A method for optimizing the bright temperature characteristics of lunar basalts and classifying geological units

Abstract

The application discloses a kind of lunar basalt bright temperature feature optimization and geological unit classification method, comprising: obtaining the characteristic image of lunar surface different frequency noon bright temperature and midnight bright temperature, and extracting bright temperature difference feature;Combining artificial interpretation and existing geological map calibration research sample;Propose a kind of bright temperature feature optimization method;Optimized bright temperature feature is processed by principal component analysis dimension reduction, and the redundancy of bright temperature feature is reduced;Supervised classification is carried out to the geological structure in the dimension-reduced bright temperature data using random forest model, and the geological structure diagram is obtained;Bright temperature feature is recombined according to frequency and acquisition time and bright temperature difference respectively, and supervised classification is carried out respectively, to further clarify the specific role of each bright temperature feature in the classification process.The method effectively fills the blank of the existing lunar bright temperature feature on basalt geological unit classification and the contribution of the research, and provides a unified guidance research scheme for the optimization method of lunar bright temperature feature.

Description

A method for optimizing the brightness temperature characteristics of lunar basalts and classifying geological units. Technical Field

[0001] This invention belongs to the field of microwave remote sensing, and in particular relates to a method for analyzing and classifying microwave radiometer data in the field of lunar exploration. Background Technology

[0002] Lunar remote sensing technology is of great significance for understanding the lunar geological structure, revealing the origin of the Moon, and detecting the mineral composition of the lunar surface. Microwave radiometers (MRMs) are a passive microwave remote sensing technology that can acquire brightness temperature information from different depths of the lunar surface, reflecting the microwave thermal radiation of the lunar surface and subsurface. Numerous studies have shown that compared to traditional optical remote sensing methods, lunar surface brightness temperature can more effectively reveal the deep geological structure and mineral composition of the Moon, as well as detect anomalous physical and chemical phenomena on the lunar surface. Therefore, microwave radiometer brightness temperature data is an effective means of probing the lunar geological structure and the mineral composition of the lunar regolith.

[0003] However, current research on lunar microwave remote sensing largely relies on manual interpretation, consuming significant time and resources. Furthermore, microwave remote sensing data often involves multiple channels with different acquisition times and electromagnetic frequency bands. Manual interpretation methods struggle to comprehensively consider the effective information from each channel, making it difficult to discover implicit correlations and interactions between channels. There is also a lack of effective evaluation and calculation methods for information redundancy between channels. Moreover, there is currently a lack of effective quantitative evaluation methods for the observational capabilities of features in each channel on the lunar surface. Although machine learning feature engineering methods have been introduced into the field of remote sensing to select and optimize the classification contribution of different features, the complexity of machine learning models often makes it difficult for traditional single feature optimization methods to comprehensively characterize the observational and classification capabilities of a specific feature for surface materials.

[0004] Therefore, in the field of lunar microwave remote sensing geological observation, proposing a feature optimization and geological unit classification method is an urgent problem to be solved. Summary of the Invention

[0005] The purpose of this invention is to provide a method for optimizing the brightness temperature characteristics of lunar basalt and classifying geological units. This method can effectively select and optimize various features in lunar brightness temperature data, and realize automated mapping of lunar basalt geological units.

[0006] This method mainly includes the following modules:

[0007] I. Brightness Temperature Feature Acquisition

[0008] Obtain brightness temperature images of the lunar surface at noon and midnight using microwave radiometers at various frequency bands, and obtain brightness temperature difference images at noon and midnight.

[0009] II. Labeling Research Samples

[0010] Based on brightness temperature images (or optical remote sensing images and mineral abundance data), a preliminary geological classification is made using visual interpretation methods. Areas whose geological categories can be determined are marked and designated as research samples of the corresponding geological categories for subsequent experimental analysis.

[0011] III. Brightness-Temperature Feature Optimization Method

[0012] The correlation between features and between geological categories was calculated using the Pearson coefficient. Two distance measures (normalized distance and JS divergence) were used to calculate the class separability of each feature. Two machine learning feature evaluation methods (ReliefF and OOB importance) were used to calculate the classification contribution of each feature. Based on these calculated indicators, the most valuable brightness temperature feature combination was determined.

