Mining area land utilization type classification method based on random forest and storage medium

Abstract

The invention discloses a mining area land utilization type classification method based on a random forest and a storage medium, and the method comprises the steps: obtaining remote sensing images of a target region at different time points, the remote sensing images being multispectral remote sensing images, and the target region comprising a mining area; respectively calculating a remote sensing index corresponding to each remote sensing image, and generating a feature image set according to each remote sensing image, the corresponding remote sensing index and the digital elevation image of the target area; constructing a classification model based on a random forest algorithm according to the feature image set, and training the classification model; and obtaining a to-be-identified remote sensing image of the target area, and performing identification classification on the to-be-identified remote sensing image through the trained classification model to obtain a classification result, the classification result including the land utilization type of the target area. According to the invention, the land utilization type classification precision can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of land use classification, and in particular to a random forest-based mining area land use type classification method and storage medium. Background Art

[0002] As the global demand for mineral resources increases, mining activities are gradually expanding. However, large-scale mining activities have had a profound impact on the surrounding environment, ecosystems and land use patterns, causing a series of problems such as environmental pollution, biodiversity decline and land degradation. Therefore, monitoring of mining areas and their surrounding environments is of great significance, which not only helps to rationalize resource development, but also provides data support for environmental protection and ecological restoration.

[0003] In recent years, some studies have delineated mining areas through manual visual interpretation based on satellite images. These studies have several limitations: 1) they rely entirely on the experience of professionals and cannot guarantee the accuracy of database delineation; 2) the satellite images used for delineation cannot guarantee temporal consistency because these images are spliced ​​from remote sensing base maps from different periods, most of which come from Google Earth or Microsoft Bin.

[0004] In the research field of mining area land use remote sensing extraction and classification, traditional remote sensing image processing methods mainly include supervised classification and unsupervised classification. In the existing technology, commonly used classification algorithms such as maximum likelihood classification, support vector machine (SVM), K-means clustering, etc. have been widely used in land use classification and extraction. However, these methods have certain limitations when processing complex mining area remote sensing images and are difficult to meet the needs of practical applications, such as: Insufficient classification accuracy: Due to the complex and dynamic changes in the surface features of the mining area, there is uncertainty in the demarcation of the land use boundaries of the mining area based solely on manual visual interpretation, and the overall extraction accuracy cannot be guaranteed; and the traditional remote sensing classification method has limited accuracy when classifying remote sensing images of the mining area, and is easily affected by factors such as sample imbalance and noise interference.

[0005] Low computational efficiency: Methods such as SVM have high computational complexity when processing large-scale remote sensing images, resulting in a long classification process and making it difficult to meet the needs of large-scale mining area monitoring.

[0006] Insufficient generalization ability: Some traditional methods (such as maximum likelihood classification) have too strict assumptions about data and cannot effectively adapt to the complex and diverse land distribution characteristics of mining areas.

[0007] Lack of robustness: Unsupervised classification methods (such as K-means clustering) are sensitive to initial conditions, the classification results are unstable, and the ability to distinguish land types is weak in applications in complex mining areas. Summary of the invention

[0008] The technical problem to be solved by the present invention is to provide a mining area land use type classification method and storage medium based on random forest, which can improve the accuracy of land use type classification.

[0009] In order to solve the above technical problems, the technical solution adopted by the present invention is: a mining area land use type classification method based on random forest, comprising: Acquire remote sensing images of a target area at different time points, wherein the remote sensing images are multispectral remote sensing images, and the target area includes a mining area; Calculate the remote sensing index corresponding to each remote sensing image respectively, and generate a feature image set according to each remote sensing image and its corresponding remote sensing index and the digital elevation image of the target area; According to the feature image set, a classification model is constructed based on a random forest algorithm, and the classification model is trained; The remote sensing image to be identified is obtained, and the remote sensing image to be identified is identified and classified using a trained classification model to obtain a classification result, wherein the classification result includes the land use type of the target area.

