Hyperspectral imaging data processing method for analyzing shapes and mineral components of rock-soil particles

Abstract

The invention discloses a hyperspectral imaging data processing method for analyzing rock-soil particle shapes and mineral components, and the method comprises the following steps: obtaining hyperspectral data of to-be-detected mineral particles, and preprocessing the hyperspectral data of the to-be-detected mineral particles; obtaining a mineral component test set based on the preprocessed hyperspectral data of the to-be-tested mineral particles; and calling the generated mineral component model weight file to identify and classify the mineral component test set to obtain corresponding end member mineral components, and performing statistics on the end member mineral components to obtain mineral components and the relative content of each mineral component so as to realize mineral component analysis of the to-be-detected mineral particles. The method can accurately realize the conjoint analysis of the shapes of the mineral particles and the mineral components.

Description

Technical Field

[0001] The present invention relates to the technical field of hyperspectral data processing, and in particular to a hyperspectral imaging data processing method for analyzing rock and soil particle shapes and mineral components. Background Art

[0002] The engineering mechanical properties of rock and soil are closely related to its physical and chemical properties such as color, structure, chemical and mineral composition, and particle size and shape. Therefore, in the fields of engineering geological survey, shield construction, oil drilling, etc., exploration and sampling, in-situ testing and rock cuttings logging are widely carried out at the engineering site to achieve the layered naming and engineering mechanical property evaluation of underground rock and soil, and guide the design and construction of underground projects. At present, the artificial test method of naked eye observation and description is mainly used for the obtained rock and soil samples, but this method has problems such as large sampling disturbance, strong subjectivity and experience, low efficiency and lack of accuracy.

[0003] In recent years, the introduction of hyperspectral imaging technology has significantly improved the information acquisition capabilities of rock and soil bodies, for example, from three discrete channels of RGB (red, green, blue) color images to more than 100 continuous channels of hyperspectral data. Hyperspectral imaging technology has the ability to acquire information in continuous wide bands in the ultraviolet, visible and infrared bands. It uses a spectroscopic device to continuously modulate the wavelength of the input light, and can simultaneously obtain the spatial position and spectral information of the target object in the imaging area in one imaging. Among them, the spectral information of the object can reflect its material structure and composition characteristics, and by statistically analyzing the position and distribution of pixels in two-dimensional space, the physical properties of the object in the imaging area, such as color, size and shape, can be obtained. The data obtained by hyperspectral imaging is a three-dimensional data cube, which is usually composed of hundreds of single-channel grayscale images superimposed. The length and width of the image are two spatial dimensions, and the direction perpendicular to the image plane is called the spectral dimension.

[0004] Based on the current research status, hyperspectral imaging technology is mainly used for mineral composition or element analysis of rock and soil bodies, while the color, size and shape characteristics of particles still rely on manual visual or color image-based recognition methods, but these methods are difficult to identify minerals with low reflectivity characteristics. In order to overcome the shortcomings of the above-mentioned existing technologies and improve the level of detection, identification and comprehensive analysis of rock and soil bodies, it is necessary to provide a hyperspectral imaging data processing method for analyzing rock and soil particle shape and mineral composition. Summary of the invention

[0005] The present invention aims to solve the technical problems in the related art at least to a certain extent. To this end, the purpose of the present invention is to provide a hyperspectral imaging data processing method for analyzing the shape of rock and soil particles and mineral components, so as to accurately realize the joint analysis of the shape of mineral particles and mineral components.

[0006] To achieve the above object, the present invention is implemented through the following technical solutions:

[0007] A hyperspectral imaging data processing method for analyzing rock and soil particle shape and mineral composition, comprising:

[0008] Acquire hyperspectral data of the mineral particles to be tested, and preprocess the hyperspectral data of the mineral particles to be tested;

[0009] A mineral component test set is obtained based on the preprocessed hyperspectral data of the mineral particles to be tested;

[0010] The generated mineral component model weight file is called to identify and classify the mineral component test set to obtain the corresponding end-member mineral composition, and the mineral composition and the relative content of each mineral component are obtained based on the end-member mineral composition statistics to realize the mineral composition analysis of the mineral particles to be tested.

[0011] Preferably, the hyperspectral data of the mineral particles to be measured are preprocessed, including:

[0012] The basic information in the hyperspectral data of the mineral particles to be tested is read, and the hyperspectral data of the mineral particles to be tested is subjected to data partition correction; wherein the basic information includes data format, spectral range, image depth, image size and sampling interval; the data partition correction includes region of interest delineation, baseline correction, reflectance calibration, spectral denoising, reflectance parameter normalization processing and calculation of derivative spectrum.