[0013] IV. Feature Dimensionality Reduction

[0014] Principal component analysis (PCA) was used to reduce the dimensionality of the original brightness temperature features, thereby reducing feature redundancy. The number of principal components was determined by the cumulative contribution rate of the principal component variance.

[0015] V. Classification of Geological Units for Supervision

[0016] Based on the optimal feature combination and a certain number of principal components, a random forest classifier is used to perform supervised classification of the study area to obtain a geological structure map.

[0017] VI. Analysis of Characteristic Combinations in Ablation Experiments

[0018] A feature optimization and visualization method based on ablation experiments is designed: features are combined according to brightness temperature frequency, acquisition time and brightness temperature difference, and a random forest (RF) model is used for supervised classification for each feature combination to obtain the corresponding geological structure map, further clarifying the role of brightness temperature features in geological classification.

[0019] In summary, this invention proposes a method for optimizing the brightness temperature characteristics of lunar basalts and classifying geological units. This method effectively selects and optimizes brightness temperature features, reduces redundant information in brightness temperature data, and enables supervised classification and automated mapping of lunar geological units. This invention provides a guiding and universal method for lunar brightness temperature data processing and analysis.

[0020] The beneficial effects of this invention include:

[0021] (I) This invention proposes a feature optimization method for lunar basalt brightness temperature data. Compared to traditional single feature evaluation methods, this comprehensive evaluation system can analyze brightness temperature features from multiple perspectives, including feature redundancy, inter-class distance, and contribution to the classification of supervised classifiers based on different principles. The optimization method proposed in this invention can effectively analyze the specific role and contribution of each brightness temperature feature in geological structure classification. Currently, lunar brightness temperature feature optimization is still a research gap, and research in this field is very insufficient. This invention is the first feature optimization method for lunar surface brightness temperature data.

[0022] (ii) This invention uses a supervised learning method to classify lunar geological structures in brightness temperature data. Compared with existing manual interpretation methods, this method can achieve an adaptive learning process and can comprehensively consider the effective information in each brightness temperature channel.

[0023] (III) This invention employs feature dimensionality reduction technology to reduce the dimensionality of features extracted from brightness temperature data. Brightness temperature data often contains multi-channel information, including a large amount of redundant and invalid information. Feature dimensionality reduction can effectively enhance the effective information in each feature and improve the accuracy of geological structure classification.

[0024] (iv) This invention introduces an ablation experiment method for the first time to select brightness temperature features. Features with the same properties are combined and classified. Through visualization analysis of the classification results, the advantages and disadvantages of each combination of brightness temperature features for each geological category can be effectively analyzed. Attached Figure Description

[0025] To more clearly illustrate the specific technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be shown below.

[0026] Figure 1 is a flowchart of the present invention;

[0027] Figure 2 is a rich image of optical and microwave brightness temperature of the sea area provided by an embodiment of the present invention;

[0028] (a) Optical image (b) Microwave brightness temperature image

[0029] Figure 3 shows the rich marine geological classification results provided by the embodiments of the present invention;

[0030] (a) Optimization features (b) Dimensionality reduction features

[0031] Figure 4 shows the rich marine geological structure mapping based on frequency feature combinations provided in an embodiment of the present invention;

[0032] (a)3.0GHz(b)7.8GHz(c)19.35GHz(d)37.0GHz

[0033] Figure 5 shows a rich marine geological structure map provided by an embodiment of the present invention, based on the combination of acquisition time and brightness temperature difference features.

[0034] (a) Noon brightness temperature characteristic combination (b) Midnight brightness temperature characteristic combination (c) Brightness temperature difference characteristic combination. Detailed Implementation

[0035] To better explain the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0036] This invention proposes a method for optimizing the brightness temperature characteristics of lunar basalts and classifying geological units, as shown in Figure 1. This embodiment uses brightness temperature data from the Chang'e-2 microwave radiometer in the Mare Fecunditatis basalt basin as an example to classify the phases and material composition of lunar mare basalts, and selects and optimizes brightness temperature characteristics to determine their influence and role in the classification of basalt phases and material composition. A schematic diagram of the optical image of the Mare Fecunditatis region is shown in Figure 2(a). It should be noted that the Chang'e-2 microwave radiometer data used in this embodiment is divided into four working frequency bands, two working times (noon and midnight), and the brightness temperature difference between noon and midnight for each frequency band, totaling 12 features. Specific parameters are shown in Table 1.

