Intelligent lithology mapping method based on multi-source remote sensing data and machine learning
By using multi-source remote sensing data collaborative preprocessing and machine learning models, the problems of one-sided data utilization and insufficient intelligence in traditional remote sensing lithology mapping methods have been solved, achieving high-precision and high-efficiency automatic lithology identification and mapping.
Patent Information
- Application Number
- CN202511409026.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-29
- Publication Date
- 2026-01-27
AI Technical Summary
Traditional remote sensing lithology mapping methods rely on a single data source, making it difficult to fully capture rock mass characteristics. The feature extraction level is shallow, the level of intelligence is low, and the fusion of multi-source data is insufficient, resulting in low accuracy and efficiency in lithology classification.
Multi-source remote sensing data collaborative preprocessing is adopted to extract multi-dimensional features. Lithology is automatically identified through machine learning models. By combining Gabor filters, Teracorder algorithms and thermal infrared data inversion, high-dimensional fused feature vectors are constructed for intelligent lithology classification and result optimization.
It achieves high-precision and high-efficiency lithology identification, reduces reliance on the experience of geological experts, and improves the automation level of lithology classification and the model's generalization ability.
Smart Images
Figure CN121414883A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of geological survey technology, specifically relating to an intelligent lithological mapping method based on multi-source remote sensing data and machine learning. Background Technology
[0002] Lithological mapping is a fundamental task in geological surveys and resource exploration. Traditional remote sensing lithological mapping methods mainly rely on single types of optical remote sensing data, processed through manual visual interpretation or traditional machine learning algorithms. In the field of hyperspectral remote sensing, methods based on the similarity of spectral curve morphology, such as spectral angle mapping (SAM), are commonly used. These methods typically perform independent analysis on a single data source or employ simple band overlay when fusing multi-source data. The overall process is highly dependent on the prior knowledge of geological experts and human intervention, requiring significant human involvement from feature selection to model parameter optimization.
[0003] These methods have several limitations: First, data utilization is incomplete. Relying on a single data source makes it difficult to comprehensively capture the integrated characteristics of rock masses in terms of space, spectral, and physical properties. Insufficient information dimensions lead to a theoretical upper limit on the accuracy of lithological classification. Second, feature extraction is superficial. Traditional machine learning methods rely on limited manually designed features, while methods such as SAM cannot quantify the actual physical parameters of spectral absorption features, are sensitive to noise, and struggle to uncover deeper patterns hidden in the data. Third, the process lacks intelligence. The method structure is rigid, automation is low, and it relies heavily on expert experience, introducing subjective bias and leading to inefficiency. Furthermore, simple data stacking and fusion methods fail to achieve effective complementarity and synergistic enhancement between multiple information sources, limiting the model's generalization ability. Summary of the Invention
[0004] The technical problem to be solved by the present invention is to overcome the existing defects and provide an intelligent lithological mapping method based on multi-source remote sensing data and machine learning, so as to solve the problems mentioned in the background art.
[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent lithological mapping method based on multi-source remote sensing data and machine learning, comprising the following steps:
[0006] Step 1: Multi-source remote sensing data acquisition and collaborative preprocessing: Simultaneously acquire sub-meter resolution high-resolution remote sensing data, hyperspectral remote sensing data, and thermal infrared remote sensing data covering the same area, and perform collaborative preprocessing on the three types of data, including radiometric calibration, atmospheric correction, and geometric registration, so that they have consistent spatial reference and comparable spectral information.
[0007] Step 2: Multi-dimensional feature extraction: Based on the three types of data after collaborative preprocessing, features representing different lithological attributes are extracted respectively;
[0008] Step 3, Feature-level fusion: The multidimensional features extracted from different data sources in Step 2 are standardized to eliminate differences in scale. Then, they are concatenated along the feature dimensions to construct a unified, high-dimensional fusion feature vector set.
[0009] Step 4, Intelligent Lithology Classification: The fused feature vector set obtained in Step 3 is used as input and imported into a pre-trained machine learning model for automatic identification and classification of lithology types, and a preliminary lithology classification map is output.