[0010] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method described above is implemented.

[0011] The beneficial effects of the present invention are as follows: by calculating the remote sensing index and combining the multi-spectral bands of the remote sensing image, the remote sensing index and the elevation data as the multi-feature input of the classification model, these features are comprehensively applied to the classification process, which can improve the recognition ability of different land types and enhance the distinction between different land types; by constructing a classification model based on the random forest algorithm, it is possible to capture complex and nonlinear feature relationships and adapt to complex land distribution characteristics, thereby improving classification accuracy, and it has strong robustness, can more effectively resist the influence of noise and outliers, and enhance the stability and consistency of the classification results. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A flowchart of a method for classifying land use types in mining areas based on random forests according to the present invention; Figure 2 This is a flow chart of a method according to Embodiment 1 of the present invention. DETAILED DESCRIPTION

[0013] In order to explain the technical content, achieved objectives and effects of the present invention in detail, the following is an explanation in conjunction with the implementation modes and the accompanying drawings.

[0014] Please refer to Figure 1 , a classification method of land use types in mining areas based on random forest, including: Acquire remote sensing images of a target area at different time points, wherein the remote sensing images are multispectral remote sensing images, and the target area includes a mining area; Calculate the remote sensing index corresponding to each remote sensing image respectively, and generate a feature image set according to each remote sensing image and its corresponding remote sensing index and the digital elevation image of the target area; According to the feature image set, a classification model is constructed based on a random forest algorithm, and the classification model is trained; The remote sensing image to be identified is obtained, and the remote sensing image to be identified is identified and classified using a trained classification model to obtain a classification result, wherein the classification result includes the land use type of the target area.

[0015] From the above description, it can be seen that the beneficial effect of the present invention is that the accuracy of land use type classification can be improved.

[0016] Furthermore, after acquiring the remote sensing images of the target area at different time points, the method further includes: The remote sensing image is subjected to radiation correction, atmospheric correction and cloud removal processing.

[0017] From the above description, it can be seen that the quality of the image can be guaranteed by performing radiation correction, atmospheric correction and cloud removal.

[0018] Furthermore, generating a feature image set according to each remote sensing image and its corresponding remote sensing index and a digital elevation image of the target area includes: The remote sensing index corresponding to each remote sensing image is added to each remote sensing image as a remote sensing index feature, and the elevation data in the digital elevation image of the target area is added to each remote sensing image as an elevation feature to generate each feature image and obtain a feature image set.

[0019] From the above description, it can be seen that the subsequent training of the classification model can perform multi-dimensional feature selection and combination according to different remote sensing indices and terrain characteristics, ensuring that the classification model has higher accuracy when dealing with complex and diverse land use types.

[0020] Furthermore, the step of constructing a classification model based on a random forest algorithm according to the feature image set and training the classification model includes: Randomly sampling the feature image set with replacement to obtain at least one training sample subset; Constructing corresponding decision trees according to each training sample subset, and training each decision tree according to the training sample subset corresponding to each decision tree; When a node of a decision tree is split, a preset number of features are randomly selected from the features to be selected as a feature subset, and based on the Gini index, the best splitting feature is determined in the feature subset; the features to be selected include spectral band features, remote sensing index features and elevation features, and the remote sensing index features include vegetation features, water body features and bare soil features.

[0021] From the above description, we can see that by constructing multiple decision trees and integrating their results, we can capture complex, nonlinear feature relationships, adapt to complex land distribution characteristics, and thus improve classification accuracy. It is also highly robust to the errors of a single decision tree, and can more effectively resist the influence of noise and outliers, thus enhancing the stability of classification results. At the same time, the optimal feature is automatically selected for decision making when each node is split, which can automatically process high-dimensional data and use a combination of remote sensing index and elevation data to improve classification efficiency and accuracy.

