Mine reclamation effect evaluation method based on multi-temporal remote sensing image

Abstract

The invention discloses a mine reclamation effect evaluation method based on a multi-temporal remote sensing image, and belongs to the field of mine reclamation effect evaluation, and the method comprises the steps: obtaining and preprocessing multi-source remote sensing data, and obtaining a standardized multi-temporal data set; performing image registration and radiation correction on the standardized multi-temporal data set to obtain a correction data set with consistent space and consistent radiation; constructing a multi-dimensional feature data set based on the correction data set, wherein the multi-dimensional feature data set comprises surface morphological features, vegetation health features, soil temperature features and underground water level change features; extracting a surface form feature vector according to the multi-dimensional feature data set, and generating a vegetation health distribution map, a soil temperature feature map and an underground water level change distribution map; and fusing the surface form feature vector, the vegetation health distribution map, the soil temperature feature map and the underground water level change distribution map, constructing a comprehensive index system, and outputting a mine reclamation effect evaluation result. According to the method, the accuracy and comprehensiveness of reclamation effect evaluation are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of mine reclamation effect evaluation technology, and in particular relates to a method for evaluating the effect of mine reclamation based on multi-temporal remote sensing images. Background Technology

[0002] Mine reclamation is a crucial link in ecological restoration, directly impacting the sustainable use of land resources and the restoration of the ecological environment. With increasing global emphasis on environmental protection, the scientific assessment of mine reclamation effectiveness has become an important means of ensuring the quality of ecological restoration. Monitoring mine reclamation progress using remote sensing technology can efficiently acquire large-scale surface information, providing data support for formulating restoration strategies. However, existing assessment methods have significant limitations when processing complex surface information, primarily manifested in the difficulty of integrating different types of remote sensing data and the challenge of comprehensive analysis of multi-dimensional indicators. These shortcomings often result in assessment results that fail to fully reflect the actual effects of mine reclamation, limiting the scientific guidance of restoration work.

[0003] In practical applications, mine reclamation involves changes in multiple aspects, including landform, vegetation health, soil temperature, and groundwater level. These elements are typically acquired from different types of remote sensing data (such as optical, radar, hyperspectral, and thermal infrared). The spatial resolution, temporal resolution, and radiometric characteristics of data from different sensors vary significantly, making image registration and radiometric correction challenging. For example, optical images may lose crucial information due to cloud cover, while radar images, although able to penetrate clouds, struggle to capture vegetation spectral characteristics. This data heterogeneity complicates the construction of a unified time-series database, thus affecting the collaborative analysis of multi-source data. Furthermore, even with data integration, extracting comprehensive indicators encompassing landform and vegetation health from multi-source data remains a technical bottleneck. For instance, in a mine reclamation project, while vegetation cover may appear to have increased, abnormal soil temperature could indicate a declining groundwater level; a single-dimensional indicator cannot reveal this potential ecological imbalance. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a method for evaluating the effectiveness of mine reclamation based on multi-temporal remote sensing images, thereby resolving the issues present in the existing technologies.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a method for evaluating the effectiveness of mine reclamation based on multi-temporal remote sensing imagery, comprising:

[0006] Multi-source remote sensing data for mine reclamation were acquired, including optical, radar, hyperspectral, and thermal infrared images. Cloud cover interference was eliminated through preprocessing to obtain a standardized multi-temporal dataset.

[0007] For standardized multi-temporal datasets, an image registration algorithm is used to adjust the spatial resolution based on feature point matching to obtain a spatially consistent registration dataset.

[0008] Radiometric characteristics are extracted from the registration dataset, and a radiometric correction algorithm is used to unify the radiometric standards of different sensors to obtain a radiometrically consistent correction dataset.

[0009] Based on the calibration dataset, a time series database is constructed, and the characteristics of landform, vegetation health, soil temperature and groundwater level changes are integrated to obtain a multidimensional feature dataset.

[0010] For multidimensional feature datasets, principal component analysis algorithm is used to extract the main patterns of landform changes and obtain landform feature vectors;

[0011] From the multidimensional feature dataset, the normalized vegetation index is used to calculate the vegetation health status, and combined with the landform feature vector, a vegetation health distribution map is obtained.

[0012] Based on the multidimensional feature dataset, the soil temperature distribution was analyzed using thermal infrared data, and the surface morphology feature vector and vegetation health distribution map were fused to obtain the soil temperature feature map.

[0013] If the soil temperature characteristic map shows abnormal areas, then the groundwater level change trend is analyzed by radar data to obtain a groundwater level change distribution map;

[0014] By integrating surface morphology feature vectors, vegetation health distribution maps, soil temperature feature maps, and groundwater level change distribution maps, a comprehensive index system is constructed using a weighted linear combination algorithm to obtain the evaluation results of mine reclamation effectiveness.

[0015] Preferably, the acquisition of multi-source remote sensing data for mine reclamation includes optical, radar, hyperspectral, and thermal infrared imagery. Cloud cover interference is eliminated through preprocessing to obtain a standardized multi-temporal dataset, including:

[0016] Acquire multi-source remote sensing data for mine reclamation, and extract raw datasets from optical images, radar images, hyperspectral images, and thermal infrared images;

[0017] The support vector machine algorithm is used to perform cloud detection on optical images, determine the cloud coverage area, and generate cloud mask images.

[0018] By fusing radar imagery and cloud mask imagery, if there is cloud cover in the optical imagery, the radar imagery is used to fill the cloud-covered area to obtain a cloud-free imagery.

[0019] For cloud-removed images, principal component analysis algorithm is used to extract the main features of hyperspectral images and generate feature-enhanced images;

[0020] Based on the feature-enhanced image and the thermal infrared image, image registration is performed. If the registration deviation is less than a preset threshold, a multi-source fused image is generated.

[0021] Standardized images are obtained by standardizing multi-source fused images using the mean normalization method.

[0022] For standardized images, temporal alignment is performed. If the temporal deviation is less than a preset threshold, a multi-temporal dataset is generated.

[0023] Preferably, the step of using an image registration algorithm to adjust the spatial resolution based on feature point matching for a standardized multi-temporal dataset to obtain a spatially consistent registration dataset includes:

[0024] A multi-temporal dataset is obtained, and feature points of each temporal image are extracted using a feature point detection algorithm to obtain a feature point set;

[0025] Based on the feature point set, the scale-invariant feature transformation algorithm is used to match feature points and determine matching point pairs;

[0026] If the number of matching point pairs exceeds a preset threshold, then erroneous matching points are removed using a random sampling consensus algorithm to obtain optimized matching point pairs.

[0027] Based on the optimized matching point pairs, the geometric transformation matrix of each temporal image is calculated, and geometric correction parameters are generated;

[0028] By adjusting the spatial resolution of images at different time phases using geometric correction parameters, spatially consistent registered images are obtained.

[0029] For the registered images, calculate the registration accuracy index and determine whether the registration dataset meets the preset accuracy requirements;

[0030] If the registration accuracy does not reach the preset threshold, return to the optimized matching point pair step, perform geometric correction again, and obtain the updated registration dataset.

[0031] Preferably, the step of extracting radiometric characteristics from the registration dataset and using a radiometric correction algorithm to unify the radiometric standards of different sensors to obtain a radiometrically consistent correction dataset includes:

[0032] The raw radiometric values ​​are obtained from the registration dataset, and the radiometric features are determined by the feature extraction algorithm.

[0033] The radiation characteristics are processed using a preset radiation correction algorithm to unify the radiation standards of different sensors and obtain the corrected radiation value.

[0034] If the corrected radiation value deviates from the preset standard deviation of radiation by more than a threshold, the correction parameters are adjusted through an iterative optimization algorithm to obtain the optimized radiation value.