[0010] Step 5: Post-processing optimization of classification results: Spatial filtering is performed on the preliminary lithology classification map obtained in Step 4 to eliminate isolated noise points in the map and smooth the boundaries of lithology units, ultimately generating a high-precision lithology mapping result with good spatial continuity and optimized mapping effect.
[0011] Preferably, in step two, extracting spatial texture and morphological features from sub-meter resolution high-resolution remote sensing data includes:
[0012] ① The panchromatic or multispectral bands of the sub-meter high-resolution remote sensing data are processed using a multi-scale, multi-directional Gabor filter bank;
[0013] ② Extract texture features that reflect the spatial structure, striations, joints, and porphyritic texture of rocks;
[0014] ③ Simultaneously, an image segmentation algorithm is used to segment independent rock mass objects from the data and calculate their morphological characteristics.
[0015] Preferably, in step two, extracting mineral quantification features from hyperspectral remote sensing data involves using a physical model-based spectral analysis method. Specifically, this involves performing least-squares fitting on the characteristic absorption valleys in the hyperspectral data based on a modified Gaussian model to invert and obtain physical parameters of the spectral absorption features characterizing the mineral composition. These physical parameters include at least the depth, width, area, and symmetry of the absorption features, and these parameters are used as feature vectors for quantifying the mineral composition.
[0016] Preferably, the physical model-based spectral analysis method is implemented using the Teracorder algorithm framework to achieve quantitative inversion of mineral spectral characteristics, replacing the traditional qualitative comparison method of spectral angle mapping.
[0017] Preferably, in step two, extracting thermophysical features from thermal infrared remote sensing data includes:
[0018] ① Based on multiple thermal infrared bands of ASTER thermal infrared data, the emissivity spectrum and surface temperature of the Earth's surface are obtained by inversion using the radiative transfer equation;
[0019] ② The focus is on extracting the spectral shape characteristics of thermal infrared emissivity that reflect lithological information such as silicate minerals in the 8-12μm wavelength range.
[0020] Preferably, the feature-level fusion in step three involves standardizing the spatial texture and morphological features extracted from sub-meter high-resolution data, the mineral quantification features extracted from hyperspectral data, and the thermophysical features extracted from thermal infrared data, and then splicing them along the feature channel dimension to form an ultra-high-dimensional fused feature vector that integrates spatial, mineral composition, and thermophysical properties, providing complementary input data for machine learning models.
[0021] Preferably, the machine learning model in step four is an ensemble learning model or a deep learning model. During the training phase, the machine learning model is trained using samples of known lithological categories and their corresponding fused feature vectors. The model is then tuned through hyperparameter optimization experiments to address the high-dimensional and nonlinear characteristics of the lithological data, enabling the model to automatically learn the complex rules of lithological classification without relying on manual feature selection.
[0022] Preferably, the spatial filtering process in step five specifically employs the majority voting method or the conditional random field algorithm to improve the spatial coherence of the preliminary classification results, making the boundaries of the final lithological mapping results smoother and the distribution of land features more reasonable.
[0023] Preferably, the sub-meter high-resolution remote sensing data in step one is data from the GF-2 domestic satellite with a spatial resolution of 0.8 meters, and the thermal infrared remote sensing data is data from five thermal infrared bands of the ASTER satellite.
[0024] Preferably, the above method constructs an end-to-end automated process from multi-source data input to lithological mapping results generation, which significantly reduces the reliance on the manual experience of geological experts in traditional methods and achieves high-precision and high-efficiency identification and mapping of lithologies with similar spectral characteristics.
[0025] Compared with existing technologies, this invention provides an intelligent lithological mapping method based on multi-source remote sensing data and machine learning, which has the following beneficial effects:
[0026] 1. This invention realizes the deep feature extraction of multi-source remote sensing information, extracts spatial texture features from sub-meter high-resolution images, inverts the quantitative parameters of mineral composition from hyperspectral data based on physical models (such as Teracoder or MGM), and obtains the thermophysical characteristics of rock mass from thermal infrared data, thereby realizing the comprehensive characterization of multiple physical parameters of lithology.