[0022] Furthermore, the method further comprises: building a classification model based on a random forest algorithm according to the feature image set, and training the classification model; The accuracy of the trained classification model is evaluated through the validation sample set.

[0023] Furthermore, the evaluation index of the accuracy assessment includes at least one of a confusion matrix, an overall accuracy and a Kappa coefficient.

[0024] From the above description, it can be seen that through accuracy evaluation, the classification accuracy of the classification model is ensured to reach the expected standard.

[0025] Furthermore, it also includes: Perform a visual display based on the classification results.

[0026] From the above description, it can be seen that support can be provided for the dynamic monitoring and management of land use in the target area.

[0027] Furthermore, the remote sensing index includes a normalized difference vegetation index, a normalized difference water index, an enhanced vegetation index, a soil exposure index and a building index; and the land use type includes at least one of vegetation, water bodies, bare land and building areas.

[0028] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method described above is implemented.

[0029] Embodiment 1 Please refer to Figure 2Embodiment 1 of the present invention is: a mining area land use type classification method based on random forest, which can be applied to mining area environmental monitoring, land resource management, ecological environment protection and mining area development planning, and is particularly suitable for dynamic monitoring and evaluation of mining area environment.

[0030] This embodiment takes the target area as a key mineral mining area as an example for explanation, and uses Google EarthEngine (GEE) as a data processing platform, using its powerful cloud computing capabilities and global multi-temporal satellite image library to perform large-scale, high-resolution land use extraction of the mining area. In other optional embodiments, the target area can also be other application scenarios.

[0031] like Figure 2 As shown, the following steps are included: S1: Acquire remote sensing images of the target area at different time points. In this embodiment, the remote sensing images are Sentinel-2 multispectral remote sensing images.

[0032] Specifically, Sentinel-2 multispectral remote sensing images are acquired through the GEE platform, covering multiple time points in the mining area to ensure the timeliness and coverage of the data.

[0033] Furthermore, the remote sensing images are subjected to radiation correction, atmospheric correction and cloud removal to ensure the image quality.

[0034] S2: Calculate the remote sensing index corresponding to each remote sensing image respectively, and generate a feature image set according to each remote sensing image and its corresponding remote sensing index and the digital elevation image of the target area.

[0035] This embodiment is based on Sentinel-2 multispectral image data and combines a series of remote sensing indices to extract the characteristics of the mining area and the surrounding environment and generate high-precision classification data.

[0036] Among them, remote sensing indices include NDVI (Normalized Difference Vegetation Index), NDWI (Normalized Difference Water Index), EVI (Enhanced Vegetation Index), BSI (Bare Soil Index) and IBI (Index-Based Built-up Index). Each index targets specific surface features and can effectively distinguish between mining areas, vegetation, water bodies, soil and artificial surfaces. The following are the definitions, calculation methods, bands used and their uses of each remote sensing index.

[0037] 1. NDVI (Normalized Difference Vegetation Index) The calculation formula is:

[0038] Among them, NIR represents the reflectivity of the near-infrared light band, and Red represents the reflectivity of the red light band.

[0039] Usage: NDVI is the most commonly used vegetation index, mainly used to identify and monitor the density and health of surface vegetation. Higher NDVI values ​​indicate good vegetation cover and higher chlorophyll content. For remote sensing classification of mining areas, NDVI can distinguish mining areas from surrounding areas with dense vegetation, helping to identify ecological differences inside and outside mining areas.

[0040] 2. NDWI (Normalized Difference Water Index) The calculation formula is:

[0041] Among them, Green represents the reflectivity of the green light band.

[0042] Application: NDWI is mainly used for water body detection. It uses a combination of green and near-infrared bands to highlight the reflection characteristics of water bodies and suppress other ground objects such as vegetation and exposed soil. For mining areas, NDWI can effectively extract the characteristics of water bodies around the mining area to monitor the impact of mining activities on water resources.