[0037] Table 1. Parameters of the microwave radiometer used in this embodiment.

[0038] Channel 1, Channel 2, Channel 3, Channel 4; Operating frequencies: 3.0GHz, 7.8GHz, 19.35GHz, 37.0GHz; Bandwidth: 100MHz, 200MHz, 500MHz, 500MHz; Penetration depth: 1m-2m, 38.5cm-75cm, 15.5cm-31cm, 8.1cm-16.2cm. surface

[0039] The specific implementation method of the present invention is as follows:

[0040] (I) Brightness Temperature Feature Acquisition

[0041] The brightness temperature difference characteristics are obtained by subtracting the noon brightness temperature characteristics of each operating frequency band from the midnight brightness temperature characteristics of the corresponding frequency:

[0042]

[0043] in and dT represents the brightness temperature characteristics at noon and midnight, respectively. B The brightness temperature difference characteristics are shown in Figure 2(b). A schematic diagram of the brightness temperature difference characteristics at 37.0 GHz is shown in Figure 2(b).

[0044] (ii) Labeling research samples

[0045] This embodiment combines visual interpretation with the geological structure map of Mare Fecunditatis proposed by Kramer et al. (from the literature "Kramer, GY; Jolliff, BL; Neal, CR: Searching for high alumina mare basalts using Clementine UVVIS and Lunar Prospector GRS data: Mare Fecunditatis and Mare Imbrium. Icarus. 2008, 198, 7-18.") to mark the research samples. The geological structure map of Mare Fecunditatis proposed by Kramer et al. divides Mare Fecunditatis into five units: Ihtm (Late? Immura, medium-high titanium, high-iron basalt), Chtr (extended region of high titanium basalt), Iltm (Immura, low titanium, medium-iron basalt), Im (Early? Immura, very low titanium, low-iron basalt), and Cc (impact crater ejecta). This embodiment also refers to this geological map for marking the research samples.

[0046] (III) Brightness Temperature Feature Optimization Method

[0047] Pearson coefficients were used to calculate feature correlation and category correlation, respectively. If two features are highly correlated (coefficient close to 1 or -1), it indicates redundancy between features. In this embodiment, the abundant marine brightness temperature features of the selected area contain a large amount of redundant information, highlighting the necessity of feature dimensionality reduction techniques in subsequent steps. In the category correlation analysis of this embodiment, there is a high correlation between Iltm, Im, and Cc, indicating that these three geological units are difficult to distinguish. The inter-category correlation is calculated in the order Ihtm → Chtr → Iltm.

[0048] The sequence →Im→Cc decreases monotonically, which is highly consistent with the eruption periods and igneous processes of lunar mare basalts.

[0049] Normalized distance is used to visually represent the Euclidean distance between two classes and eliminates the influence of variance. In this embodiment, the noon brightness temperature feature is suitable for classifying most basaltic geological structures except for Iltm, Im, and Cc. For the three geological units Iltm, Im, and Cc, the midnight brightness temperature feature is suitable for distinguishing them. However, the midnight brightness temperature feature has poor distinguishing ability for geological structures other than these three geological units. The brightness temperature difference feature is not suitable for distinguishing the Iltm, Im, and Cc geological structures, but is very suitable for classifying other categories. Furthermore, among all features, the normalized distance value between Ihtm and Chtr is smaller than the distance values ​​between other classes, indicating that they are not easily distinguishable.

[0050] JS divergence characterizes the degree of separation in a statistical histogram; a larger divergence indicates less confusion between the two classes and stronger separability. In this embodiment, the JS divergence values ​​for most categories corresponding to the noon brightness temperature feature are higher than those for the midnight brightness temperature feature, further demonstrating the effectiveness of the noon brightness temperature feature. The midnight brightness temperature feature shows significantly better separation for Cc-Im and Cc-Iltm than the noon brightness temperature and brightness temperature difference features. The brightness temperature difference feature demonstrates a unique advantage in distinguishing most basalt categories other than Cc, Im, and Iltm.