[0027] 2. This invention constructs a feature-level deep fusion mechanism to integrate the above-mentioned multi-source high-dimensional features into a unified feature vector, and inputs it into an optimized machine learning model to achieve automatic learning of complex feature relationships and intelligent lithology classification;
[0028] 3. This invention establishes an end-to-end automated processing flow, which significantly reduces the reliance on prior human knowledge and subjective intervention, and achieves significant improvements in lithology identification accuracy, automation level and model generalization ability. Attached Figure Description
[0029] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention, but do not constitute a limitation thereof. In the drawings:
[0030] Figure 1 This is a schematic diagram of the process proposed in this invention. Detailed Implementation
[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0032] Please see Figure 1 This invention provides a technical solution: an intelligent lithological mapping method based on multi-source remote sensing data and machine learning, comprising the following steps:
[0033] Step 1: Multi-source remote sensing data acquisition and collaborative preprocessing: Simultaneously acquire sub-meter resolution high-resolution remote sensing data, hyperspectral remote sensing data, and thermal infrared remote sensing data covering the same area, and perform collaborative preprocessing on the three types of data, including radiometric calibration, atmospheric correction, and geometric registration, so that they have consistent spatial reference and comparable spectral information.
[0034] Step 2: Multi-dimensional feature extraction: Based on the three types of data after collaborative preprocessing, features representing different lithological attributes are extracted respectively;
[0035] Step 3, Feature-level fusion: The multidimensional features extracted from different data sources in Step 2 are standardized to eliminate differences in scale. Then, they are concatenated along the feature dimensions to construct a unified, high-dimensional fusion feature vector set.
[0036] Step 4, Intelligent Lithology Classification: The fused feature vector set obtained in Step 3 is used as input and imported into a pre-trained machine learning model for automatic identification and classification of lithology types, and a preliminary lithology classification map is output.
[0037] Step 5: Post-processing optimization of classification results: Spatial filtering is performed on the preliminary lithology classification map obtained in Step 4 to eliminate isolated noise points in the map and smooth the boundaries of lithology units, ultimately generating a high-precision lithology mapping result with good spatial continuity and optimized mapping effect.
[0038] In this invention, preferably, step two, extracting spatial texture and morphological features from sub-meter resolution high-resolution remote sensing data, includes:
[0039] ① Utilize multi-scale, multi-directional Gabor filter banks to process panchromatic or multispectral bands of sub-meter high-resolution remote sensing data;
[0040] ② Extract texture features that reflect the spatial structure, striations, joints, and porphyritic texture of rocks;
[0041] ③ Simultaneously, an image segmentation algorithm is used to segment independent rock mass objects from the data and calculate their morphological characteristics.
[0042] In this invention, preferably, in step two, the extraction of mineral quantification features from hyperspectral remote sensing data is carried out using a physical model-based spectral analysis method. Specifically, it involves performing least-squares fitting on the characteristic absorption valleys in the hyperspectral data based on a modified Gaussian model to invert and obtain physical parameters of spectral absorption features characterizing mineral composition. The physical parameters include at least the depth, width, area, and symmetry of the absorption features, and these parameters are used as feature vectors for quantifying mineral composition.
[0043] In this invention, preferably, the physical model-based spectral analysis method is implemented using the Teracorder algorithm framework to achieve quantitative inversion of mineral spectral characteristics, replacing the traditional qualitative comparison method of spectral angle mapping.
[0044] In this invention, preferably, step two, extracting thermophysical features from thermal infrared remote sensing data, includes:
[0045] ① Based on multiple thermal infrared bands of ASTER thermal infrared data, the emissivity spectrum and surface temperature of the Earth's surface are obtained by inversion using the radiative transfer equation;
[0046] ② The focus is on extracting the spectral shape characteristics of thermal infrared emissivity that reflect lithological information such as silicate minerals in the 8-12μm wavelength range.