[0043] 3. EVI (Enhanced Vegetation Index) The calculation formula is:

[0044] Among them, Blue represents the reflectivity of the blue light band.

[0045] Application: Compared with NDVI, EVI is more suitable for areas with high-density vegetation, has a higher dynamic range, and can reduce atmospheric influence. EVI can effectively capture the distribution of high-density vegetation inside and outside the mining area, and is suitable for monitoring the ecological impact around the mining area.

[0046] 4. BSI (Bare Soil Index) The calculation formula is:

[0047] Among them, SWIR1 represents the reflectivity of the shortwave infrared band (SWIR-1).

[0048] Application: BSI is used to identify exposed soil and has a good ability to distinguish exposed soil areas in mining areas. The exposed areas in mining areas are large, and the surrounding environment often has a high vegetation coverage. BSI can clearly separate the exposed mining area and the surrounding environment, providing an important basis for the identification and classification of mining activities in mining areas. BSI is more stable and versatile, suitable for a wide range of bare soil detection needs, especially in areas with many types of objects and complex changes.

[0049] 5. IBI (Index-Based Built-up Index) The calculation formula is:

[0050] Use: The IBI index is used to identify buildings or artificial surfaces, and can effectively distinguish between buildings, infrastructure and other non-natural features in mining areas. The index is suitable for analyzing changes and expansions in mining infrastructure, and helps understand the potential impact of mining development activities on the regional environment. Based on a more complex index combination, not only short-wave infrared and near-infrared are used, but also visible green light bands are combined. By introducing multiple ratios, IBI reduces the interference of vegetation and water bodies, and can more accurately distinguish between built-up areas and natural surfaces, especially in complex urban environments. The complex calculation method of IBI increases the sensitivity to buildings, and is more suitable for fine urban classification or studying the distribution of complex features. Under the requirements of high-resolution imagery and fine classification, IBI usually performs better than NDBI.

[0051] After each index is calculated, it is added to the Sentinel-2 multi-band image to form multiple feature layers including vegetation, water, soil and artificial surfaces. NASA's digital elevation image (DEM) data is then added to the image to further enhance the recognizability of terrain features. The final image is composed of the following bands: Sentinel-2 band: B2(Blue), B3(Green), B4(Red), B5(Red Edge 1), B6(Red Edge2), B7(Red Edge 3), B8(NIR), B8A(Narrow Near Infrared), B11(SWIR1), B12(SWIR2) Remote sensing index: NDVI, NDWI, EVI, BSI, IBI Elevation data: DEM After these bands and indices are cropped and synthesized, the resulting feature images are used to train a random forest model to classify the mining area and the surrounding environment.

[0052] By calculating and combining the above remote sensing indices, we can effectively distinguish different types of land cover in the mining area and its surrounding areas, and significantly improve the classification accuracy. Especially in the training stage of the random forest classifier, the added remote sensing features can significantly enhance the recognition ability of different land surface types, providing strong data support for mining area monitoring and environmental impact analysis.

[0053] S3: According to the feature image set, a classification model is constructed based on a random forest algorithm, and the classification model is trained.

[0054] This embodiment uses the random forest algorithm as a classifier to train the extracted remote sensing index data. In the process of feature quantity processing, the random forest uses bagging sampling and random feature selection in node splitting to give full play to the role of remote sensing feature quantities.

[0055] Specifically, bagging sampling is performed on the feature image set to obtain multiple training sample subsets, and then corresponding decision trees are constructed according to each training sample subset, and each decision tree is trained according to the training sample subset corresponding to each decision tree. In other words, the classification model in this embodiment includes multiple decision trees, each decision tree will output a prediction result, and the classification model takes the class that accounts for the majority of the prediction results of all decision trees as the final prediction result.

[0056] Among them, Bagging sampling (Bootstrap aggregating, also known as bagging algorithm) refers to randomly extracting N samples from the original training data set with replacement to form a subset (the size of each subset is equal to the original data set). During the sampling process, the same sample may be selected multiple times, while some samples may not be extracted.