[0035] A radiation consistency model is constructed based on the optimized radiation values, and a calibration dataset is generated through data fusion technology.

[0036] Consistency features are extracted from the calibration dataset, and the support vector machine algorithm is used to determine data consistency to obtain consistency evaluation results.

[0037] Based on the consistency assessment results, outlier data is removed using data filtering techniques to obtain the final corrected dataset;

[0038] For the final calibrated dataset, data storage technology is used to save the radiometrically consistent data, resulting in a persistent dataset.

[0039] Preferably, the step of constructing a time-series database based on the calibration dataset, and fusing features of landform, vegetation health, soil temperature, and groundwater level changes to obtain a multi-dimensional feature dataset, including:

[0040] Landform, vegetation health, soil temperature and groundwater level data were obtained from the calibration dataset. Time series analysis methods were used to construct a multidimensional time series database to obtain a multidimensional time series dataset.

[0041] Based on the multidimensional time series dataset, feature extraction methods were used to separate land surface morphology features, vegetation health index, soil temperature data and groundwater level changes to obtain independent feature sets.

[0042] For independent feature sets, if the correlation between landform features and vegetation health index is lower than a preset threshold, principal component analysis is used to extract the main change features and obtain the principal feature set.

[0043] Based on the main feature set, data fusion technology is used to integrate landform features, vegetation health index, soil temperature data and groundwater level changes to generate a fused feature dataset.

[0044] For the fused feature dataset, if the deviation between groundwater level changes and soil temperature data exceeds a preset threshold, the feature weights are adjusted using a linear regression algorithm to obtain an optimized feature dataset.

[0045] Based on the optimized feature dataset, cluster analysis was used to determine the dynamic correlation pattern between landform features and vegetation health index, resulting in a dynamic correlation dataset.

[0046] For dynamically correlated datasets, data storage technology is used to persist feature data that integrates changes in landform, vegetation health, soil temperature, and groundwater level, resulting in a multidimensional feature dataset.

[0047] Preferably, the step of extracting the main patterns of landform changes using principal component analysis algorithm for the multidimensional feature dataset to obtain landform feature vectors includes:

[0048] Time series analysis methods are used to separate time series data of landform changes from multidimensional feature datasets to obtain landform time series sets;

[0049] Principal component analysis algorithm is used to extract the main change patterns from the landform time series set and obtain the main feature vector of landform.

[0050] If the correlation between the main feature vector of landform and the vegetation health index is lower than the preset threshold, then data fusion technology is used to integrate the main feature vector of landform and the vegetation health index to obtain the landform vegetation fusion feature set.

[0051] Based on the landform and vegetation integration feature set, cluster analysis was used to determine the dynamic correlation pattern between landform changes and vegetation health index, and a dynamic correlation feature set was obtained.

[0052] If the deviation between the groundwater level change and the soil temperature data in the dynamic correlation feature set exceeds a preset threshold, the feature weights in the dynamic correlation feature set are adjusted by a linear regression algorithm to obtain an optimized dynamic correlation feature set.

[0053] Based on the optimized dynamic association feature set, data storage technology is used to persist the feature data of landform changes, vegetation health index, soil temperature data and groundwater level changes, resulting in a multidimensional feature dataset.

[0054] Preferably, the step of calculating vegetation health status from the multidimensional feature dataset using the normalized vegetation index and combining it with landform feature vectors to obtain a vegetation health distribution map includes:

[0055] From the multidimensional feature dataset, data preprocessing methods are used to extract the spectral data and landform feature data required for the normalized vegetation index, and a preprocessed feature set is obtained.

[0056] Based on the preprocessed feature set, the normalized vegetation index algorithm is used to calculate the vegetation health status and generate a vegetation health index set. The normalized vegetation index formula is NDVI = (NIR - RED) / (NIR + RED), where NIR represents the near-infrared band value and RED represents the red band value.

[0057] If the values ​​in the vegetation health index set are lower than the preset threshold, outliers are removed using data cleaning techniques to obtain a cleaned vegetation health index set.

[0058] Based on the cleaned vegetation health index set, combined with the landform feature vector, feature vector fusion technology is used to integrate the vegetation health index and landform features to generate a fused feature set.

[0059] Based on the fused feature set, a spatial interpolation algorithm is used to generate a spatial distribution pattern of vegetation health distribution, resulting in a vegetation health distribution map.

[0060] If the spatial distribution pattern in the vegetation health distribution map does not match the preset spatial uniformity threshold, the spatial resolution of the distribution map is adjusted by grid division technology to obtain an optimized vegetation health distribution map.

[0061] Based on the optimized vegetation health distribution map, data storage technology is used to persist the vegetation health distribution map and the fused feature set, generating a multidimensional feature dataset.

[0062] Preferably, the step of analyzing soil temperature distribution using thermal infrared data based on a multidimensional feature dataset, and fusing landform feature vectors and vegetation health distribution maps to obtain a soil temperature feature map includes:

[0063] Acquire thermal infrared data and landform feature vectors, remove noise through preprocessing, and obtain the first temperature dataset and the first terrain dataset;

[0064] Principal component analysis algorithm is used to extract key feature vectors from the first temperature dataset and the first terrain dataset to obtain the first feature set;

[0065] If the spatial resolution of the first feature set is lower than a preset threshold, the spatial resolution is enhanced by an interpolation algorithm to obtain the second feature set.

[0066] The vegetation health index was extracted from the vegetation health distribution map and normalized to obtain the first vegetation dataset.

[0067] A weighted fusion algorithm is used to fuse the second feature set with the first vegetation dataset to generate a comprehensive feature set;

[0068] By analyzing the spatial distribution, a soil temperature distribution model is constructed based on a comprehensive feature set, resulting in a soil temperature feature map.

[0069] If the variance of a local area in the soil temperature feature map exceeds a preset threshold, a smoothing filtering algorithm is used for optimization to obtain the final soil temperature feature map.

[0070] Preferably, if the soil temperature characteristic map shows an abnormal area, then the groundwater level change trend is analyzed through radar data to obtain a groundwater level change distribution map, including:

[0071] Soil temperature data is acquired through sensors, a soil temperature feature map is constructed, and abnormal areas are identified.

[0072] If the abnormal area in the soil temperature feature map exceeds the preset threshold, the groundwater level information is extracted from the radar data to obtain the original water level dataset.

[0073] Time series analysis was used to process the raw water level dataset and calculate the trend of groundwater level changes.

[0074] Based on the trend of groundwater level changes, and combined with spatial interpolation algorithms, a groundwater level change distribution map is generated;

[0075] By comparing the distribution maps of soil temperature anomalies and groundwater level changes, the spatial distribution correlation can be determined.

[0076] If the correlation is higher than the preset threshold, the soil temperature and groundwater level data will be merged to generate a comprehensive environmental monitoring distribution map.

[0077] Machine learning classification algorithms were used to analyze the comprehensive environmental monitoring distribution map to identify areas with abnormal hydrological characteristics.

[0078] Preferably, the integrated landform feature vector, vegetation health distribution map, soil temperature feature map, and groundwater level change distribution map are used to construct a comprehensive index system using a weighted linear combination algorithm to obtain the mine reclamation effect evaluation results, including:

[0079] Land surface morphology feature vectors, vegetation health distribution maps, soil temperature feature maps, and groundwater level change distribution maps are obtained from multi-source remote sensing data. Standardization processing is then used to obtain a feature dataset in a unified format.

[0080] For the feature dataset, principal component analysis algorithm is used to extract the main components of land surface morphology features, vegetation health distribution, soil temperature features and groundwater level changes, and obtain a set of feature vectors after dimensionality reduction.