[0013] Preferably, the preprocessed hyperspectral data of the mineral particles to be tested are dissolved into spectral vectors to be tested according to different end members to obtain the mineral component test set.

[0014] Preferably, the steps of generating the mineral component model weight file include:

[0015] Acquire hyperspectral data of standard mineral particles, wherein the standard mineral particles are mineral particles with known mineral composition;

[0016] Generate a label array and superimpose it on the hyperspectral data of standard mineral particles to form labeled standard hyperspectral data;

[0017] The annotated standard hyperspectral data are dissolved into standard spectral vectors according to different end members, and the mineral component training set and the mineral component verification set are obtained according to the data distribution ratio;

[0018] The mineral component training set and the mineral component verification set are input into the one-dimensional deep learning model for iterative training and verification to generate the mineral component model weight file.

[0019] Preferably, the annotation array includes boundary elements and type elements;

[0020] The gray value corresponding to the boundary element is 0, and it is located at the particle boundary position of the standard mineral particle, and the boundary element of a single mineral particle forms a closed loop;

[0021] The grayscale value corresponding to the type element is in the range of 1 to 255, is located inside the standard mineral particle, and the type element of a single mineral particle forms a connected domain;

[0022] The grayscale value of the type element corresponds one-to-one to the mineral composition of the standard mineral particles.

[0023] Preferably, the method further comprises performing particle segmentation and shape analysis on the mineral particles to be measured, and performing particle segmentation and shape analysis on the mineral particles to be measured comprises:

[0024] A particle segmentation test set is obtained based on the preprocessed hyperspectral data of the mineral particles to be tested;

[0025] The generated particle segmentation model weight file is called to perform instance segmentation on the particle segmentation test set, obtain the instance boundary point coordinates of the mineral particles to be tested, and generate a two-dimensional instance mask;

[0026] The connected domain analysis is performed on the two-dimensional instance mask to obtain the shape parameters of the two-dimensional instance mask, and the single particle shape and particle size distribution information of the mineral particles to be tested are statistically obtained to achieve particle segmentation and shape analysis.

[0027] Preferably, the preprocessed hyperspectral data of the mineral particles to be tested are decomposed into spectral images to be tested according to different wavelengths to obtain the particle segmentation test set.

[0028] Preferably, the step of generating the particle segmentation model weight file includes:

[0029] Synthesize feature grayscale image;

[0030] The feature grayscale image and the annotation array are superimposed, and after random data augmentation operations, they are randomly cropped into images and annotation files of several preset sizes, and then allocated according to the data allocation ratio to obtain a particle segmentation training set and a particle segmentation verification set;

[0031] The particle segmentation training set and the particle segmentation verification set are input into the two-dimensional deep learning model for iterative training and verification to generate the particle segmentation model weight file.

[0032] Preferably, synthesizing the characteristic grayscale image includes:

[0033] Comparing and analyzing the mineral component data set based on the mineral component identification result to determine the characteristic absorption peak of the mineral component, extracting the central wavelength and the half-maximum full width of the characteristic absorption peak, so as to calculate the difference characteristic wavelength, the mineral component data set includes the mineral component training set and the mineral component verification set;

[0034] A difference characteristic wavelength search function is defined, and a difference characteristic wavelength search is performed within the spectral range by using the difference characteristic wavelength search function to extract a grayscale image at a central wavelength and a grayscale image at a difference characteristic wavelength, and the extracted images are processed by a linear superposition and contrast enhancement method to synthesize the characteristic grayscale image.

[0035] Preferably, the random data enhancement operation includes at least noise injection, blurring, rotation, flipping and color transformation operations.

[0036] The present invention has at least the following technical effects:

[0037] The present invention provides a hyperspectral imaging data processing method for analyzing rock and soil particle shape and mineral composition. The method can analyze the particle shape and mineral composition of rock and soil particles at the same time through a single hyperspectral imaging, thereby forming a multi-parameter data joint acquisition and analysis capability, further simplifying the system structure, and avoiding the adverse effects of secondary measurement on the measurement and analysis speed and accuracy; at the same time, based on the mineral component identification and classification results, the particle segmentation data is acquired and screened, effectively improving the quantity and quality of deep learning training and verification data, and can significantly reduce the annotation workload.