[0057] By generating multiple different subsets of training samples through bagging, each decision tree sees different data, resulting in a unique training process for each tree. This data diversity reduces the model's dependence on individual samples and helps random forests cope with noise and imbalance problems in complex remote sensing data.

[0058] In each training sample subset, the multi-band and remote sensing index features of remote sensing images are repeatedly sampled. The sampling of a certain category (such as mining area) in each subset may show different responses on different features (for example, mining area may be different from soil in B2 and B8 bands), which allows different trees to capture various potential feature combinations of mining area and surrounding environment.

[0059] During the construction of each decision tree, each node does not consider all features, but randomly selects a certain number of feature subsets (assuming m features) from all features, and only searches for the best split feature in the subset. This random feature selection is repeated at each node of each tree to ensure that the tree has the opportunity to select different features at different split nodes.

[0060] Random feature selection ensures that various remote sensing feature quantities (such as bands, indices, DEM) are likely to become splitting bases at different nodes, allowing the model to identify more subtle surface differences. For example: Vegetation features (NDVI, EVI): At the initial node, the model may give priority to NDVI or EVI for splitting, so that the vegetation area can be separated from the non-vegetation area earlier. Especially in an environment with multiple vegetation types, these features can help the model quickly distinguish the differences in vegetation coverage inside and outside the mining area.

[0061] Water body feature (NDWI): During the random feature selection process, NDWI may be selected by some nodes and is particularly suitable for identifying water body features in mining areas and separating them from surface types such as mine pits and earth and rock piles.

[0062] Bare soil characteristics in mining areas (BSI, IBI): Bare soil and buildings are important features in mining area detection. Features such as BSI and IBI can be used in the random selection of split nodes to further refine the classification effect of mining areas by capturing the spectral characteristics of bare soil and mining area buildings.

[0063] This feature randomization not only prevents the model from being overly dependent on a certain feature, but also ensures the independence of each tree, so that in mining area detection, each tree can learn the surface information in remote sensing images from different perspectives. For example, some trees may rely more on NDVI and EVI split nodes to identify vegetation, while others may rely more on IBI and BSI to identify bare soil and mine pits.

[0064] Furthermore, in random forests, the splitting decision for each node is determined by the performance of the candidate features. Specifically, the model evaluates each randomly selected feature when splitting a node and selects the feature that maximizes the "purity".

[0065] In this embodiment, random forest evaluates node purity by calculating the Gini index (or information gain). A low Gini index means that the samples in the node are more likely to be of the same class, and the model reduces the complexity by splitting layer by layer.

[0066] Remote sensing features contribute differently to the Gini index at different nodes. For example, when the NDVI split node can clearly divide a part of the samples into vegetation categories, the Gini index will drop significantly, so the model prefers to choose the NDVI split node to achieve higher purity. At some nodes, using IBI combined with B8 band splitting may be more effective, because IBI reflects the distribution of mining areas and buildings, while the B8 band can further refine the differences between objects in different mining areas.

[0067] By selecting features at different nodes, each tree can split layer by layer, gradually parsing the high-dimensional features of remote sensing images into categories such as mining areas, vegetation, and water bodies. Since each tree is generated under bagging and feature randomization conditions, the diversity of split decisions makes the model more robust in mining area classification.

[0068] S4: Obtain the remote sensing image to be identified in the target area, and use the trained classification model to identify and classify the remote sensing image to obtain the classification result.

[0069] Specifically, through the trained classification model, GEE's cloud computing capabilities are used to classify and extract the land use types in the mining area on a large scale. The classification results cover the main land use types in the mining area, including vegetation, water bodies, bare land, and built-up areas.

[0070] S5: Visualize the classification results, that is, visualize the classification results and render the mining areas and non-mining areas in colors to provide support for dynamic monitoring and management of land use in mining areas.

