting several closest central grids according to the product data, and marking the corresponding undetermined grid set as the target area; S6. Set grid W i+1 , the side length of the grid is L i+1 , L i+1 =mL i , m∈(0,1), divide the target area into several grids W i+1 , will be with the center grid W i The inner grid W i+1 As the undetermined grid W i+1 , with the grid to be determined W i+1 Calculate the center and all the peripheral grids W in turn i+1 The spectral difference between them is smaller than the preset threshold value, and the outer grid W i+1 With the undetermined grid W i+1 Mark them together as the new target area; repeat the grid division until the grid W is reached n , get the grid W n Corresponding to the final target area, the spectral characteristic parameters and texture characteristic parameters of the remote sensing data of the final target area are extracted, and compared with the mineral spectral curve in the spectral library, and the mineral identification result is output.

2. The mineral identification method based on full-spectrum hyperspectral remote sensing data according to claim 1 is characterized in that: In S1, the preprocessing includes: The full-spectrum hyperspectral remote sensing data are subjected to band removal, radiation calibration and atmospheric correction. The band removal is to check the radiation DN value of each band and remove the bands with DN value of 0 or signal-to-noise ratio lower than the standard. The radiation calibration is to convert the radiation DN value output by the sensor into the surface radiation brightness value through the radiation calibration coefficient. The atmospheric correction is to correct the data based on the MODTRAN correction model through statistical correction parameters.

3. The mineral identification method based on full-spectrum hyperspectral remote sensing data according to claim 1 is characterized in that: In S3, the calculation of the spectral difference S is specifically as follows: ； Among them, A and B represent the interval values ​​of the target spectral range, x represents the spectral range, f(x) represents the hyperspectral curve of the undetermined grid, and g(x) represents the hyperspectral curve of the adjacent grid.

4. The mineral identification method based on full-spectrum hyperspectral remote sensing data according to claim 1 is characterized in that: In S4, when the number of grids closest to the mean is greater than one, the grid at which the geometric center of the grid closest to the mean is located is marked as the center grid.

5. The mineral identification method based on full-spectrum hyperspectral remote sensing data according to claim 1 is characterized in that: In S5, the specific process of selecting the closest central grids according to the product data is as follows: For any central grid, obtain the product data of the central grid and other central grids, and sort the product data from small to large, establish a rectangular coordinate system, use the sequence number as the horizontal coordinate and the product value as the vertical coordinate, draw a scatter plot, and connect the scatter plots in sequence, and mark the obtained curve as curve q(t), t represents the sequence number, and connect the first coordinate point with the last coordinate point, and mark the obtained straight line as h(t), calculate the integral difference of h(t)-q(t), mark the one with the largest difference as the target sequence, and perform on the target sequence; in the scatter plot where the target sequence is located, connect the first coordinate point as the starting point and the other coordinate points as the end points in sequence to obtain several line segments, mark the end point corresponding to the first line segment whose slope is greater than the preset slope as a node, and the central grid corresponding to the coordinate point of the product data before the node is the closest several central grids.

6. The mineral identification method based on full-spectrum hyperspectral remote sensing data according to claim 1 is characterized in that: In S6, the grid is divided repeatedly until the grid W is reached. n , get the grid W n The specific process of the corresponding final target area is: For the i-th target area, use the grid W i+1 Divide it and calculate the undetermined grid W in the central grid i+1 With the outer grid W i+1 The spectral difference between the two, the periphery includes adjacent and non-adjacent, and the peripheral grid W with a difference less than a preset threshold is i+1 With the undetermined grid W i+1 They are collectively marked as the i+1th target area until i=n, ​​and the final target area is obtained.

7. The mineral identification method based on full-spectrum hyperspectral remote sensing data according to claim 1 is characterized in that: In S6, extracting the spectral characteristic parameters and texture characteristic parameters of the remote sensing data of the final target area includes: The spectral feature parameters include spectral curve features, spectral transformation features and spectral measurement features, and the texture feature parameters include grayscale co-occurrence matrix, grayscale run length, local binary pattern and Gabor filter.

Images