[0038] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 The present invention is a flowchart of a method for processing hyperspectral imaging data for analyzing rock and soil particle shapes and mineral components according to an embodiment of the present invention.

[0040] Figure 2 This is a specific implementation flow chart of the hyperspectral imaging data processing method for analyzing rock and soil particle shape and mineral composition according to an embodiment of the present invention.

[0041] Figure 3 Schematic diagram of a standard sample base according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The present embodiment is described in detail below, and examples of the embodiment are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0043] The hyperspectral imaging data processing method for analyzing rock and soil particle shapes and mineral components according to this embodiment is described below with reference to the accompanying drawings.

[0044] Figure 1 Flow chart of the hyperspectral imaging data processing method for analyzing rock and soil particle shape and mineral composition according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0045] Step S101: acquiring hyperspectral data of the mineral particles to be tested, and preprocessing the hyperspectral data of the mineral particles to be tested.

[0046] The preprocessing of the hyperspectral data of the mineral particles to be tested includes reading the basic information in the hyperspectral data of the mineral particles to be tested, and performing data partition correction on the hyperspectral data of the mineral particles to be tested. The basic information includes data format, spectral range, image depth, image size and sampling interval; the data partition correction includes region of interest delineation, baseline correction, reflectance calibration, spectral noise reduction, reflectance parameter normalization and calculation of derivative spectrum.

[0047] Step S102: obtaining a mineral component test set based on the preprocessed hyperspectral data of the mineral particles to be tested.

[0048] Specifically, the preprocessed hyperspectral data of the mineral particles to be tested are dissolved into spectral vectors to be tested according to different end members to obtain a mineral component test set.

[0049] Step S103: Call the generated mineral component model weight file to identify and classify the mineral component test set, obtain the corresponding end-member mineral components, and obtain the mineral components and the relative content of each mineral component based on the end-member mineral components to achieve mineral component analysis of the mineral particles to be tested.

[0050] Among them, the steps for generating the mineral component model weight file include: obtaining the hyperspectral data of standard mineral particles, which are mineral particles with known mineral composition; generating a labeled array and superimposing it on the hyperspectral data of standard mineral particles to form labeled standard hyperspectral data; decomposing the labeled standard hyperspectral data into standard spectral vectors according to different end members, and distributing them according to the data distribution ratio to obtain a mineral component training set and a mineral component verification set; inputting the mineral component training set and the mineral component verification set into a one-dimensional deep learning model for iterative training and verification, and thus generating a mineral component model weight file.

[0051] In this embodiment, the annotation array includes boundary elements and type elements. The grayscale value corresponding to the boundary element is 0, located at the particle boundary position of the standard mineral particle, and the boundary element of a single mineral particle forms a closed loop. The grayscale value corresponding to the type element is in the range of 1 to 255, located inside the standard mineral particle, and the type element of a single mineral particle forms a connected domain. It should be noted that the grayscale value of the type element corresponds one-to-one to the mineral composition of the standard mineral particle.

[0052] In one embodiment of the present invention, the method also includes performing particle segmentation and shape analysis on the mineral particles to be tested, and the particle segmentation and shape analysis on the mineral particles to be tested includes: obtaining a particle segmentation test set based on the preprocessed hyperspectral data of the mineral particles to be tested; calling the generated particle segmentation model weight file to perform instance segmentation on the particle segmentation test set to obtain the instance boundary point coordinates of the mineral particles to be tested, and generate a two-dimensional instance mask; performing connected domain analysis on the two-dimensional instance mask to obtain the shape parameters of the two-dimensional instance mask, and statistically obtaining the single particle shape and particle size distribution information of the mineral particles to be tested, thereby realizing particle segmentation and shape analysis.

[0053] Among them, the pre-processed hyperspectral data of the mineral particles to be tested are decomposed into spectral images to be tested according to different wavelengths to obtain the particle segmentation test set.

[0054] In one embodiment of the present invention, the step of generating a particle segmentation model weight file includes: synthesizing a feature grayscale image, superimposing the feature grayscale image with an annotation array, and after performing random data enhancement operations such as injecting noise, blurring, rotating, flipping, and color transformation, randomly cropping the image and annotation files into a number of preset sizes, and distributing the particle segmentation training set and the particle segmentation verification set according to the data distribution ratio. The particle segmentation training set and the particle segmentation verification set are input into a two-dimensional deep learning model for iterative training and verification, and the particle segmentation model weight file can be generated.

