Lithology classification method of hyperspectral remote sensing image
By combining unsupervised clustering and supervised classification methods, hyperspectral remote sensing images are preprocessed and classified for lithology. This solves the problem of the difficulty in applying hyperspectral remote sensing in lithology classification, and achieves efficient and reliable lithology classification, which is suitable for large-area remote sensing lithology classification mapping.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING RES INST OF URANIUM GEOLOGY
- Filing Date
- 2023-11-01
- Publication Date
- 2026-05-12
AI Technical Summary
Hyperspectral remote sensing technology is difficult to apply in lithology classification. Existing methods are time-consuming and costly, and it is difficult to guarantee the generalization ability of lithology classification models. There is also a lack of standard spectral libraries as a reference.
By combining unsupervised clustering and supervised classification methods, a lithology classification model is constructed by preprocessing, denoising, selecting bands, clustering, and assigning lithology classification labels to hyperspectral remote sensing images, ultimately achieving lithology classification.
It improves the efficiency and reliability of hyperspectral remote sensing lithology classification, is easy to implement semi-automated processing, and provides technical support for large-area remote sensing lithology classification mapping.
Smart Images
Figure CN117496239B_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present invention relate to a method for geological identification using optical imaging, specifically to a lithological classification method for hyperspectral remote sensing images. Background Technology
[0002] Lithological classification is one of the main directions of remote sensing geology applications. Hyperspectral remote sensing, due to its spectral subdivision characteristics, has a significant advantage over traditional multispectral remote sensing in rock and mineral identification. Different types of minerals have their own characteristic spectral bands, and some standard mineral spectral libraries have been established based on this, providing a reference and theoretical basis for hyperspectral remote sensing mineral mapping. However, for lithological identification, there is still a lack of standard spectral libraries as a reference. Therefore, hyperspectral remote sensing mineral identification methods, represented by spectral similarity matching, are not applicable to lithological identification. Summary of the Invention
[0003] In view of the above problems, this application provides a lithology classification method for hyperspectral remote sensing images, aiming to quickly achieve lithology classification of hyperspectral remote sensing images.
[0004] This application provides a lithological classification method for hyperspectral remote sensing images, comprising: preprocessing the hyperspectral remote sensing image to obtain a reflectance image; processing the reflectance image to remove noise; selecting reflectance image data corresponding to certain bands in the noise-removed reflectance image; performing clustering processing on the selected reflectance image data to divide the reflectance image data into multiple classification clusters; determining the separability of the reflectance images between each classification cluster; classifying the reflectance image data into multiple categories based on the clustering results and the separability, and assigning lithological classification labels to each category; determining a lithological classification model for the reflectance image based on the reflectance image data and its corresponding lithological classification labels; and classifying the hyperspectral remote sensing image for lithology based on the lithological classification model and the reflectance image.
[0005] The method provided in the embodiments of this application can effectively improve the efficiency of hyperspectral remote sensing lithology classification, while having high reliability and being easy to implement semi-automated processing, providing technical support for large-area remote sensing lithology classification mapping. Attached Figure Description
[0006] Other objects and advantages of the invention will become apparent from the following description of embodiments of the invention with reference to the accompanying drawings, and will help to provide a comprehensive understanding of the invention.
[0007] Figure 1 This is a flowchart of the lithological classification method for hyperspectral remote sensing images provided in the embodiments of this application.
[0008] Figure 2This is a flowchart of the process of processing reflectance images provided in the embodiments of this application.
[0009] Figure 3 This is a flowchart of a lithological classification model for determining reflectance images provided in an embodiment of this application.
[0010] Figure 4 This is a schematic diagram of hyperspectral remote sensing images and lithological classification results of a uranium mining area in Xinjiang, provided in an embodiment of this application. Detailed Implementation
[0011] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only one embodiment of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the described embodiments of this application without creative effort are within the scope of protection of this application.
[0012] It should be noted that, unless otherwise defined, the technical or scientific terms used in this application should have the ordinary meaning understood by a person with ordinary skill in the art to which this application pertains. Where the terms "first," "second," etc., are used throughout the text, they are used only to distinguish similar objects and should not be construed as indicating or implying their relative importance, order of precedence, or implicitly specifying the number of technical features indicated. It should be understood that the data described by "first," "second," etc., can be interchanged where appropriate. Where "and / or" appears throughout the text, it means including three parallel solutions. Taking "A and / or B" as an example, it includes solution A, or solution B, or a solution that satisfies both A and B. Furthermore, for ease of description, spatial relative terms such as "above," "below," "top," "bottom," etc., may be used here, only to describe the spatial positional relationship between one device or feature as shown in the figure and other devices or features. It should be understood that this also includes different orientations in use or operation besides those shown in the figure.