[0081] Based on the feature vector set, a weighted linear combination algorithm is used to merge the feature vectors through a preset weight allocation scheme to generate a comprehensive index system.

[0082] If the values ​​of each indicator in the comprehensive indicator system meet the preset threshold conditions, then a spatial interpolation algorithm is used to generate a spatial distribution map of the mine reclamation effect, and the spatial characteristics of the reclamation effect are obtained.

[0083] Based on the spatial characteristics of reclamation effects, a clustering analysis algorithm is used to divide the spatial distribution map into regions, and the classification results of reclamation effects are obtained.

[0084] The distribution patterns of reclamation effects are extracted from the classification results, and statistical analysis methods are used to generate quantitative evaluation results of reclamation effects.

[0085] Compared with the prior art, the present invention has the following advantages and technical effects:

[0086] This invention discloses a comprehensive evaluation method for mine reclamation effects based on multi-source remote sensing data fusion. Addressing the challenge of uniformly quantifying and evaluating changes in landform, vegetation health, soil temperature, and groundwater level during mine reclamation, this method preprocesses multi-source remote sensing data to eliminate cloud cover interference, generating a standardized multi-temporal dataset. Image registration and radiometric correction algorithms ensure spatial and radiometric consistency, constructing a multi-dimensional feature dataset. Principal component analysis is used to extract landform feature vectors, combined with the normalized vegetation index to generate a vegetation health distribution map. Soil temperature distribution is analyzed using thermal infrared data, and groundwater level trends are monitored by fusing radar data. Finally, a weighted linear combination algorithm is used to construct a comprehensive index system, achieving accurate evaluation of mine reclamation effects. This invention significantly improves the accuracy and comprehensiveness of reclamation effect evaluation through multi-source data fusion and multi-dimensional feature analysis, providing a scientific basis for mine ecological restoration. Attached Figure Description

[0087] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0088] Figure 1 This is a flowchart of a method for evaluating the effectiveness of mine reclamation according to an embodiment of the present invention;

[0089] Figure 2 This is a schematic diagram of the mine reclamation effect evaluation method according to an embodiment of the present invention;

[0090] Figure 3 This is another schematic diagram of the mine reclamation effect evaluation method according to an embodiment of the present invention. Detailed Implementation

[0091] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0092] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0093] Example 1

[0094] like Figure 1-3 As shown, this embodiment provides a method for evaluating the effectiveness of mine reclamation based on multi-temporal remote sensing imagery, including:

[0095] S101. Acquire multi-source remote sensing data for mine reclamation, including optical, radar, hyperspectral, and thermal infrared images. Eliminate cloud cover interference through preprocessing to obtain a standardized multi-temporal dataset.

[0096] Multi-source remote sensing data for mine reclamation was acquired, and raw datasets were extracted from optical, radar, hyperspectral, and thermal infrared images. Support vector machine (SVM) algorithms were used to detect clouds in the optical images, identifying cloud-covered areas and generating cloud mask images. By fusing radar and cloud mask images, if cloud cover was present in the optical images, the radar images were used to fill the cloud-covered areas, resulting in de-clouded images. For the de-clouded images, principal component analysis (PCA) was used to extract the main features of the hyperspectral images, generating feature-enhanced images. Image registration was performed between the feature-enhanced images and the thermal infrared images; if the registration deviation was less than a preset threshold, the images were fused to generate multi-source fused images. The multi-source fused images were then standardized using mean normalization to obtain standardized images. For the standardized images, temporal alignment was performed; if the temporal deviation was less than a preset threshold, a multi-temporal dataset was generated.

[0097] For example, multi-source remote sensing data processing involves the collaborative analysis of optical imagery, radar imagery, hyperspectral imagery, and thermal infrared imagery, aiming to generate high-quality multi-temporal datasets. The following analysis and examples cover each technical topic, focusing on a single scenario of remote sensing image processing.

[0098] For example, cloud detection in optical imagery employs the Support Vector Machine (SVM) algorithm. SVM distinguishes between cloud and non-cloud regions by classifying pixel features. Assuming a Landsat-8 optical image, pixel spectral values ​​are extracted as feature vectors, and training data includes cloud, vegetation, and water samples, the algorithm labels cloud regions as 1 and non-cloud regions as 0, generating a binary cloud mask image. This cloud mask image identifies cloud-covered areas, such as areas with 30% coverage, which helps in subsequent processing to clearly identify areas requiring restoration. The advantages of this method are high-precision classification, reduced false positives, and improved image usability.

[0099] In one possible implementation, radar imagery is fused with cloud-masked imagery to remove cloud interference. Radar imagery is unobstructed by clouds and can penetrate cloud layers to obtain surface information.

[0100] For example, Sentinel-1 radar imagery is combined with optical imagery cloud masks. After identifying cloud-covered areas, data from the same area in the radar imagery is used to fill in the missing parts of the optical imagery. Assuming the cloud-covered area is 5000 pixels, the radar imagery provides surface reflectance data, which is used to replace the cloud pixels to generate a cloud-free image. This fusion improves image integrity and is suitable for scenarios in agricultural monitoring that require continuous surface information.

[0101] Specifically, principal component analysis (PCA) is used to extract hyperspectral image features from cloud-free images. Hyperspectral images have numerous bands, resulting in significant data redundancy. PCA reduces the dimensionality of high-dimensional bands to principal components; for example, compressing 200 bands into 5 principal components while retaining 95% of the information variance. This feature enhancement highlights the spectral characteristics of ground features, such as distinguishing different crop types, improving classification accuracy, and providing high-quality input for subsequent image fusion.

[0102] For example, feature-enhanced images are registered with thermal infrared images. Registration ensures spatial alignment between the two images. Assuming a feature-point-based registration method is used, corner points are extracted from the images. If the registration deviation is less than a 0.5-pixel threshold, the alignment is considered successful. After registration, the images are fused to generate a multi-source fused image, such as combining hyperspectral crop information with thermal infrared land surface temperature to create a comprehensive image. This fusion improves the accuracy of land cover classification and environmental monitoring.

[0103] In one possible implementation, the multi-source fused images are standardized using mean normalization. Assuming the image pixel value range is 0-10000, the normalized values ​​are adjusted to the range of 0-1 to eliminate dimensional differences.

[0104] For example, normalizing the pixel values ​​of crop areas facilitates cross-image comparisons and improves the training effect of machine learning models.

[0105] It should be noted that standardized imagery is time-series aligned to generate a multi-temporal dataset. Assuming the imagery was acquired between March and June 2025, alignment is considered successful if the time deviation is less than the 1-day threshold.

[0106] For example, aligning monthly image sequences creates a temporally continuous crop growth monitoring dataset, suitable for analyzing crop growth cycles. This method ensures temporal consistency and improves the reliability of dynamic monitoring.

[0107] Understandably, the above process, through multi-source data collaboration and gradual optimization, significantly improves the usability and analysis accuracy of remote sensing imagery, providing high-quality data support for applications such as agriculture and environmental monitoring.

[0108] S102. For standardized multi-temporal datasets, an image registration algorithm is used to adjust the spatial resolution based on feature point matching to obtain a spatially consistent registration dataset.