[0071] Furthermore, an independent validation sample set is used to evaluate the accuracy of the classification model. In this embodiment, evaluation indicators such as confusion matrix, overall accuracy, and Kappa coefficient are used to ensure that the classification accuracy reaches the expected standard.

[0072] Specifically, a confusion matrix is ​​generated based on the classification results of the validation sample set and the actual category labels. The confusion matrix is ​​a matrix containing indicators such as classification accuracy and misclassification rate, which is used to evaluate the prediction performance of the model. In the confusion matrix, the number of correctly classified samples is located in the diagonal position, while the number of samples in the non-diagonal position indicates the misclassification of the model.

[0073] Overall Accuracy is the ratio of all correctly classified samples to the total samples in the confusion matrix. The calculation formula is: Overall Accuracy = Number of correctly classified samples / Total number of validation samples. This indicator represents the overall classification accuracy of the model for all categories and reflects the performance of the model on the entire test set.

[0074] Kappa coefficient is a standardized indicator for evaluating classification accuracy. It can quantify whether the classification accuracy of the model is better than random classification. The calculation formula is: Kappa coefficient = (p o -p e ) / (1-p e ), where p o indicates the observed overall accuracy, p e Indicates the expected random accuracy.

[0075] The value range of the Kappa coefficient is between -1 and 1. The closer the value is to 1, the better the classification effect. If it is close to 0, it means that there is no significant difference between the model and random classification. A negative value means that the classification result is even worse than random classification.

[0076] Producer's Accuracy indicates how many real samples of a certain category are correctly classified, which is used to reflect the recognition accuracy of the model for this category. User's Accuracy indicates how many samples of a certain category in the classification results are correct, which is used to reflect the reliability of the classification results of this category. These two accuracy indicators respectively represent the omission and misclassification of classification. By calculating the confusion matrix for each category, it can help identify which categories have better or worse classification effects.

[0077] Finally, the generated confusion matrix, overall accuracy, and Kappa coefficient are output in the Google Earth Engine platform and displayed in the console to facilitate the evaluation of the performance of the classification model.

[0078] Furthermore, the classified images are displayed using layers, and the classification results are compared with the actual surface conditions to further verify the accuracy of the classification model.

[0079] This embodiment has the following advantages: 1. Efficiency and automation: By utilizing the powerful computing power and cloud processing mechanism of the Google Earth Engine (GEE) platform, efficient and automated processing of remote sensing images of large-scale mining areas can be achieved without manual intervention, greatly improving the speed of classification and extraction. Compared with manual visual interpretation, this embodiment can quickly process large-scale data and complete land use extraction tasks in a short time.

[0080] 2. High classification accuracy: The random forest algorithm is used to make full use of the multispectral information of the image and a variety of remote sensing indices (such as NDVI, NDWI, EVI, BSI, IBI), and these features are comprehensively applied to the classification process to enhance the distinction between different land types. By constructing multiple decision trees, the random forest algorithm has stronger robustness and generalization ability, making the classification accuracy much higher than traditional methods and manual visual interpretation, especially showing obvious advantages when dealing with complex mining environments.

[0081] 3. Robustness and stability: The random forest algorithm reduces the sensitivity of a single decision tree to noise and abnormal data by constructing multiple decision trees. It has strong robustness and noise resistance, ensuring the stability and consistency of classification results. In addition, the GEE platform supports large-scale parallel computing to ensure that reliable classification results can be obtained in different mining areas and different time ranges.

[0082] 4. Ability to process multi-source data: It can flexibly combine multi-source data, including multi-spectral bands of Sentinel-2 images, various remote sensing indices (NDVI, NDWI, EVI, BSI, IBI) and digital elevation models (DEM). Through the multi-feature input of the random forest algorithm, it improves the ability to identify different land use types.