[0013] The inventors of this application discovered that rocks are composed of a mixture of various minerals, and the spectral characteristics of lithology include the spectral characteristics of the minerals. Rock-forming minerals do not exhibit significant spectral characteristics in the 0.4-2.5 μm wavelength range commonly used in hyperspectral remote sensing. Therefore, the spectral characteristics of metamorphic rocks mainly originate from altered minerals, while the spectral characteristics of other rock types without alteration are not obvious, making them difficult to identify using hyperspectral remote sensing. Consequently, the application of hyperspectral remote sensing technology in lithology classification is challenging. Furthermore, using deep learning methods for remote sensing lithology classification requires substantial time investment and makes it difficult to guarantee the generalization ability of the final lithology classification model.
[0014] Therefore, embodiments of this application provide a lithological classification method for hyperspectral remote sensing images. Please refer to [link to relevant documentation]. Figure 1 The method includes the following steps S10 to S80.
[0015] Step S10: Preprocess the hyperspectral remote sensing image to obtain a reflectance image.
[0016] Step S20: Process the reflectance image to remove noise from the reflectance image.
[0017] Step S30: Select reflectance image data corresponding to a portion of the bands in the noise-removed reflectance image.
[0018] Step S40: Perform clustering processing on the selected reflectance image data to divide the reflectance image data into multiple classification clusters.
[0019] Step S50: Determine the separability of reflectance images between each classification cluster.
[0020] Step S60: Based on the results of clustering and separability, the reflectance image data is divided into multiple categories, and each category is assigned a lithological classification label.
[0021] Step S70: Determine the lithological classification model of the reflectance image based on the reflectance image data and its corresponding lithological classification labels.
[0022] Step S80: Classify the hyperspectral remote sensing image based on the lithology classification model and reflectance image.
[0023] The method provided in this embodiment combines computationally efficient unsupervised clustering methods with supervised classification methods from machine learning, and comprehensively considers the spectral features of the hyperspectral imagery itself during processing. This enables efficient and accurate lithological classification of hyperspectral remote sensing images, providing technical support for large-area remote sensing lithological classification mapping with a large number of images and a need for timely processing. The obtained classification results can both discover content not filled in during geological mapping, thus supplementing the geological mapping, and also have good reliability.
[0024] In some embodiments, when preprocessing the hyperspectral remote sensing image to obtain a reflectance image in step S10, the hyperspectral remote sensing image can be processed by radiometric calibration, geometric correction, atmospheric correction and spectral reconstruction to eliminate errors and the influence of atmospheric and illumination factors on the reflection of ground objects, so as to ensure the accuracy of the obtained reflectance image.
[0025] Figure 2 One implementation of processing reflectance images is shown; please refer to [link to relevant documentation]. Figure 2Step S20 includes steps S201 to S203.
[0026] Step S201: Perform a minimum noise separation forward transform on the reflectance image to obtain the forward transformed data, wherein each band in the forward transformed data corresponds to a feature value; Step S202: Select the data of the band whose feature values meet the predetermined conditions based on the feature values of each band after the forward transform; Step S203: Perform a minimum noise separation reverse transform on the selected band data to obtain the reflectance image data after noise removal.
[0027] In some embodiments, during the processing of the reflectance image in step S20, the Minimum Noise Separation Transform (MNF) technique can be used to denoise the reflectance image obtained in step S10. In step S201, a forward transform is first performed on the reflectance image. The forward transform includes two principal component transforms, converting the reflectance image data to a new feature space, which is beneficial for identifying and removing noise. In the resulting forward-transformed data, each band corresponds to a feature value.
[0028] In some embodiments, in step S202, bands with larger eigenvalues can be selected based on the eigenvalues of each band after the forward transformation. A larger eigenvalue indicates a greater amount of information contained within it. Optionally, the first 15-25 bands with larger eigenvalues can be selected.
[0029] In some embodiments, in step S203, the data corresponding to the selected band can be subjected to a minimum noise separation inverse transformation, that is, the data corresponding to the selected band can be transformed from the new feature space back to the feature space of the original data, and finally the reflectance image data after noise removal can be obtained.
[0030] It should be noted that denoising the reflectance image data is the basis for the clustering process in step S40 and the determination of the lithology classification model in step S70. This is because the noise in the data has a significant impact on the results of K-means clustering. The lithology classification model may overfit in some classification problems with high sample data noise, resulting in poor generalization ability.