[0109] A multi-temporal dataset is acquired, and feature points for each temporal image are extracted using a feature point detection algorithm to obtain a feature point set. Based on the feature point set, a scale-invariant feature transformation algorithm is used for feature point matching to determine matching point pairs. If the number of matching point pairs exceeds a preset threshold, erroneous matching points are removed using a random sampling consensus algorithm to obtain optimized matching point pairs. Based on the optimized matching point pairs, the geometric transformation matrix for each temporal image is calculated to generate geometric correction parameters. Using the geometric correction parameters, the spatial resolution of each temporal image is adjusted to obtain a spatially consistent registered image. For the registered image, a registration accuracy index is calculated to determine whether the registration dataset meets the preset accuracy requirements. If the registration accuracy does not reach the preset threshold, the process returns to the optimized matching point pair step, and geometric correction is performed again to obtain an updated registration dataset.

[0110] For example, feature point detection is a crucial step in acquiring multi-temporal datasets. Feature point detection algorithms are commonly used to extract points with significant features in images, such as corner points or edge points. One possible implementation is the Harris corner detection algorithm, which identifies feature points with high contrast and stability by calculating the image grayscale gradient. Assuming 1000 corner points are detected in a set of optical images, these points exhibit high repeatability across different temporal images, aiding in subsequent matching.

[0111] For example, in a 5000×5000 pixel image, detected corner points may be located at building edges or boundaries of ground features; these points are crucial for subsequent registration. During the feature point matching stage, the Scale Invariant Feature Transform (SIFT) algorithm can be used to determine matching point pairs. SIFT extracts illumination- and rotation-invariant feature descriptors by constructing a scale space.

[0112] For example, 1000 and 1200 feature points are extracted from two temporal images, respectively. After SIFT matching, 800 initial matching point pairs are obtained. These point pairs may contain some incorrect matches, so further optimization is needed.

[0113] It should be noted that the SIFT algorithm determines the matching relationship through the Euclidean distance of feature descriptors, thus ensuring the robustness of the matching.

[0114] Specifically, the Random Sample Consensus Algorithm (RANSAC) is used to remove incorrect matches. RANSAC iteratively selects a subset of matching points, estimates the transformation model, and removes points that do not conform to the model.

[0115] For example, from 800 pairs of matched points, RANSAC might remove 100 outliers, retaining 700 optimized matched point pairs. These optimized point pairs can be used to compute high-precision geometric transformation matrices.

[0116] Preferably, RANSAC can be set to a maximum of 1000 iterations and a minimum matching point threshold of 50 pairs to ensure computational efficiency and accuracy. In the calculation of the geometric transformation matrix, optimizing matching point pairs can be used to estimate affine or projective transformation parameters.

[0117] For example, in a set of multi-temporal images, translation, rotation, and scaling parameters are calculated using 700 pairs of matching points, generating geometric correction parameters. These parameters can control the spatial deviation of the images to within 1 pixel.

[0118] It should be noted that the accuracy of the transformation matrix directly affects the subsequent registration results. Spatial resolution adjustment is an important step in generating registered images.

[0119] For example, if a set of images has spatial resolutions of 10 meters and 15 meters respectively, they can be uniformly adjusted to 10 meters resolution using interpolation methods (such as bilinear interpolation) to generate spatially consistent registered images. This adjustment ensures the spatial comparability of images from different time periods. In registration accuracy evaluation, the root mean square error (RMSE) can be calculated as an accuracy indicator.

[0120] For example, by comparing the feature point positions of the registered image and the reference image, the RMSE is calculated to be 0.5 pixels. If the preset threshold is 1 pixel, then the registration dataset meets the requirements. If the RMSE is 1.5 pixels, then the optimization matching point pair step needs to be returned to, and geometric correction needs to be performed again.

[0121] For example, the number of SIFT feature points extracted can be increased to 1500, and the matching and optimization can be repeated until the accuracy is achieved. This iterative optimization method can significantly improve the quality of the registration dataset.

[0122] Specifically, if the registration accuracy is not up to standard, it can be optimized by adjusting the number of RANSAC iterations or the threshold for feature point detection.

[0123] For example, reducing the grayscale gradient threshold of Harris corner detection by 10% may increase the number of feature points by 200, thereby increasing the number of matching point pairs. This method, through multi-faceted optimization, ensures the spatial consistency and accuracy of the final registration dataset, providing a reliable data foundation for subsequent multi-temporal analysis.

[0124] S103. Extract radiation characteristics from the registration dataset, and use a radiation correction algorithm to unify the radiation standards of different sensors to obtain a radiation-consistent correction dataset.

[0125] Raw radiometric values ​​are obtained from the registration dataset, and radiometric features are determined using a feature extraction algorithm. A pre-defined radiometric correction algorithm is then used to process the radiometric features, unifying the radiometric standards of different sensors to obtain corrected radiometric values. If the deviation between the corrected radiometric value and the pre-defined radiometric standard exceeds a threshold, the correction parameters are adjusted using an iterative optimization algorithm to obtain optimized radiometric values. A radiometric consistency model is constructed based on the optimized radiometric values, and a correction dataset is generated using data fusion techniques.

[0126] For example, the process of obtaining raw radiometric values ​​from a registration dataset and performing radiometric correction involves multiple technical topics, including feature extraction, radiometric correction, iterative optimization, data fusion, consistency assessment, data filtering, and data storage. The following sections, within the context of remote sensing image processing and drawing on historical discussions of image registration, provide detailed explanations through examples and analysis to ensure the content is closely related and logically sound.

[0127] For example, when obtaining raw radiometric values ​​from a registration dataset, radiometric features can be extracted by analyzing the grayscale distribution of the image. Suppose a multi-temporal remote sensing image dataset contains Landsat-8 and Sentinel-2 images, and the raw radiometric values ​​are represented as digitally quantized values ​​(DN values) for each band.

[0128] In one possible implementation, principal component analysis (PCA) is used as the feature extraction algorithm to reduce the DN values ​​of multi-band images to the main radiometric feature vectors.

[0129] For example, the visible light band DN value range of Landsat-8 is 0-65535. The first two principal components are extracted by PCA, retaining 90% of the radiation information, and a feature vector is generated to facilitate subsequent correction.

[0130] Specifically, the radiometric correction algorithm can employ absolute radiometric correction methods, such as the 6S model based on the radiative transfer model. For the radiometric characteristics of Landsat-8 and Sentinel-2 imagery, the 6S model converts DN values ​​into surface reflectivity by inputting atmospheric parameters (such as water vapor content and aerosol optical thickness), thus achieving a unified radiometric standard across different sensors.

[0131] For example, the red band DN value of Landsat-8 is converted to a reflectance range of 0.1-0.3 after correction, which is consistent with the reflectance standard of Sentinel-2, eliminating the deviation caused by sensor differences.

[0132] In one embodiment, if the corrected radiance value deviates from a preset standard deviation by more than a threshold (e.g., a reflectance deviation greater than 0.05), the correction parameters are adjusted using an iterative optimization algorithm. For example, if the near-infrared reflectance of the Landsat-8 image is too high during the initial correction, the atmospheric parameters can be iteratively adjusted using the gradient descent method, and the 6S model can be rerun until the deviation is less than 0.05. The optimized radiance value is closer to the true surface reflectance, which helps improve the accuracy of subsequent analyses.

[0133] For example, when constructing a radiometric consistency model, optimized radiometric values ​​from multiple temporal images can be integrated using data fusion techniques. Assuming the fusion of Landsat-8 and Sentinel-2 corrected data, a weighted average fusion method is used, assigning weights according to time intervals (e.g., 0.7 for recent images and 0.3 for earlier ones) to generate a radiometrically consistent fused dataset. This method effectively smooths radiometric differences between temporal phases and improves the continuity of the dataset.

[0134] Specifically, the Support Vector Machine (SVM) algorithm can be used for consistency assessment. For the fused dataset, consistency features, such as the mean and variance of reflectance for each band, are extracted and input into the SVM model to determine whether the data meets the consistency requirements.