[0083] Embodiment 2 This embodiment is a computer-readable storage medium corresponding to the above embodiment, on which a computer program is stored. When the program is executed by the processor, it implements the various steps of a mining area land use type classification method based on random forest in the above embodiment, and can achieve the same technical effect, which will not be repeated here.

[0084] In summary, the present invention provides a mining area land use type classification method and storage medium based on random forest. By constructing a classification model based on the random forest algorithm, and using multispectral remote sensing images combined with a variety of remote sensing indexes and digital elevation models as multi-feature inputs of the classification model, efficient and accurate extraction of land use types can be achieved; not only the accuracy of remote sensing image classification is improved, but also the problems of low computational efficiency and insufficient classification accuracy of traditional methods when processing large-scale data are effectively solved. By efficiently and accurately classifying and extracting remote sensing image data, automatic identification and dynamic monitoring of mining area land use types can be achieved, providing scientific decision-making support for relevant departments. In addition, the present invention also has broad application prospects in the fields of mining area environmental impact assessment, land reclamation, resource exploration, and disaster warning.

[0085] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's specification and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for classifying land use types in mining areas based on random forests, characterized in that: include: Acquire remote sensing images of a target area at different time points, wherein the remote sensing images are multispectral remote sensing images, and the target area includes a mining area; Calculate the remote sensing index corresponding to each remote sensing image respectively, and generate a feature image set according to each remote sensing image and its corresponding remote sensing index and the digital elevation image of the target area; According to the feature image set, a classification model is constructed based on a random forest algorithm, and the classification model is trained; The remote sensing image to be identified is obtained, and the remote sensing image to be identified is identified and classified using a trained classification model to obtain a classification result, wherein the classification result includes the land use type of the target area.

2. The method for classifying mining area land use types based on random forest according to claim 1, characterized in that: After acquiring the remote sensing images of the target area at different time points, the method further includes: The remote sensing image is subjected to radiation correction, atmospheric correction and cloud removal processing.

3. The method for classifying mining area land use types based on random forest according to claim 1, characterized in that: The method of generating a feature image set according to each remote sensing image and its corresponding remote sensing index and a digital elevation image of a target area includes: The remote sensing index corresponding to each remote sensing image is added to each remote sensing image as a remote sensing index feature, and the elevation data in the digital elevation image of the target area is added to each remote sensing image as an elevation feature to generate each feature image and obtain a feature image set.

4. The method for classifying mining area land use types based on random forest according to claim 1, characterized in that: The step of constructing a classification model based on a random forest algorithm according to the feature image set and training the classification model includes: Randomly sampling the feature image set with replacement to obtain at least one training sample subset; Constructing corresponding decision trees according to each training sample subset, and training each decision tree according to the training sample subset corresponding to each decision tree; When a node of a decision tree is split, a preset number of features are randomly selected from the features to be selected as a feature subset, and based on the Gini index, the best splitting feature is determined in the feature subset; the features to be selected include spectral band features, remote sensing index features and elevation features, and the remote sensing index features include vegetation features, water body features and bare soil features.

5. The method for classifying mining area land use types based on random forest according to claim 1, characterized in that: The method further comprises: building a classification model based on a random forest algorithm according to the feature image set and training the classification model; The accuracy of the trained classification model is evaluated through the validation sample set.

6. The method for classifying mining area land use types based on random forest according to claim 5, characterized in that: The evaluation index of the accuracy assessment includes at least one of a confusion matrix, an overall accuracy and a Kappa coefficient.

7. The method for classifying mining area land use types based on random forest according to claim 1, characterized in that: Also includes: Perform a visual display based on the classification results.

8. The method for classifying mining area land use types based on random forests according to any one of claims 1 to 7, characterized in that: The remote sensing index includes a normalized difference vegetation index, a normalized difference water index, an enhanced vegetation index, a bare soil index and a building index; and the land use type includes at least one of vegetation, water, bare land and building areas.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images