[0031] In some embodiments, when there is no geological reference map, step S30, when selecting reflectance image data corresponding to a portion of the bands in the noise-removed reflectance image, may include: step S311, selecting a predetermined number of bands with relatively large feature values from the forward-transformed data, and performing color enhancement synthesis on the reflectance images corresponding to the selected bands; step S312, selecting reflectance image data of multiple bands in multiple regions with different image features based on the image features of the synthesized reflectance image. In this embodiment, by selecting reflectance images from different regions with different image features, it is possible to classify based on image feature differences during subsequent clustering processing, obtain reliable classification results, and thus ensure the reliability of the lithological classification model constructed based on the classification results.
[0032] In some embodiments, in the data after the minimum noise separation forward transformation in step S201, a predetermined number of bands with relatively large feature values are selected, and the reflectance images corresponding to the selected bands are color-enhanced and synthesized. That is, the image is enhanced by transforming the multi-band black-and-white image into a color image to highlight the differences between different ground features. Optionally, the predetermined number can be three, that is, in the data after the forward transformation, the top three bands with large feature values are selected, and the reflectance images corresponding to the top three bands are color-enhanced and synthesized.
[0033] In some embodiments, the image features of the reflectance image include color and image texture. In step S312, based on the color differences and image texture feature differences of the synthesized reflectance image, a certain number of reflectance images can be selected from several different regions with significant differences for spectral analysis to determine the spectral features of the reflectance image, which is beneficial for band selection.
[0034] In some embodiments, when performing band selection on the reflectance image in step S312, bands that can better represent the spectral characteristics of the lithology can be selected, and the image data after band selection can be retained.
[0035] In some embodiments, a geological reference map may also be obtained to provide a reference when selecting characteristic bands, and the geological reference map may also provide a reference for setting lithology categories during clustering processes.
[0036] Furthermore, in some embodiments, when a geological reference map is available, in step S30, selecting reflectance image data corresponding to a portion of the bands in the noise-removed reflectance image includes: step S321, spatially overlaying the reflectance image and the geological reference map to obtain an overlaid geological map; step S322, determining multiple geological map units with different lithologies in the geological map based on the overlaid geological map; and step S323, selecting reflectance image data of multiple bands in the regions corresponding to each geological map unit based on the geological map units.
[0037] In some embodiments, in step S321, the acquired geological reference map can be spatially registered, and the reflectance image can be spatially superimposed with the geological reference map to achieve the registration and superposition of the reflectance image and the geological reference map in terms of geographical location, thereby obtaining the superimposed geological map.
[0038] In some embodiments, multiple geological map units of different lithologies can be determined based on the superimposed geological map. Each geological map unit represents the distribution area of a lithology on the geological map, thereby selecting reflectance images from different lithologies to improve the reliability of subsequent clustering results and the lithology classification model. In step S323, a certain number of reflectance images can be randomly selected from the region corresponding to each geological map unit of each lithology for spectral analysis to determine the spectral characteristics of the reflectance images, facilitating band selection.
[0039] In some embodiments, when performing band selection on the reflectance image in step S323, bands that can better represent the spectral characteristics of the lithology can be selected, and the image data after band selection can be retained.
[0040] In some embodiments, when selecting reflectance image data corresponding to a portion of the bands in steps S312 and S323, for reflectance images with characteristic absorption peaks, a band range that is wider than the band corresponding to the characteristic absorption peaks is selected.
[0041] In some embodiments, each pixel in the reflectance image corresponds to a spectral reflectance curve. For a spectral reflectance curve with a characteristic absorption peak, in addition to selecting the band range corresponding to the characteristic absorption peak, a certain number of bands larger than that range are selected. The purpose is to comprehensively consider the peak shape of the band corresponding to the characteristic absorption peak and its relationship with adjacent bands. For a spectral reflectance curve without a characteristic absorption peak, such as the spectral curves of various rocks showing a gentle trend, only a small number of bands are selected in this case to avoid classifying the lithology of various rocks as consistent.
[0042] Optionally, for spectral reflectance curves with characteristic absorption peaks, 5-10 bands are selected beyond the band range corresponding to the characteristic absorption peak; for example, if the band corresponding to the characteristic absorption peak is the 30th to 40th, then the 25th to 45th bands, which are wider than that band, are selected. For spectral reflectance curves without characteristic absorption peaks, only 3-5 bands are selected. Furthermore, water vapor absorption bands near 1.4 and 1.9 μm are not selected.