[0135] For example, the SVM model confirms the radial consistency of the fused dataset with a classification accuracy of 90%, and outlier data points (such as cloud-covered areas) are marked as inconsistent for easy removal later.

[0136] In one possible implementation, data filtering techniques can remove outlier data through threshold filtering. For example, if a consistency assessment identifies an area with an abnormal reflectance (e.g., a red band value higher than 0.5), this area can be marked as a cloud or shadow region and removed, retaining radiometrically consistent pixels to generate the final calibrated dataset. This filtering ensures high-quality datasets, providing a reliable foundation for subsequent analysis.

[0137] For example, data storage techniques can employ a distributed file system (such as HDFS) to store the final calibration dataset. Assuming the dataset contains 10TB of remote sensing imagery, it can be stored in HDFS fragments, each 512MB, with attached metadata (such as image time and sensor type) for easy retrieval and long-term preservation. Persistent storage ensures data reusability, supporting subsequent dynamic monitoring tasks. Through the implementation of these techniques, the generated calibration dataset exhibits significant advantages in radiometric consistency and data quality, making it suitable for applications such as land surface monitoring and change detection in the field of remote sensing.

[0138] S104. Based on the calibration dataset, construct a time series database and integrate the characteristics of landform, vegetation health, soil temperature and groundwater level changes to obtain a multidimensional feature dataset.

[0139] Landform, vegetation health, soil temperature, and groundwater level data were obtained from a calibration dataset. A multidimensional time-series database was constructed using time-series analysis methods, resulting in a multidimensional time-series dataset. Based on this dataset, feature extraction methods were used to separate landform features, vegetation health index, soil temperature data, and groundwater level changes, yielding independent feature sets. For each independent feature set, if the correlation between landform features and the vegetation health index was below a preset threshold, principal component analysis was used to extract the main changing features, resulting in a principal feature set. Based on this principal feature set, data fusion techniques were employed to integrate landform features, vegetation health index, soil temperature data, and groundwater level changes, generating a fused feature dataset. For this fused feature dataset, if the deviation between groundwater level changes and soil temperature data exceeded a preset threshold, linear regression was used to adjust the feature weights, resulting in an optimized feature dataset. Based on this optimized feature dataset, cluster analysis was used to determine the dynamic correlation pattern between landform features and the vegetation health index, yielding a dynamically correlated dataset. For dynamically correlated datasets, data storage technology is used to persist feature data that integrates changes in landform, vegetation health, soil temperature, and groundwater level, resulting in a multidimensional feature dataset.

[0140] For example, a time-series database is constructed based on a calibrated dataset. First, landform, vegetation health, soil temperature, and groundwater level changes are acquired using remote sensing imagery and sensor data. Landform data utilizes Landsat 8 satellite imagery to extract a digital elevation model (DEM) with a resolution of 30 meters. The slope is calculated using ArcGIS's slope analysis algorithm; for example, the average slope of a certain area is 5.2 degrees with a standard deviation of 1.8 degrees, used to characterize terrain complexity. Vegetation health is calculated using the Normalized Difference Vegetation Index (NDVI), based on the near-infrared and red bands of Landsat 8. The formula is NDVI = (NIR - RED) / (NIR + RED). In a certain area, the average NDVI in June 2025 was 0.65, indicating good vegetation cover. Soil temperature data is acquired from ground sensors. Assuming the average daily soil temperature in a certain area in June is 22.3℃, time-series interpolation (linear interpolation) is used to fill in missing data, ensuring hourly data integrity. Groundwater level changes are recorded using downhole water level sensors. Assuming an average annual drop of 0.15 meters in a certain area, wavelet transform is used to decompose the time series data and extract periodic fluctuation features, such as a 12-month periodic signal strength of 0.08 meters. When fusing multidimensional features, the data is standardized (Z-score method, mean 0, variance 1), and dimensionality is reduced using principal component analysis (PCA) while retaining 90% of the variance, resulting in a 4-dimensional feature vector. The time series database is stored in HBase, with the row key designed as "region ID + timestamp," and column families including landform, vegetation health, soil temperature, and groundwater level. The timestamp accuracy is down to the second to ensure efficient querying.

[0141] For example, querying data for a specific region on June 1, 2025, returns a slope of 5.2 degrees, an NDVI of 0.65, a soil temperature of 22.3℃, and a water level drop of 0.15 meters. The analysis process uses K-means clustering (K=3) to classify the multidimensional dataset into high, medium, and low-risk areas. A particular area is classified as medium-risk, and based on business needs (such as agricultural irrigation), optimized water resource allocation is recommended. This method ensures data consistency and analytical accuracy, logically forming a closed loop from data acquisition and processing to storage and analysis, thus meeting the requirements for multidimensional feature fusion.

[0142] S105. For the multidimensional feature dataset, the principal component analysis algorithm is used to extract the main patterns of landform changes and obtain the landform feature vector.

[0143] Time series analysis was employed to separate time series data of landform changes from a multidimensional feature dataset, resulting in a landform time series set. Principal component analysis was then used to extract the main change patterns from this landform time series set, yielding the principal feature vectors of the landform.

[0144] For example, suppose there is a dataset containing multidimensional features of landform, with 1000 sample points, each sample point having 5 features: elevation (meters), slope (degrees), aspect (degrees), curvature (unitless), and surface roughness (unitless). The data is first standardized by adjusting the mean of each feature to 0 and the variance to 1 to eliminate dimensional differences.

[0145] For example, the elevation data ranges from 50 to 200 meters. After standardization, the mean is 0 and the standard deviation is 1. The calculation formula is z = (x - μ) / σ, where μ is the mean and σ is the standard deviation. This can be implemented using Python's `sklearn.preprocessing.StandardScaler`. Next, Principal Component Analysis (PCA) is used to extract the main patterns of landform changes. First, the covariance matrix of the data is calculated. Assuming the covariance matrix is ​​a 5×5 matrix, the elements represent the covariance between each feature; for example, the covariance between elevation and slope is 0.75. Then, eigenvalue decomposition is performed on the covariance matrix to obtain eigenvalues ​​and eigenvectors. Assuming the eigenvalues ​​are [3.2, 1.1, 0.4, 0.2, 0.1], the corresponding eigenvectors are the column vectors of matrix V. Selecting the first two eigenvalues ​​(3.2 and 1.1), their cumulative variance contribution rate reaches 85% (3.2+1.1) / (3.2+1.1+0.4+0.2+0.1) = 85%, which meets the threshold for retaining key information. Using these two eigenvectors, the original data is projected into a new space to obtain principal component eigenvectors. For example, each sample point, after projection, yields a two-dimensional vector [x1, x2], where x1 represents the first principal component and x2 represents the second principal component. The projection formula is Y = X·V, where X is the standardized data matrix and V is the selected eigenvector matrix. Analysis of the principal component loading matrices reveals that the first principal component mainly reflects changes in elevation and slope (loading values ​​of 0.65 and 0.55, respectively), while the second principal component mainly reflects aspect and curvature (loading values ​​of 0.70 and 0.60, respectively). Ultimately, the landform feature vector can be used for subsequent classification or change detection. For example, K-means clustering can be used to divide the landform into erosion zones and deposition zones, with cluster centers of [1.5, 0.8] and [-1.2, -0.9], respectively, thereby revealing the main patterns of landform changes.

[0146] S106. From the multidimensional feature dataset, the normalized vegetation index is used to calculate the vegetation health status, and combined with the land surface feature vector, a vegetation health distribution map is obtained.