[0047] In this invention, preferably, the feature-level fusion in step three involves standardizing the spatial texture and morphological features extracted from sub-meter high-resolution data, the mineral quantification features extracted from hyperspectral data, and the thermophysical features extracted from thermal infrared data, and then splicing them together along the feature channel dimension to form an ultra-high-dimensional fused feature vector that integrates spatial, mineral composition, and thermophysical properties, providing complementary input data for machine learning models.
[0048] In this invention, preferably, the machine learning model in step four is an ensemble learning model or a deep learning model. During the training phase, the machine learning model is trained using samples of known lithological categories and their corresponding fusion feature vectors. The model is then tuned through hyperparameter optimization experiments to address the high-dimensional and nonlinear characteristics of the lithological data, enabling the model to automatically learn the complex rules of lithological classification without relying on manual feature selection.
[0049] In this invention, preferably, the spatial filtering process in step five adopts the majority voting method or the conditional random field algorithm to improve the spatial coherence of the preliminary classification results, so that the boundaries of the final lithological mapping results are smoother and the distribution of land features is more reasonable.
[0050] In this invention, preferably, the sub-meter high-resolution remote sensing data in step one is data from the GF-2 domestic satellite with a spatial resolution of 0.8 meters, and the thermal infrared remote sensing data is data from five thermal infrared bands of the ASTER satellite.
[0051] In this invention, preferably, the above method constructs an end-to-end automated process from multi-source data input to lithological mapping results generation, which significantly reduces the reliance on the manual experience of geological experts in traditional methods and achieves high-precision and high-efficiency identification and mapping of lithologies with similar spectral characteristics.
[0052] Example 1
[0053] A smart lithological mapping method based on multi-source remote sensing data and machine learning includes the following steps:
[0054] Step 1: First, simultaneously acquire sub-meter resolution high-resolution remote sensing data, hyperspectral remote sensing data, and thermal infrared remote sensing data covering the same area. The sub-meter resolution high-resolution remote sensing data uses data from the domestic GF-2 satellite with a spatial resolution of 0.8 meters, while the thermal infrared remote sensing data uses data from five thermal infrared bands of the ASTER satellite. These three types of data undergo collaborative preprocessing, including radiometric calibration, atmospheric correction, and geometric registration, to ensure that the data have a consistent spatial reference and comparable spectral information.
[0055] Step 2: Then, multi-dimensional feature extraction is performed: For sub-meter high-resolution data, multi-scale and multi-directional Gabor filter banks are used to process panchromatic or multi-spectral bands to extract texture features that reflect the spatial structure, stripes, joints, and porphyritic structures of rocks. At the same time, image segmentation algorithms are used to segment rock objects and calculate morphological features.
[0056] Step 3: For hyperspectral data, a physical model-based spectral analysis method is adopted, specifically through the Teracorder algorithm framework. Based on the modified Gaussian model, the characteristic absorption valleys are fitted with least squares to obtain the physical parameters such as the depth, width, area and symmetry of the absorption features, which are used as mineral quantification features.
[0057] Step 4: For thermal infrared data, surface emissivity spectra and temperature are inverted using the radiative transfer equation, with a focus on extracting the emissivity spectral shape characteristics in the 8-12 μm wavelength range to reflect lithological information such as silicate minerals. Then, feature-level fusion is performed.
[0058] Step 5: After standardizing the above spatial texture and morphological features, mineral quantification features and thermophysical features respectively, they are spliced together in the feature channel dimension to form a unified ultra-high-dimensional fusion feature vector; then the fusion feature vector set is input into a pre-trained machine learning model. The model is preferably of the ensemble learning or deep learning type. During the training phase, known lithological samples and corresponding feature vectors are used to optimize hyperparameters so that the model can automatically learn lithological classification rules and output a preliminary lithological classification map.