[0055] In this embodiment, synthesizing a characteristic grayscale image includes: comparing and analyzing a mineral component data set based on the mineral component identification result to determine the characteristic absorption peak of the mineral component, extracting the central wavelength and half-maximum full width of the characteristic absorption peak, so as to calculate the difference characteristic wavelength, wherein the mineral component data set includes a mineral component training set and a mineral component verification set.

[0056] Furthermore, a difference characteristic wavelength search function is defined, and a difference characteristic wavelength search is performed within the spectral range through the difference characteristic wavelength search function to extract the grayscale image at the center wavelength and the grayscale image at the difference characteristic wavelength. The extracted images are then processed by linear superposition and contrast enhancement methods to synthesize the characteristic grayscale image.

[0057] Figure 2 This is a specific implementation flow chart of the hyperspectral imaging data processing method for analyzing rock and soil particle shape and mineral composition according to an embodiment of the present invention. Figure 2 As shown, the method specifically includes:

[0058] Step 1: Hyperspectral data preprocessing

[0059] (1) Read basic information

[0060] The basic information in the hyperspectral data of the mineral particles to be measured includes data format, spectral range (starting wavelength SW, ending wavelength EW), image depth, image size and sampling interval SI, etc.

[0061] (2) Data partition correction

[0062] The hyperspectral data of the mineral particles to be tested are subjected to data partition correction, which includes region of interest (ROI) delineation, baseline correction, reflectance calibration, spectral denoising, normalization and calculation of derivative spectra.

[0063] Region of interest delineation is to remove some data that are irrelevant and weakly related to the test sample within the hyperspectral scanning imaging area to reduce the amount of calculation and improve processing and analysis efficiency.

[0064] Baseline correction is to remove the offset effect of factors such as lenses, spectrometers and light-guiding fibers on the sample spectrum during light propagation, so as to reflect the true spectral information of the test sample.

[0065] Reflectivity calibration is to use the spectral data of the black standard diffuse reflectance target plate and the white standard diffuse reflectance target plate to calibrate and convert the original light intensity parameters into reflectivity parameters. Figure 3In the figure, reference numerals 22 and 23 are arranged in a square groove of a standard sample base 21 . A standard mineral particle 24 is also arranged in the square groove of the standard sample base 21 . A positioning target 20 is also arranged on the standard sample base 21 .

[0066] Spectral noise reduction uses digital signal filtering technology to reduce random noise in spectral data caused by factors such as the system itself and the test environment, and reduce invalid information in spectral data.

[0067] Normalization processing is to normalize the reflectance parameters to the range of [0, 1] to enhance the comparability of spectral data.

[0068] Derivative spectroscopy includes calculating first-order and second-order derivative spectra to highlight subtle features of spectral curves, separate close / overlapping peaks, and eliminate baseline drift.

[0069] Step 2: Generate a deep learning dataset

[0070] (1) Data labeling

[0071] Data annotation is to generate an annotation array and superimpose it on the hyperspectral data of standard mineral particles (mineral particles with known mineral composition) to form annotated standard hyperspectral data.

[0072] Furthermore, the annotation array includes boundary elements and type elements, and the size of the annotation array is the same as the image size, which is 2560px*1920px.

[0073] Furthermore, the gray value corresponding to the boundary element is 0, which is located at the particle boundary position of the standard mineral particle, and for a single mineral particle, its boundary element forms a closed loop.

[0074] Furthermore, the grayscale value corresponding to the type element is in the range of [1, 255], located inside the standard mineral particle, and for a single mineral particle, its type element forms a connected domain.

[0075] Furthermore, the grayscale value of the type element corresponds one-to-one to the mineral composition of the standard mineral particles.

[0076] (2) Generate mineral composition dataset

[0077] The mineral component dataset includes a mineral component training set, a mineral component validation set, and a mineral component test set.

[0078] The mineral component training set and the mineral component verification set are obtained by dissolving the annotated standard hyperspectral data into standard spectral vectors according to different end members and distributing them according to the data distribution ratio.

[0079] The mineral component test set is obtained by dissolving the preprocessed hyperspectral data of the mineral particles to be tested into spectral vectors to be tested according to different end members.

[0080] (3) Generate a particle segmentation test set

[0081] The particle segmentation test set is obtained by decomposing the preprocessed hyperspectral data of the mineral particles to be tested into spectral images to be tested according to different wavelengths.

[0082] Step 3: Identification and classification of mineral components

[0083] (1) The mineral component training set and the mineral component verification set are input into the one-dimensional deep learning model for iterative training and verification to obtain the mineral component model weight file.