[0147] From the multidimensional feature dataset, data preprocessing methods are used to extract the spectral data and landform feature data required for the Normalized Difference Vegetation Index (NDVI), resulting in a preprocessed feature set. Based on the preprocessed feature set, the NDVI algorithm is used to calculate vegetation health status, generating a vegetation health index set. The formula for the NDVI is NDVI = (NIR - RED) / (NIR + RED), where NIR represents the near-infrared band value and RED represents the red band value. If the values ​​in the vegetation health index set are lower than a preset threshold, outliers are removed using data cleaning techniques, resulting in a cleaned vegetation health index set. Based on the cleaned vegetation health index set and landform feature vectors, feature vector fusion technology is used to integrate the vegetation health index and landform features, generating a fused feature set. Based on the fused feature set, a spatial interpolation algorithm is used to generate a spatial distribution pattern of vegetation health, resulting in a vegetation health distribution map.

[0148] For example, suppose there is a dataset containing landform and vegetation-related features, with 800 sample points. Each sample point has four features: near-infrared spectral value (NIR, range 0.2-0.9, unitless), red spectral value (R, range 0.1-0.7, unitless), elevation (range 100-300 meters), and slope (range 0-45 degrees). First, the data is normalized by scaling each feature to the [0,1] interval to eliminate dimensional differences. The formula is x' = (x - x_min) / (x_max - x_min), implemented using Python's sklearn.preprocessing.MinMaxScaler.

[0149] For example, the NIR value, after normalization to 0.6, becomes (0.6-0.2) / (0.9-0.2) = 0.571. Next, the Normalized Difference Vegetation Index (NDVI) is calculated using the formula NDVI = (NIR-R) / (NIR+R). Taking a sample point as an example, with NIR = 0.8 and R = 0.3, after normalization, NIR' = 0.857, R' = 0.333, and NDVI = (0.857-0.333) / (0.857+0.333) = 0.440, reflecting the vegetation health status. The NDVI value ranges from -1 to 1, with values ​​closer to 1 indicating healthy vegetation. Then, the NDVI is combined with elevation and slope features to construct a three-dimensional feature vector [X_NDVI, X_elevation, X_slope]. To extract the main patterns, Linear Discriminant Analysis (LDA) was used to calculate the inter-class and intra-class scatter matrices, assuming the target was vegetation health classification (healthy / unhealthy). The covariance between inter-class scatter matrix elements, such as NDVI and elevation, was 0.62, while the intra-class scatter matrix was calculated using the sample variance. LDA optimized the projection direction to obtain a one-dimensional feature vector; for example, the projected sample point value was [2.1], representing the weight of the health distribution. Analyzing the projection coefficients, the NDVI weight was 0.72, the elevation weight was 0.45, and the slope weight was 0.33, indicating that NDVI contributed the most to the health distribution. Finally, a spatial interpolation algorithm (such as Inverse Distance Weighted, IDW) was used to generate a vegetation health distribution map using the sample point coordinates and projected values. The weight function was w = 1 / d^2, where d is the distance, generating a raster map with a resolution of 10 meters. The health value ranged from [0,3], with higher values ​​indicating healthier vegetation.

[0150] S107. Based on the multidimensional feature dataset, analyze the soil temperature distribution using thermal infrared data, and fuse the land surface morphology feature vector and vegetation health distribution map to obtain a soil temperature feature map.

[0151] Thermal infrared data and landform feature vectors are acquired. Noise is removed through preprocessing to obtain a first temperature dataset and a first topographic dataset. Principal component analysis (PCA) is used to extract key feature vectors from these datasets, resulting in a first feature set. If the spatial resolution of the first feature set is lower than a preset threshold, an interpolation algorithm is used to enhance the spatial resolution, resulting in a second feature set. A vegetation health index is extracted from the vegetation health distribution map and normalized to obtain a first vegetation dataset. A weighted fusion algorithm is used to fuse the second feature set with the first vegetation dataset, generating a comprehensive feature set. A soil temperature distribution model is constructed based on the comprehensive feature set through spatial distribution analysis, resulting in a soil temperature feature map. If the local variance of the soil temperature feature map exceeds a preset threshold, a smoothing filter algorithm is used for optimization, yielding the final soil temperature feature map.

[0152] For example, to analyze soil temperature distribution based on a multidimensional feature dataset, thermal infrared data processing is first employed. Assuming Landsat-8 TIRS band 10 data with a resolution of 100 meters is used, thermal infrared images of the study area on July 23, 2025, are obtained, with radiance values ​​ranging from 0.1 to 0.5 W / (m²). 2 ·sr·μm). Surface temperature was calculated using the radiative transfer equation Lλ=εσT^4, where ε is the soil emissivity (taken as 0.95) and σ is the Stefan-Boltzmann constant (5.67×10^-8 W / (m²)). 2 ·K^4), yielding a temperature range of 293K-313K (20℃-40℃). To improve accuracy, the atmospheric correction algorithm FLAASH is applied, with an input atmospheric water vapor content of 1.5g / cm³. 2 After correction, the temperature error was controlled within ±1K. Next, landform feature vectors were fused, DEM data (30-meter resolution) was extracted, and slope (0-30°) and aspect (0-360°) were calculated. A weighted overlay method was used, with weights of slope 0.4, aspect 0.3, and thermal infrared temperature 0.3, to generate a landform influence factor map. Areas with slopes greater than 15° showed a temperature reduction of approximately 2℃. Subsequently, vegetation health distribution maps were fused, and NDVI was calculated using Sentinel-2 imagery, ranging from 0.2 to 0.8. Areas with NDVI above 0.6 indicated healthy vegetation and a temperature reduction of approximately 1.5℃. Using a random forest algorithm, with input temperature, slope, aspect, and NDVI features, a model was trained to predict soil temperature. R0 2 The NDVI reached 0.85. A soil temperature characteristic map was finally generated with a resolution of 30 meters, showing a temperature range of 18℃-42℃. Areas with high slope and low NDVI had higher temperatures. Analysis indicates that vegetation cover and landform significantly affect temperature distribution, and NDVI is negatively correlated with temperature (Pearson coefficient -0.72), providing a basis for agricultural management.

[0153] S108. If the soil temperature characteristic map shows an abnormal area, then the groundwater level change trend is analyzed by radar data to obtain a groundwater level change distribution map.

[0154] Soil temperature data is acquired using sensors to construct a soil temperature feature map and identify abnormal areas. If the abnormal areas in the soil temperature feature map exceed a preset threshold, groundwater level information is extracted from radar data to obtain a raw water level dataset. Time series analysis is used to process the raw water level dataset to calculate the groundwater level change trend. Based on the groundwater level change trend, a groundwater level change distribution map is generated using a spatial interpolation algorithm.

[0155] For example, soil temperature characteristic maps are acquired using remote sensing technology. An infrared thermal imaging sensor scans a 100 square kilometer farmland area, generating temperature distribution data with a resolution of 1 meter x 1 meter, in degrees Celsius. Assuming a temperature anomaly is detected in a certain area with an average temperature of 32.5°C, 5°C higher than the surrounding area, this area is marked as an anomaly, possibly due to changes in thermal conductivity caused by abnormal groundwater levels. Next, synthetic aperture radar (SAR) data is used to analyze groundwater level trends. C-band SAR images with a resolution of 10 meters are acquired, covering the same area over a period of 6 months, with data acquired monthly. The InSAR algorithm is used to calculate ground subsidence. The monthly subsidence is calculated using the phase difference formula Δφ = 4πΔh / λ (λ is the wavelength, 5.6 cm). Assuming a cumulative subsidence of 0.02 meters in an anomaly area over 6 months, and combining this with a linear relationship model between groundwater level and subsidence (a 1-meter drop in water level corresponds to 0.01-meter subsidence), the estimated water level drop is approximately 2 meters. Furthermore, based on the Kriging interpolation algorithm, the settlement data was interpolated to generate a groundwater level change distribution map with a grid resolution of 5 meters. The map showed that the abnormal water level drop was concentrated in the northwest corner, with a maximum drop of 2.3 meters, and a drop of 0.5-1.2 meters in the surrounding area. The analysis process combined soil porosity (assumed to be 0.4) and hydrogeological parameters to verify the correlation between water level drop and temperature anomalies, with a correlation coefficient of 0.85, indicating a high correlation between high-temperature areas and water level drop. For business applications, the system automatically output the distribution map to the agricultural irrigation decision platform, recommending increased irrigation for areas with water level drop, accurate to approximately 200 cubic meters of additional water per hectare.