[0059] Step Six: Finally, the classification results are post-processed and optimized. Spatial filtering is performed using majority voting or conditional random field algorithms to eliminate isolated noise points, smooth lithological boundaries, and generate high-precision lithological mapping results. This embodiment significantly reduces reliance on human experience through an end-to-end automated process, improving the ability to distinguish spectrally similar lithologies and mapping efficiency.
[0060] Example 2
[0061] A smart lithological mapping method based on multi-source remote sensing data and machine learning includes the following steps:
[0062] Step 1: When implementing the intelligent lithological mapping method, firstly acquire sub-meter high-resolution data, hyperspectral data, and ASTER thermal infrared data of the same area, and perform collaborative preprocessing to ensure data consistency;
[0063] Step 2, Feature Extraction Stage: Multi-scale Gabor texture and morphological features are extracted from sub-meter high-resolution data, and emissivity spectral shape features are inverted from thermal infrared data. Hyperspectral data processing employs the Teracorder algorithm framework for deep mining. By modifying the Gaussian model and using least squares fitting, the physical parameters of spectral absorption features, such as depth, width, area, and symmetry, are quantitatively inverted. These parameters directly characterize mineral composition, replacing the qualitative comparison of traditional spectral angle mapping, and greatly improving the geological interpretation and lithological differentiation accuracy of the features.
[0064] Step 3: The multi-source features are then standardized and concatenated into a high-dimensional fusion vector, which is then input into an optimized machine learning model (such as a random forest or convolutional neural network) for classification. The model adapts to the high-dimensional nonlinear features through extensive tuning and automatically completes lithology identification.
[0065] Step 4: The preliminary results are post-processed using the majority voting method to optimize boundary coherence. This embodiment highlights the core advantages of hyperspectral physical model feature extraction, realizing the transformation from spectral morphology similarity judgment to quantitative identification of mineral content, effectively overcoming the limitations of traditional methods in complex geological environments.
[0066] Example 3
[0067] A smart lithological mapping method based on multi-source remote sensing data and machine learning includes the following steps:
[0068] Step 1: In the data preparation stage, GF-2 sub-meter resolution high-resolution data, hyperspectral data, and ASTER thermal infrared data are acquired simultaneously. After radiometric calibration, atmospheric correction, and geometric registration preprocessing, three types of features are extracted: spatial texture and morphological features are extracted from the high-resolution data based on Gabor filters and image segmentation; mineral quantification features are inverted from the hyperspectral data based on the Teracorder algorithm; and emissivity spectral features are extracted from the thermal infrared data based on the radiative transfer equation.
[0069] Step 2, the feature fusion step, is not a simple superposition of image bands, but rather a process of standardizing the high-dimensional feature vectors mentioned above to eliminate dimensional differences, and then splicing them along the feature dimensions to construct a unified feature set that integrates spatial, mineral, and thermophysical properties, providing complementary input for machine learning;
[0070] Step 3: The fused feature set is then input into the optimized ensemble learning or deep learning model for training and classification. The model adapts to high-dimensional characteristics through hyperparameter tuning and automatically mines lithological classification rules.
[0071] Step 4: The final classification image is post-processed using a conditional random field algorithm to improve spatial continuity. This embodiment demonstrates the crucial role of feature-level fusion in information complementarity and model input optimization, overcoming the bottleneck of simply stacking multi-source data.
[0072] Example 4
[0073] A smart lithological mapping method based on multi-source remote sensing data and machine learning includes the following steps:
[0074] Step 1: During implementation, first acquire multi-source remote sensing data, including GF-2 high-resolution data, hyperspectral data, and ASTER thermal infrared data. After collaborative preprocessing, extract spatial texture features, mineral quantification features, and thermophysical features respectively.
[0075] Step 2: After feature fusion, a high-dimensional vector set is formed, which serves as the input to the machine learning model. For model selection, ensemble learning (such as gradient boosting trees) or deep learning (such as deep neural networks) models are adopted. During the training phase, a large number of hyperparameter optimization experiments are conducted to address the high-dimensional and nonlinear characteristics of lithological data, enabling the model to automatically learn complex classification rules without the need for manual feature selection.