[0084] (2) The mineral component test set is identified and classified by calling the mineral component model weight file to obtain the corresponding end-member mineral composition, and further statistics are obtained to obtain information such as the mineral composition and the relative content of each component.

[0085] Step 4: Particle segmentation and shape analysis

[0086] (1) Calculate the difference characteristic wavelength

[0087] Based on the mineral component identification results, the mineral component data set is compared and analyzed to obtain the characteristic absorption peaks of the mineral components, the central wavelength CW and the full width at half maximum FWHM of the characteristic absorption peaks are extracted, and the difference characteristic wavelengths (CWL, CWR) are calculated:

[0088]

[0089] Among them, CWL and CWR represent the left difference characteristic wavelength and the right difference characteristic wavelength respectively.

[0090] (2) Synthesize feature grayscale image

[0091] Define the difference characteristic wavelength search function SF(x):

[0092]

[0093] Where n = INT[(x-SW) / SI], x represents the independent variable of the difference characteristic wavelength search function SF(x), SW represents the starting wavelength of the spectral range, SI is the sampling interval, n represents the intermediate variable, which is a non-negative integer, and INT represents the floor function.

[0094] A difference characteristic wavelength search is performed within the spectral range, and the grayscale image Img(SF(CW)) at the center wavelength and the grayscale images Img(SF(CWL)) and Img(SF(CWR)) at the difference characteristic wavelengths are extracted, and the characteristic grayscale image SCI is synthesized by linear superposition and contrast enhancement method:

[0095]

[0096] Among them, CGI represents the fused image, and HE is the histogram equalization function.

[0097] (3) Generate particle segmentation training set and particle segmentation verification set

[0098] The particle segmentation training set and particle segmentation validation set are obtained by superimposing the feature grayscale image and the annotation array, randomly cropping them into several 224px*224px images and annotation files after random data augmentation operations, and distributing them according to the data distribution ratio.

[0099] Preferably, the random data enhancement operation includes injecting noise, blurring, rotating, flipping and color transformation.

[0100] (4) Particle segmentation training and verification

[0101] The particle segmentation training set and the particle segmentation verification set are input into the two-dimensional deep learning model for iterative training and verification to obtain the particle segmentation model weight file.

[0102] (5) Particle segmentation test

[0103] By calling the particle segmentation model weight file, the particle segmentation test set is instance segmented to obtain the instance boundary point coordinates of the mineral particles to be tested, and a two-dimensional instance mask is generated.

[0104] (6) Shape analysis

[0105] Connected domain analysis is performed on the two-dimensional instance mask to obtain shape parameters of the two-dimensional instance mask, and then further statistics are obtained to obtain information such as single particle shape and particle size distribution of the mineral particles to be tested.

[0106] In summary, the present invention provides a hyperspectral imaging data processing method for analyzing rock and soil particle shape and mineral composition. The method can analyze the particle shape and mineral composition of rock and soil particles at the same time through a single hyperspectral imaging, thereby forming a multi-parameter data joint acquisition and analysis capability, further simplifying the system structure, and avoiding the adverse effects of secondary measurement on the measurement and analysis speed and accuracy; at the same time, the particle segmentation data is acquired and screened based on the mineral component identification and classification results, which effectively improves the quantity and quality of deep learning training and verification data, and can significantly reduce the annotation workload.

[0107] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0108] Although the content of the present invention has been described in detail through the above preferred embodiments, it should be appreciated that the above description should not be considered as a limitation of the present invention. After reading the above content, it will be apparent to those skilled in the art that various modifications and substitutions of the present invention will occur. Therefore, the protection scope of the present invention should be limited by the appended claims.

Claims

1. A hyperspectral imaging data processing method for analyzing rock and soil particle shape and mineral composition, characterized in that: include: Acquire hyperspectral data of the mineral particles to be tested, and preprocess the hyperspectral data of the mineral particles to be tested; A mineral component test set is obtained based on the preprocessed hyperspectral data of the mineral particles to be tested; The generated mineral component model weight file is called to identify and classify the mineral component test set to obtain the corresponding end-member mineral composition, and the mineral composition and the relative content of each mineral component are obtained based on the end-member mineral composition statistics to realize the mineral composition analysis of the mineral particles to be tested.