[0156] S109. By integrating surface morphology feature vectors, vegetation health distribution maps, soil temperature feature maps, and groundwater level change distribution maps, a comprehensive index system is constructed using a weighted linear combination algorithm to obtain the evaluation results of mine reclamation effect.

[0157] Landform feature vectors, vegetation health distribution maps, soil temperature feature maps, and groundwater level change distribution maps are obtained from multi-source remote sensing data. Standardization is applied to obtain a unified feature dataset. Principal component analysis (PCA) is used to extract the main components of landform features, vegetation health distribution, soil temperature features, and groundwater level changes from this dataset, resulting in a dimensionality-reduced feature vector set. Based on this feature vector set, a weighted linear combination algorithm is used to fuse the feature vectors through a pre-defined weight allocation scheme, generating a comprehensive index system. If the values ​​of each index in the comprehensive index system meet pre-defined threshold conditions, a spatial interpolation algorithm is used to generate a spatial distribution map of the mine reclamation effect, obtaining the spatial characteristics of the reclamation effect. Based on the spatial characteristics of the reclamation effect, a clustering analysis algorithm is used to divide the spatial distribution map into regions, obtaining classification results for the reclamation effect. The distribution patterns of the reclamation effect are extracted from the classification results, and statistical analysis methods are used to generate a quantitative evaluation result of the reclamation effect.

[0158] For example, firstly, when fusing the landform feature vector, it is assumed that landform features, including slope, aspect, and topographic relief, are extracted from a high-resolution digital elevation model (DEM). Using ArcGIS software, 30-meter resolution DEM data is processed to calculate slope (range 0-45°, average 15.2°), aspect (0-360°, uniformly distributed), and topographic relief (standard deviation 2.5 meters). These features are standardized to the [0,1] interval, and principal component analysis (PCA) is used to extract the first three principal components, which contribute 85.3%, forming the landform feature vector. Next, a vegetation health distribution map is generated, and the Normalized Difference Vegetation Index (NDVI) is calculated using Landsat-8 imagery, ranging from -0.1 to 0.8, with an average of 0.45, reflecting vegetation cover. NDVI data were spatially interpolated (Kriging) to generate a continuous distribution map, categorized into low (<0.3), medium (0.3-0.6), and high (>0.6) health levels, accounting for 20%, 50%, and 30%, respectively. Then, soil temperature characteristic maps were obtained through surface temperature inversion, based on MODIS data, with a temperature range of 15-35℃ and an average of 25.6℃. Inverse distance weighted (IDW) interpolation was used to generate a spatial distribution, dividing the area into high-temperature (>30℃), moderate (20-30℃), and low-temperature (<20℃) regions, accounting for 15%, 70%, and 15%, respectively. The groundwater level change distribution map was generated based on monitoring well data (50 points, depth 2-10 meters, average decrease of 0.5 meters / year). A rate of change map was generated using ordinary kriging interpolation, dividing the area into stable (<0.2 meters / year), slightly decreasing (0.2-0.5 meters / year), and significantly decreasing (>0.5 meters / year) regions, accounting for 30%, 40%, and 30% respectively. Finally, a weighted linear combination algorithm was used to construct a comprehensive index system, with weights determined according to the entropy method: surface morphology 0.25, vegetation health 0.35, soil temperature 0.20, and groundwater level 0.20. After standardization of each index, a comprehensive score (range 0-1) was calculated using the formula S = 0.25 × S1 + 0.35 × S2 + 0.20 × S3 + 0.20 × S4, where S1-S4 are the standardized values ​​of each index. The results showed that the average score for mine reclamation effectiveness was 0.68, categorized as excellent (>0.8, 20%), good (0.6-0.8, 50%), and poor (<0.6, 30%), indicating relatively good reclamation results but with some areas needing optimization. Logically, surface morphology influences vegetation restoration, vegetation health is related to soil temperature, and groundwater level changes affect long-term ecological stability. These indicators are weighted and integrated to form a comprehensive assessment, demonstrating practical relevance.

[0159] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for evaluating the effectiveness of mine reclamation based on multi-temporal remote sensing imagery, characterized in that, Includes the following steps: Multi-source remote sensing data for mine reclamation were acquired, including optical, radar, hyperspectral, and thermal infrared images. Cloud cover interference was eliminated through preprocessing to obtain a standardized multi-temporal dataset. For standardized multi-temporal datasets, an image registration algorithm is used to adjust the spatial resolution based on feature point matching to obtain a spatially consistent registration dataset. Radiometric characteristics are extracted from the registration dataset, and a radiometric correction algorithm is used to unify the radiometric standards of different sensors to obtain a radiometrically consistent correction dataset. Based on the calibration dataset, a time series database is constructed, and the characteristics of landform, vegetation health, soil temperature and groundwater level changes are integrated to obtain a multidimensional feature dataset. For multidimensional feature datasets, principal component analysis algorithm is used to extract the main patterns of landform changes and obtain landform feature vectors; From the multidimensional feature dataset, the normalized vegetation index is used to calculate the vegetation health status, and combined with the landform feature vector, a vegetation health distribution map is obtained. Based on the multidimensional feature dataset, the soil temperature distribution was analyzed using thermal infrared data, and the surface morphology feature vector and vegetation health distribution map were fused to obtain the soil temperature feature map. If the soil temperature characteristic map shows abnormal areas, then the groundwater level change trend is analyzed by radar data to obtain a groundwater level change distribution map; By integrating surface morphology feature vectors, vegetation health distribution maps, soil temperature feature maps, and groundwater level change distribution maps, a comprehensive index system is constructed using a weighted linear combination algorithm to obtain the evaluation results of mine reclamation effectiveness.

2. The method according to claim 1, characterized in that, The acquisition of multi-source remote sensing data for mine reclamation includes optical, radar, hyperspectral, and thermal infrared imagery. Cloud cover interference is removed through preprocessing to obtain a standardized multi-temporal dataset, including: Acquire multi-source remote sensing data for mine reclamation, and extract raw datasets from optical images, radar images, hyperspectral images, and thermal infrared images; The support vector machine algorithm is used to perform cloud detection on optical images, determine the cloud coverage area, and generate cloud mask images. By fusing radar imagery and cloud mask imagery, if there is cloud cover in the optical imagery, the radar imagery is used to fill the cloud-covered area to obtain a cloud-free imagery. For cloud-removed images, principal component analysis algorithm is used to extract the main features of hyperspectral images and generate feature-enhanced images; Based on the feature-enhanced image and the thermal infrared image, image registration is performed. If the registration deviation is less than a preset threshold, a multi-source fused image is generated. Standardized images are obtained by standardizing multi-source fused images using the mean normalization method. For standardized images, temporal alignment is performed. If the temporal deviation is less than a preset threshold, a multi-temporal dataset is generated.