[0051] ReliefF is a filtering selection method that calculates contribution based on the average Euclidean distance. In this embodiment, the number of nearest neighbor data points is set to 10. ReliefF results show that, compared to the midnight brightness temperature feature, the midday brightness temperature feature is suitable for distinguishing Ihtm-Chtr, Ihtm-Iltm, Ihtm-Cc, Ihtm-Im, Chtr-Cc, Chtr-Iltm, and Chtr-Im. The midnight brightness temperature feature has an advantage in classifying Cc-Im, Cc-Iltm, and Im-Iltm. The brightness temperature difference feature is suitable for classifying most categories. The overall feature contribution of the midday brightness temperature feature is higher than that of the midnight brightness temperature feature, indicating that the midday brightness temperature feature can effectively distinguish more categories.

[0052] Out-of-Band (OOB) importance is a wrap-around selection method, a step in a random forest classifier, that indirectly assesses feature importance by adding random perturbations and observing their impact on the classifier. In this embodiment, in most cases, the pairwise OOB importance value of the midnight brightness temperature feature is equal to 0. However, the overall feature OOB importance value of the midnight brightness temperature feature is higher than that of the noon brightness temperature feature. This peculiarity stems from the unique advantage of the midnight brightness temperature feature in distinguishing the three geological structures: Iltm, Im, and Cc.

[0053] Based on the results of the above indicators, considering that the inter-class distance and classification contribution of the midnight brightness temperature features of 7.8GHz and 19.35GHz are low, and there is a large redundancy among the brightness temperature difference features, in this embodiment, four features, namely the midnight brightness temperature features and brightness temperature difference features of 7.8GHz and 19.35GHz, are filtered out, and the other eight features are retained.

[0054] (iv) Feature Dimensionality Reduction

[0055] Principal component analysis was performed on the eight brightness temperature features after feature optimization to obtain a new feature space distribution. In this embodiment, the cumulative contribution rate of the principal component variance reached 99.6929% in the first three principal components; therefore, the first three principal components were determined as the number of principal components in this embodiment. In subsequent embodiments, the data from the first three principal components were selected for geological classification and mapping.

[0056] (V) Classification of Supervision

[0057] A random forest model was used to classify basalt geology, resulting in a geological structure map of the rich sea area. In this embodiment, the number of trees in the random forest was set to 125. The original feature classification results and the feature classification results after dimensionality reduction are shown in Figure 3. Clearly, the Im geological units on the Hainan side of the rich sea area were more accurately identified after feature dimensionality reduction, while the Langrenus impact crater area was not identified as Im geological units.

[0058] (VI) Analysis of Feature Combinations in Ablation Experiments

[0059] To further analyze the role of features in geological classification, this embodiment employs an ablation experiment for feature combination analysis. Geological classification and mapping were performed using 3.0 GHz feature combinations (3.0 GHz brightness temperature feature at noon, 3.0 GHz brightness temperature feature at midnight, and 3.0 GHz brightness temperature difference feature, hereinafter the same), 7.8 GHz feature combinations, 19.35 GHz feature combinations, and 37.0 GHz feature combinations, as shown in Figure 4. In the results of the 3.0 GHz and 7.8 GHz feature combinations, some Chtr and Iltm were misclassified as Cc, and the Taruntius crater was almost completely misclassified. However, the classification performance of the 19.35 GHz and 37.0 GHz features was generally better than that of the 3.0 GHz and 7.8 GHz features. From the feature evaluation indicators, the contribution of frequency was not significantly different, but in this ablation experiment, the classification effect corresponding to high-frequency features was found to be the best.

[0060] Geological classification and mapping were performed by combining noon brightness temperature features, midnight brightness temperature features, and brightness temperature difference features, as shown in Figure 5. The noon brightness temperature feature combination Cc showed confusion with Iltm, but other categories showed good classification results. The midnight brightness temperature feature combination Cc, Im, and Iltm showed good classification results, but Iltm and Chtr units showed large-scale misclassification. The dTB feature Cc showed confusion with Im. Therefore, each brightness temperature feature has a unique role in enriching marine mapping.