2. The method for processing hyperspectral imaging data for analyzing rock and soil particle shape and mineral composition according to claim 1, characterized in that: Preprocess the hyperspectral data of the mineral particles to be tested, including: The basic information in the hyperspectral data of the mineral particles to be tested is read, and the hyperspectral data of the mineral particles to be tested is subjected to data partition correction; wherein the basic information includes data format, spectral range, image depth, image size and sampling interval; the data partition correction includes region of interest delineation, baseline correction, reflectance calibration, spectral denoising, reflectance parameter normalization processing and calculation of derivative spectrum.

3. The method for processing hyperspectral imaging data for analyzing rock and soil particle shape and mineral composition according to claim 1, characterized in that: The preprocessed hyperspectral data of the mineral particles to be tested are dissolved into spectral vectors to be tested according to different end members to obtain the mineral component test set.

4. The method for processing hyperspectral imaging data for analyzing rock and soil particle shape and mineral composition according to claim 1, characterized in that: The steps to generate the mineral component model weight file include: Acquire hyperspectral data of standard mineral particles, wherein the standard mineral particles are mineral particles with known mineral composition; Generate a label array and superimpose it on the hyperspectral data of standard mineral particles to form labeled standard hyperspectral data; The annotated standard hyperspectral data are dissolved into standard spectral vectors according to different end members, and the mineral component training set and the mineral component verification set are obtained according to the data distribution ratio; The mineral component training set and the mineral component verification set are input into the one-dimensional deep learning model for iterative training and verification to generate the mineral component model weight file.

5. The method for processing hyperspectral imaging data for analyzing rock and soil particle shape and mineral composition according to claim 4, characterized in that: The annotation array includes boundary elements and type elements; The gray value corresponding to the boundary element is 0, and it is located at the particle boundary position of the standard mineral particle, and the boundary element of a single mineral particle forms a closed loop; The grayscale value corresponding to the type element is in the range of 1 to 255, located inside the standard mineral particle, and the type element of a single mineral particle forms a connected domain; The grayscale value of the type element corresponds one-to-one to the mineral composition of the standard mineral particles.

6. The method for processing hyperspectral imaging data for analyzing rock and soil particle shape and mineral composition according to claim 4, characterized in that: It also includes particle segmentation and shape analysis of the mineral particles to be tested, including: A particle segmentation test set is obtained based on the preprocessed hyperspectral data of the mineral particles to be tested; The generated particle segmentation model weight file is called to perform instance segmentation on the particle segmentation test set, obtain the instance boundary point coordinates of the mineral particles to be tested, and generate a two-dimensional instance mask; The connected domain analysis is performed on the two-dimensional instance mask to obtain the shape parameters of the two-dimensional instance mask, and the single particle shape and particle size distribution information of the mineral particles to be tested are statistically obtained to achieve particle segmentation and shape analysis.

7. The method for processing hyperspectral imaging data for analyzing rock and soil particle shape and mineral composition according to claim 6, characterized in that: The preprocessed hyperspectral data of the mineral particles to be tested are decomposed into spectral images to be tested according to different wavelengths to obtain the particle segmentation test set.

8. The method for processing hyperspectral imaging data for analyzing rock and soil particle shape and mineral composition according to claim 6, characterized in that: The steps to generate the particle segmentation model weight file include: Synthesize feature grayscale image; The feature grayscale image and the annotation array are superimposed, and after random data augmentation operations, they are randomly cropped into images and annotation files of several preset sizes, and then allocated according to the data allocation ratio to obtain a particle segmentation training set and a particle segmentation verification set; The particle segmentation training set and the particle segmentation verification set are input into the two-dimensional deep learning model for iterative training and verification to generate the particle segmentation model weight file.

9. The method for processing hyperspectral imaging data for analyzing rock and soil particle shape and mineral composition according to claim 8, characterized in that: Synthesize feature grayscale images, including: Comparing and analyzing the mineral component data set based on the mineral component identification result to determine the characteristic absorption peak of the mineral component, extracting the central wavelength and the half-maximum full width of the characteristic absorption peak, so as to calculate the difference characteristic wavelength, the mineral component data set includes the mineral component training set and the mineral component verification set; A difference characteristic wavelength search function is defined, and a difference characteristic wavelength search is performed within the spectral range by using the difference characteristic wavelength search function to extract a grayscale image at a central wavelength and a grayscale image at a difference characteristic wavelength, and the extracted images are processed by a linear superposition and contrast enhancement method to synthesize the characteristic grayscale image.

10. The method for processing hyperspectral imaging data for analyzing rock and soil particle shape and mineral composition according to claim 8, characterized in that: The random data enhancement operation includes at least noise injection, blurring, rotation, flipping and color transformation operations.

Images