[0076] Step 3: After the model training is completed, the fused features are intelligently classified to output a preliminary lithology map. Then, spatial filtering is performed using the majority voting method to optimize the map quality. This embodiment highlights the core role of the machine learning model in automated learning and classification. Through model optimization, high-precision and high-generalization lithology identification is achieved, reducing subjective intervention.
[0077] Example 5
[0078] A smart lithological mapping method based on multi-source remote sensing data and machine learning includes the following steps:
[0079] Step 1: After completing multi-source data acquisition, collaborative preprocessing, multi-dimensional feature extraction, feature-level fusion, and machine learning classification, a preliminary lithology classification map is obtained;
[0080] Step 2: In the post-processing stage, spatial filtering techniques, such as majority voting or conditional random field algorithms, are used to optimize the preliminary results. The majority voting method eliminates isolated misclassification points by voting by neighboring pixels, while the conditional random field considers spatial context to smooth the boundaries, making the distribution of lithological units more reasonable and the boundaries clearer.
[0081] Step 3: Throughout the process, data preprocessing ensures consistency, feature extraction covers spatial, spectral, and thermophysical properties, feature fusion constructs high-dimensional input, machine learning models achieve automatic classification, and post-processing further enhances the practicality and mapping effect of the results.
[0082] Step 4: This embodiment demonstrates the important role of post-processing in the end-to-end process, ensuring the spatial coherence and aesthetics of the final mapping results, and is suitable for actual geological survey needs.
[0083] This invention provides an intelligent lithological mapping method based on multi-source remote sensing data and machine learning. The core of this method lies in its innovative integration of sub-meter-level domestic high-resolution data (providing detailed spatial texture and morphological information), hyperspectral mineral mapping data (providing accurate spectral and mineral composition information of ground features), and ASTER thermal infrared data (providing unique information on the thermal radiation characteristics of ground features) to construct a multi-dimensional, multi-level feature dataset. Furthermore, an advanced machine learning model is employed to deeply fuse and intelligently mine these multi-source heterogeneous features, automatically learning lithological classification rules, ultimately achieving high-precision and high-efficiency automatic lithological identification and mapping.
[0084] The beneficial effects of this invention are that it breaks through the limitations of identification based on a single data source, and significantly improves the accuracy of lithology identification through the complementarity of multi-source information, especially the ability to distinguish lithologies with spectral confusion; it realizes an intelligent process from data preprocessing to lithology classification, greatly improves mapping efficiency, reduces subjective errors, and provides geologists with a new and powerful technical means, which is of great significance for promoting the digital and intelligent transformation of geological work.
[0085] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. An intelligent lithological mapping method based on multi-source remote sensing data and machine learning, characterized in that, Includes the following steps: Step 1: Multi-source remote sensing data acquisition and collaborative preprocessing: Simultaneously acquire sub-meter resolution high-resolution remote sensing data, hyperspectral remote sensing data, and thermal infrared remote sensing data covering the same area, and perform collaborative preprocessing on the three types of data, including radiometric calibration, atmospheric correction, and geometric registration, so that they have consistent spatial reference and comparable spectral information. Step 2: Multi-dimensional feature extraction: Based on the three types of data after collaborative preprocessing, features representing different lithological attributes are extracted respectively; Step 3, Feature-level fusion: The multidimensional features extracted from different data sources in Step 2 are standardized to eliminate differences in scale. Then, they are concatenated along the feature dimensions to construct a unified, high-dimensional fusion feature vector set. Step 4, Intelligent Lithology Classification: The fused feature vector set obtained in Step 3 is used as input and imported into a pre-trained machine learning model for automatic identification and classification of lithology types, and a preliminary lithology classification map is output. Step 5: Post-processing optimization of classification results: Spatial filtering is performed on the preliminary lithology classification map obtained in Step 4 to eliminate isolated noise points in the map and smooth the boundaries of lithology units, ultimately generating a high-precision lithology mapping result with good spatial continuity and optimized mapping effect.