[0043] In some embodiments, when there is no geological reference map, step S40 includes: step S411, selecting multiple bands with relatively large feature values from the forward-transformed data, and performing color enhancement synthesis on the reflectance images corresponding to the selected multiple bands; step S412, determining the number of lithological categories in the synthesized reflectance image based on the image features of the synthesized reflectance image; step S413, determining the number of classification clusters to be divided during clustering processing based on the number of lithological categories; step S414, performing clustering processing on the selected reflectance image data, dividing the reflectance image data into multiple classification clusters corresponding to the number of classification clusters.
[0044] In some embodiments, in step S411, multiple bands with relatively large eigenvalues can be selected from the data after the minimum noise separation forward transformation in step S201, and color enhancement synthesis can be performed on the reflectance images corresponding to the selected multiple bands. Optionally, the top 3 bands with relatively large eigenvalues can be selected from the forward transformed data, and color enhancement synthesis can be performed on the reflectance images corresponding to the top 3 bands.
[0045] In some embodiments, in step S412, the reflectance image can be interpreted, and the number of lithological categories in the synthesized reflectance image can be determined based on the color differences and image texture features of the synthesized reflectance image.
[0046] In some embodiments, in step S413, when determining the number of clusters to be divided during clustering based on the number of lithological categories in the synthesized reflectance image, the number of clusters can be set to twice the number of lithological categories in the reflectance image. Specifically, in the absence of a geological reference map, since the human eye can only perceive visible light, and the wavelength range of visible light is shorter than that of the reflectance image, the number of lithological categories determined during visual interpretation of the color-enhanced synthesized reflectance image in step S412 is likely to be less than the actual number of lithological categories. Therefore, to ensure the accuracy and reliability of the clustering results, the number of clusters is set to twice the number of lithological categories in the reflectance image.
[0047] In clustering, a cluster is a set of data objects. Objects in the same cluster are similar to each other and different from objects in other clusters.
[0048] In some embodiments, in step S414, when performing clustering processing on the selected reflectance image data, K-means clustering processing can be performed on the reflectance image data after band selection in step S30, and the reflectance image data can be divided into multiple clusters corresponding to the number of clusters.
[0049] In some embodiments, when a geological reference map is available, step S40 includes: step S421, determining the number of classification clusters to be divided during clustering based on the number of lithology categories in the lithological geological reference map; step S422, performing clustering processing on the selected reflectance image data to divide the reflectance image data into multiple classification clusters corresponding to the number of classification clusters.
[0050] In some embodiments, in step S421, a geological reference map of the study area can be obtained, and the number of classification clusters to be divided during clustering processing can be determined based on the number of lithology categories shown in the geological reference map. Optionally, the number of classification clusters to be divided during clustering processing can be set to twice the number of lithology categories in the geological reference map. Where a geological reference map exists, the purpose of using hyperspectral imagery for lithology identification and classification is to supplement the geological reference map by clustering the reflectance image data to identify more lithology categories than those shown in the geological reference map. If the number of classification clusters is set to be the same as the number of lithology categories shown in the geological reference map, this purpose cannot be achieved. Therefore, setting the number of classification clusters to twice the number of lithology categories in the geological reference map further supplements the reference geological map through remote sensing lithology mapping.
[0051] In some embodiments, in step S422, when performing clustering processing on the selected reflectance image data, K-means clustering processing can be performed on the reflectance image data after band selection in step S30, and the reflectance image data can be divided into multiple classification clusters corresponding to the number of classification clusters.
[0052] It should be noted that by clustering the reflectance images, the data are divided into different categories according to their similarity and dissimilarity. The data in each category are as similar as possible, while the data in different categories are as dissimilar as possible. This makes it easier to discover the hidden patterns in the data and to make a preliminary classification of the lithology.
[0053] In some embodiments, step S50 includes: randomly selecting a portion of reflectance image data from each classification cluster; determining the separability between each classification cluster based on the selected portion of reflectance image data, so as to correct the results of clustering and ensure the reliability of the final classification results.
[0054] In some embodiments, a portion of the reflectance image data from each of the categories divided in step S40 can be randomly selected to ensure data representativeness. The pixel spectral bands for each category are the multiple bands selected during band selection in step S30. Optionally, a random selection method can be used when selecting data.
[0055] In some embodiments, when determining the separability between different taxonomic clusters, the separability between different taxonomic clusters can be determined using distance and transform separation based on randomly selected partial reflectance image data. Optionally, the distance and transform separation methods can be used in combination.