[0061] The data, research area, and various parameters involved in this method need to be adjusted according to actual applications. The embodiments described above are only used to illustrate the technical solutions of the present invention and are not intended to strictly limit them. It should be particularly noted that simple modifications, substitutions, and adjustments made by those skilled in the art to the technical solutions in the foregoing embodiments still fall within the protection scope of the present invention. For example, the research area, the type and load mode of the microwave radiometer, the corresponding operating frequency band and data acquisition time, the specific categories and classification standards of geological structures, the parameter settings in each algorithm, and the selection of feature dimensions involved in this embodiment are not strictly limited. Any inventive solutions proposed without departing from the spirit and essence of the present invention should fall within the protection scope of the present invention.

Claims

1. A method for optimizing the brightness temperature characteristics of lunar basalts and classifying geological units, characterized in that, It includes the following modules:

1. Brightness temperature feature acquisition: acquires brightness temperature images of the lunar surface at noon and midnight using microwave radiometers at various frequency bands, and obtains brightness temperature difference images at noon and midnight; 2. Marking research samples: based on the brightness temperature images, uses visual interpretation to preliminarily classify geological categories, and marks areas where the geological category can be determined, designating them as research samples of the corresponding geological category; III. Brightness temperature feature optimization method: Pearson coefficient is used to calculate the correlation between features and the correlation between geological categories. Two distance measures, normalized distance and JS divergence, are used to calculate the class separability of each feature. Two machine learning feature evaluation methods, ReliefF and OOB importance, are used to calculate the classification contribution of each feature. Based on the above calculation indicators, the most effective combination of brightness temperature features is determined. IV. Feature Dimensionality Reduction: Principal Component Analysis (PCA) is used to reduce the dimensionality of the original brightness temperature features, thereby reducing feature redundancy. The cumulative contribution rate of principal component variance is used to determine the number of principal components. V. Supervised Classification of Geological Units: Based on the optimal feature combination and a selected number of principal components, a random forest classifier is used to perform supervised classification of the study area, resulting in a geological structure map. VI. Ablation Experiment Method: Feature Combination Analysis: Features are combined according to brightness temperature frequency, acquisition time, and brightness temperature difference. For each feature combination, a random forest model is used for supervised classification to obtain the corresponding geological structure map, clarifying the role of brightness temperature features in geological classification.

2. The method according to claim 1, characterized in that: Microwave radiometer data was used, divided into four operating frequency bands, two operating times (noon and midnight), and the brightness-temperature difference between noon and midnight for each frequency band, totaling 12 characteristics; specific parameters are shown below: Specifically, the following is an example: (I) Brightness temperature feature acquisition: The brightness temperature difference feature is obtained by subtracting the noon brightness temperature feature of each working frequency band from the midnight brightness temperature feature of the corresponding frequency. in and dT represents the brightness temperature characteristics at noon and midnight, respectively. B Indicates the characteristics of bright temperature difference; (II) Labeling the research samples: The rich sea area is divided into five units: Ihtm (Late Irrigation Martian, medium-high titanium, high-iron basalt), Chtr (extended region of high titanium basalt), Iltm (Irrigation Martian, low titanium, medium-iron basalt), Im (Early Irrigation Martian, extremely low titanium, low-iron basalt), and Cc (impact crater ejecta); (III) Brightness temperature feature optimization method: Four features, namely the midnight brightness temperature feature and brightness temperature difference feature at 7.8 GHz and 19.35 GHz, are filtered out, leaving the other eight features; (IV) Feature dimensionality reduction: Principal component analysis is performed on the eight brightness temperature features after feature optimization to obtain a new feature space distribution; the cumulative contribution rate of the principal component variance is... The first three principal components reached 99.6929%, so the number of principal components was determined to be the first three; the data of the first three principal components were selected for geological classification and mapping; (V) Supervised classification: the random forest model was used to classify basalt geology and obtain rich marine geological structure maps; (VI) Ablation experiment method feature combination analysis: the ablation experiment method was used for feature combination analysis; geological classification and mapping were carried out using 3.0 GHz feature combination, namely noon 3.0 GHz brightness temperature feature, midnight 3.0 GHz brightness temperature feature and 3.0 GHz brightness temperature difference feature, 7.8 GHz feature combination, 19.35 GHz feature combination and 37.0 GHz feature combination.

Images