2. The intelligent lithological mapping method based on multi-source remote sensing data and machine learning according to claim 1, characterized in that, In step two, spatial texture and morphological features are extracted from sub-meter resolution high-resolution remote sensing data, including: ① The panchromatic or multispectral bands of the sub-meter high-resolution remote sensing data are processed using a multi-scale, multi-directional Gabor filter bank; ② Extract texture features that reflect the spatial structure, striations, joints, and porphyritic texture of rocks; ③ Simultaneously, an image segmentation algorithm is used to segment independent rock mass objects from the data and calculate their morphological characteristics.
3. The intelligent lithological mapping method based on multi-source remote sensing data and machine learning according to claim 1, characterized in that, In step two, mineral quantification features are extracted from hyperspectral remote sensing data. This is done using a physical model-based spectral analysis method. Specifically, a modified Gaussian model is used to perform least-squares fitting on the characteristic absorption valleys in the hyperspectral data to invert and obtain physical parameters of the spectral absorption features that characterize the mineral composition. These physical parameters include at least the depth, width, area, and symmetry of the absorption features. These parameters are used as feature vectors for quantifying mineral composition.
4. The intelligent lithological mapping method based on multi-source remote sensing data and machine learning according to claim 3, characterized in that, The physical model-based spectral analysis method is specifically implemented using the Teracorder algorithm framework to achieve quantitative inversion of mineral spectral characteristics, replacing the traditional qualitative comparison method of spectral angle mapping.
5. The intelligent lithological mapping method based on multi-source remote sensing data and machine learning according to claim 1, characterized in that, In step two, thermal physical features are extracted from the thermal infrared remote sensing data, including: ① Based on multiple thermal infrared bands of ASTER thermal infrared data, the emissivity spectrum and surface temperature of the Earth's surface are obtained by inversion using the radiative transfer equation; ② The focus is on extracting the thermal infrared emissivity spectral shape characteristics that reflect the lithological information of silicate minerals in the 8-12μm wavelength range.
6. The intelligent lithological mapping method based on multi-source remote sensing data and machine learning according to claim 1, characterized in that, The feature-level fusion in step three involves standardizing the spatial texture and morphological features extracted from sub-meter high-resolution data, the mineral quantification features extracted from hyperspectral data, and the thermophysical features extracted from thermal infrared data, and then splicing them along the feature channel dimension to form an ultra-high-dimensional fused feature vector that integrates spatial, mineral composition, and thermophysical properties, providing complementary input data for machine learning models.
7. The intelligent lithological mapping method based on multi-source remote sensing data and machine learning according to claim 1, characterized in that, The machine learning model in step four is an ensemble learning model or a deep learning model. During the training phase, the machine learning model is trained using samples of known lithological categories and their corresponding fused feature vectors. The model is then tuned through hyperparameter optimization experiments to address the high-dimensional and nonlinear characteristics of the lithological data, enabling the model to automatically learn the complex rules of lithological classification without relying on manual feature selection.
8. The intelligent lithological mapping method based on multi-source remote sensing data and machine learning according to claim 1, characterized in that, The spatial filtering process in step five specifically employs majority voting or conditional random field algorithms to improve the spatial coherence of the preliminary classification results, making the boundaries of the final lithological mapping results smoother and the distribution of land features more reasonable.
9. The intelligent lithological mapping method based on multi-source remote sensing data and machine learning according to claim 2, characterized in that, The sub-meter high-resolution remote sensing data mentioned in step one is data from the GF-2 domestic satellite with a spatial resolution of 0.8 meters, and the thermal infrared remote sensing data is data from five thermal infrared bands of the ASTER satellite.
10. A smart lithological mapping method based on multi-source remote sensing data and machine learning according to any one of claims 1-9, characterized in that, The above method constructs an end-to-end automated process from multi-source data input to lithological mapping results generation, which significantly reduces the reliance on the manual experience of geological experts in traditional methods and achieves high-precision and high-efficiency identification and mapping of lithologies with similar spectral characteristics.