[0056] When using the distance method to determine the separability between taxa, the distance between each taxa is calculated, and the distance values are used to evaluate the separability between taxa. The distance value is then the separability between taxa. The expression for calculating the distance is as follows:
[0057]
[0058]
[0059] Where J represents the distance between two classification clusters, and its value ranges from 0 to 2. and It represents the average of a certain feature between two classification clusters. and The standard deviation of a certain feature between two classification clusters.
[0060] When using the transformation separation degree method to determine the separability between each class cluster, the separability value between each class cluster is calculated, and the separability between each class cluster is judged by the separability value. At this time, the separability value is the separability between each class cluster.
[0061] Separable value The expression is as follows:
[0062]
[0063]
[0064] in, This represents the separability value between two clusters i and j. This represents the dispersion between two clusters i and j. and For classification clusters and The covariance matrix; and For classification clusters and The average vector.
[0065] In some embodiments, step S60 includes: merging taxonomic clusters to obtain a single category based on a separability of less than or equal to 1; and classifying the taxonomic cluster as a single category based on a separability of greater than 1. Specifically, based on the separability of the taxonomic clusters obtained in step S50, two taxonomic clusters with a separability of less than or equal to 1 can be merged into a single category. For two taxonomic clusters with a separability greater than 1, both taxonomic clusters are treated as independent categories. Finally, lithological classification labels are assigned to each category obtained after analysis.
[0066] When combining distance and transformation separation methods to determine the separability of taxonomic clusters, the distance J and the separability value between the two taxonomic clusters can be calculated separately. When the distance value J≤1 or the separability value When the distance value J is ≤1, the classification clusters are merged. That is, only the calculated distance value J and the separability value are needed. If a value less than or equal to 1 appears in the value, the two classification clusters can be considered inseparable, and the two classification clusters should be merged.
[0067] This embodiment re-divides the classification clusters by using the separability between them, ensuring the accuracy of the classification results of hyperspectral reflectance image data under the joint constraints of clustering and separability criteria.
[0068] Figure 3 One implementation of a lithological classification model for determining reflectance images is shown; please refer to [link to relevant documentation]. Figure 3 Step S70 includes: Step S701, constructing an initial lithology classification model; Step S702, using reflectance image data from multiple categories and their corresponding lithology classification labels as sample data, training and optimizing the initial lithology classification model, and obtaining the optimized lithology classification model. This embodiment combines computationally efficient unsupervised clustering methods with a supervised lithology classification model, enabling efficient lithology classification of hyperspectral images.
[0069] In some embodiments, in step S701, the method for constructing the initial lithology classification model can employ the random forest method. Random forest is an ensemble learning method in machine learning, an algorithm that integrates multiple decision trees based on the idea of ensemble learning. Its basic unit is a decision tree, and the constructed "forest" is an ensemble of decision trees. Each decision tree is a classifier; for an input sample, N decision trees will produce N classification votes. The random forest integrates all classification votes and designates the category with the most votes as the final output, which is the final classification result. Optionally, when constructing the initial random forest model, the number of decision trees can be set to 50, and the maximum number of features can be set to 70%.
[0070] In some embodiments, in step S702, reflectance image data of various categories obtained in step S60 can be randomly selected as sample data to construct a random forest model. In this embodiment, by establishing a random forest model, the high computational efficiency and good classification performance of random forests can be utilized to quickly achieve hyperspectral image classification and obtain lithological classification results in a short time.
[0071] In some embodiments, the initial lithology classification model can be trained and optimized using sample data, and the model's performance can be evaluated by calculating evaluation metrics. If the model's performance is poor, model optimization is required until the model's fitting ability no longer improves significantly or the evaluation metrics reach a preset standard, at which point optimization is stopped, resulting in an optimized lithology classification model.
[0072] Optionally, the performance of the lithology classification model can be evaluated by calculating metrics such as accuracy, precision, recall, and F1 score. When model optimization is required, the number of decision trees can be increased first to improve the model's fitting ability. If the model's fitting ability does not significantly improve and the preset standard is not met, the maximum number of features can be increased to improve the fitting ability of each sub-model, until the model's fitting ability does not significantly improve or the evaluation metrics reach the preset standard.
[0073] In some embodiments, in step S80, the optimized lithology classification model from step S70 can be read, the hyperspectral reflectance image can be input into the established lithology classification model, the hyperspectral remote sensing image can be classified into lithology, and finally a lithology classification map can be obtained.