3. The method according to claim 1, characterized in that, The aforementioned standardized multi-temporal dataset employs an image registration algorithm, adjusting the spatial resolution based on feature point matching to obtain a spatially consistent registration dataset, including: A multi-temporal dataset is obtained, and feature points of each temporal image are extracted using a feature point detection algorithm to obtain a feature point set; Based on the feature point set, the scale-invariant feature transformation algorithm is used to match feature points and determine matching point pairs; If the number of matching point pairs exceeds a preset threshold, then erroneous matching points are removed using a random sampling consensus algorithm to obtain optimized matching point pairs. Based on the optimized matching point pairs, the geometric transformation matrix of each temporal image is calculated, and geometric correction parameters are generated; By adjusting the spatial resolution of images at different time phases using geometric correction parameters, spatially consistent registered images are obtained. For the registered images, calculate the registration accuracy index and determine whether the registration dataset meets the preset accuracy requirements; If the registration accuracy does not reach the preset threshold, return to the optimized matching point pair step, perform geometric correction again, and obtain the updated registration dataset.

4. The method according to claim 1, characterized in that, The process involves extracting radiometric characteristics from the registration dataset, using a radiometric correction algorithm to unify the radiometric standards of different sensors, and obtaining a radiometrically consistent correction dataset, including: The raw radiometric values ​​are obtained from the registration dataset, and the radiometric features are determined by the feature extraction algorithm. The radiation characteristics are processed using a preset radiation correction algorithm to unify the radiation standards of different sensors and obtain the corrected radiation value. If the corrected radiation value deviates from the preset standard deviation of radiation by more than a threshold, the correction parameters are adjusted through an iterative optimization algorithm to obtain the optimized radiation value. A radiation consistency model is constructed based on the optimized radiation values, and a correction dataset is generated through data fusion technology.

5. The method according to claim 1, characterized in that, Based on the corrected dataset, a time-series database is constructed, integrating features of landform, vegetation health, soil temperature, and groundwater level changes to obtain a multi-dimensional feature dataset, including: Landform, vegetation health, soil temperature and groundwater level data were obtained from the calibration dataset. Time series analysis methods were used to construct a multidimensional time series database to obtain a multidimensional time series dataset. Based on the multidimensional time series dataset, feature extraction methods were used to separate land surface morphology features, vegetation health index, soil temperature data and groundwater level changes to obtain independent feature sets. For independent feature sets, if the correlation between landform features and vegetation health index is lower than a preset threshold, principal component analysis is used to extract the main change features and obtain the principal feature set. Based on the main feature set, data fusion technology is used to integrate landform features, vegetation health index, soil temperature data and groundwater level changes to generate a fused feature dataset. For the fused feature dataset, if the deviation between groundwater level changes and soil temperature data exceeds a preset threshold, the feature weights are adjusted using a linear regression algorithm to obtain an optimized feature dataset. Based on the optimized feature dataset, cluster analysis was used to determine the dynamic correlation pattern between landform features and vegetation health index, resulting in a dynamic correlation dataset. For dynamically correlated datasets, data storage technology is used to persist feature data that integrates changes in landform, vegetation health, soil temperature, and groundwater level, resulting in a multidimensional feature dataset.

6. The method according to claim 1, characterized in that, For the multidimensional feature dataset, principal component analysis is used to extract the main patterns of landform changes, resulting in landform feature vectors, including: Time series analysis methods are used to separate time series data of landform changes from multidimensional feature datasets to obtain landform time series sets; Principal component analysis algorithm is used to extract the main change patterns from the landform time series set and obtain the main feature vector of landform.

7. The method according to claim 1, characterized in that, The process involves calculating vegetation health status from a multidimensional feature dataset using a normalized vegetation index, and combining this with landform feature vectors to obtain a vegetation health distribution map, including: From the multidimensional feature dataset, data preprocessing methods are used to extract the spectral data and landform feature data required for the normalized vegetation index, and a preprocessed feature set is obtained. Based on the preprocessed feature set, the normalized vegetation index algorithm is used to calculate the vegetation health status and generate a vegetation health index set. The normalized vegetation index formula is NDVI = (NIR - RED) / (NIR + RED), where NIR represents the near-infrared band value and RED represents the red band value. If the values ​​in the vegetation health index set are lower than the preset threshold, outliers are removed using data cleaning techniques to obtain a cleaned vegetation health index set. Based on the cleaned vegetation health index set, combined with the landform feature vector, feature vector fusion technology is used to integrate the vegetation health index and landform features to generate a fused feature set. Based on the fused feature set, a spatial interpolation algorithm is used to generate a spatial distribution pattern of vegetation health distribution, resulting in a vegetation health distribution map.

8. The method according to claim 1, characterized in that, The process involves analyzing soil temperature distribution using thermal infrared data based on a multidimensional feature dataset, fusing surface morphology feature vectors and vegetation health distribution maps to obtain a soil temperature feature map, including: Acquire thermal infrared data and landform feature vectors, remove noise through preprocessing, and obtain the first temperature dataset and the first terrain dataset; Principal component analysis algorithm is used to extract key feature vectors from the first temperature dataset and the first terrain dataset to obtain the first feature set; If the spatial resolution of the first feature set is lower than a preset threshold, the spatial resolution is enhanced by an interpolation algorithm to obtain the second feature set. The vegetation health index was extracted from the vegetation health distribution map and normalized to obtain the first vegetation dataset. A weighted fusion algorithm is used to fuse the second feature set with the first vegetation dataset to generate a comprehensive feature set; By analyzing the spatial distribution, a soil temperature distribution model is constructed based on a comprehensive feature set, resulting in a soil temperature feature map. If the variance of a local area in the soil temperature feature map exceeds a preset threshold, a smoothing filtering algorithm is used for optimization to obtain the final soil temperature feature map.

9. The method according to claim 1, characterized in that, If the soil temperature characteristic map shows an abnormal area, then the groundwater level change trend is analyzed through radar data to obtain a groundwater level change distribution map, including: Soil temperature data is acquired through sensors, a soil temperature feature map is constructed, and abnormal areas are identified. If the abnormal area in the soil temperature feature map exceeds the preset threshold, the groundwater level information is extracted from the radar data to obtain the original water level dataset. Time series analysis was used to process the raw water level dataset and calculate the trend of groundwater level changes. Based on the trend of groundwater level changes, and combined with spatial interpolation algorithms, a groundwater level change distribution map is generated.

10. The method according to claim 1, characterized in that, The integrated surface morphology feature vector, vegetation health distribution map, soil temperature feature map, and groundwater level change distribution map are used to construct a comprehensive index system using a weighted linear combination algorithm to obtain the evaluation results of the mine reclamation effect, including: Land surface morphology feature vectors, vegetation health distribution maps, soil temperature feature maps, and groundwater level change distribution maps are obtained from multi-source remote sensing data. Standardization processing is then used to obtain a feature dataset in a unified format. For the feature dataset, principal component analysis algorithm is used to extract the main components of land surface morphology features, vegetation health distribution, soil temperature features and groundwater level changes, and obtain a set of feature vectors after dimensionality reduction. Based on the feature vector set, a weighted linear combination algorithm is used to merge the feature vectors through a preset weight allocation scheme to generate a comprehensive index system. If the values ​​of each indicator in the comprehensive indicator system meet the preset threshold conditions, then a spatial interpolation algorithm is used to generate a spatial distribution map of the mine reclamation effect, and the spatial characteristics of the reclamation effect are obtained. Based on the spatial characteristics of reclamation effects, a clustering analysis algorithm is used to divide the spatial distribution map into regions, and the classification results of reclamation effects are obtained. The distribution patterns of reclamation effects are extracted from the classification results, and statistical analysis methods are used to generate quantitative evaluation results of reclamation effects.

Images