[0074] Please see Figure 4 , Figure 4 The image shows a hyperspectral remote sensing image of a uranium mining area in Xinjiang and its lithological classification results. The upper image is the hyperspectral remote sensing image, and the lower image is the corresponding lithological classification image. Different colored areas in the lithological classification image represent different lithologies.
[0075] Regarding the embodiments of the present invention, it should also be noted that, without conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other to obtain new embodiments.
[0076] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. The scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A lithological classification method for hyperspectral remote sensing images, characterized in that, Includes the following steps: The hyperspectral remote sensing image is preprocessed to obtain a reflectance image; The reflectance image is processed to remove noise from it; Select reflectance image data corresponding to a portion of the bands in the noise-removed reflectance image; The selected reflectance image data is clustered to divide the reflectance image data into multiple classification clusters; Determine the separability of the reflectance images among the various classification clusters; Based on the results of the clustering process and the separability, the reflectance image data is divided into multiple categories, and each category is assigned a lithological classification label. Based on the reflectance image data and its corresponding lithological classification labels, determine the lithological classification model of the reflectance image; Based on the lithology classification model and the reflectance image, the hyperspectral remote sensing image is classified into lithologies. The process of processing the reflectance image to remove noise from the reflectance image includes: The reflectance image is subjected to a minimum noise separation forward transform to obtain the forward-transformed data; wherein, each band in the forward-transformed data corresponds to a characteristic value; Based on the characteristic values of each band after forward transformation, select the data of the band whose characteristic values satisfy a predetermined condition; The data of the selected band are subjected to a minimum noise separation inverse transform to obtain the reflectance image data after noise removal. The step of clustering the selected reflectance image data to divide it into multiple classification clusters includes: In the forward-transformed data, multiple bands with relatively large feature values are selected, and color enhancement synthesis is performed on the reflectance images corresponding to the selected multiple bands. Based on the image characteristics of the synthesized reflectance image, determine the number of lithological categories in the synthesized reflectance image; The number of taxonomic clusters to be divided during the clustering process is determined based on the number of lithological categories. The selected reflectance image data is clustered to divide the reflectance image data into multiple clusters corresponding to the number of clusters.
2. The method according to claim 1, characterized in that, The method further includes: acquiring a geological reference map; The step of clustering the selected reflectance image data to divide it into multiple classification clusters includes: The number of taxonomic clusters is determined based on the number of lithology categories in the geological reference map; The selected reflectance image data is clustered to divide the reflectance image data into multiple clusters corresponding to the number of clusters.
3. The method according to claim 1, characterized in that, The step of selecting reflectance image data corresponding to a portion of the reflectance image after noise removal includes: In the forward-transformed data, a preset number of bands with relatively large feature values are selected, and the reflectance images corresponding to the selected bands are color-enhanced and synthesized. Based on the image characteristics of the synthesized reflectance image, reflectance image data of multiple bands are selected from multiple regions where the image characteristics differ.
4. The method according to claim 1, characterized in that, The method further includes: acquiring a geological reference map; The step of selecting reflectance image data corresponding to a portion of the reflectance image after noise removal includes: The reflectance image is spatially superimposed with the geological reference map to obtain the superimposed geological map; Based on the superimposed geological map, multiple geological map units with different lithologies are determined in the geological map; Based on the geological map unit, reflectance image data of multiple bands are selected in the region corresponding to each geological map unit.
5. The method according to claim 3 or 4, characterized in that, When selecting reflectance image data corresponding to a portion of the bands, for reflectance images with characteristic absorption peaks, a band range that is wider than the band corresponding to the characteristic absorption peak is selected.
6. The method according to claim 1, characterized in that, Determining the separability of the reflectance images among the various classification clusters includes: Randomly select a portion of reflectance image data from each classification cluster; Based on selected partial reflectance image data, the separability between each classification cluster is determined.
7. The method according to claim 6, characterized in that, Based on the results of the clustering process and the separability, the reflectance image data is divided into multiple categories, including: If the separability between the classification clusters is less than or equal to 1, the classification clusters are merged to obtain a single category. The classification cluster is classified as a category based on the separability between the classification clusters being greater than 1.
8. The method according to claim 1, characterized in that, The step of determining the lithological classification model of the reflectance image based on the reflectance image data and its corresponding lithological classification label includes: Construct an initial lithological classification model; Using reflectance image data and their corresponding lithological classification labels from the multiple categories as sample data, the initial lithological classification model is trained and optimized, and the optimized lithological classification model is obtained.