A mountainous land use monitoring method based on multi-source data fusion

Abstract

The present application relates to the field of mountain land monitoring, and more particularly to a mountain land use monitoring method based on multi-source data fusion, which comprises the following steps: S1, acquisition and preprocessing of multi-source data; S2, construction of DEM fine reconstruction training samples; S3, construction and training of the DEM fine reconstruction model; S4, input of the original low spatial resolution DEM data into the target DEM fine reconstruction model, and output of high-resolution DEM data by the target DEM fine reconstruction model; S5, extraction of topographic factors; S6, construction and fusion of multi-source data; S7, construction of a mountain land use classification model and application of a target classification model. The present application effectively compensates for the shadow area in the mountainous area by using the terrain constraint capability of DEM and the advantage of SAR being unaffected by light, and improves the overall accuracy and stability in the monitoring of mountain land use by fusing different features of SAR images, optical remote sensing images and DEM data.

Description

Technical Field

[0001] This invention relates to the field of land monitoring in mountainous areas, and in particular to a method for monitoring land use in mountainous areas based on multi-source data fusion. Background Technology

[0002] Mountainous areas have dramatic topographic relief, steep slopes, and complex aspects. During the acquisition of optical remote sensing images, the effects of solar altitude angle and topographic shading can easily lead to mountain shadows. Mountain shadows can cause distortion or even loss of spectral information of ground features in remote sensing images, seriously affecting the accuracy and stability of land cover classification in mountainous areas.

[0003] The multispectral optical remote sensing images provided by the Sentinel-2 satellite have high spatial resolution and rich spectral information, and have been widely used for land cover classification. However, the reflectance is significantly reduced in shaded areas of mountainous regions, and the spectral characteristics do not match the actual land cover types, leading to confusion and misclassification of Sentinel-2-based classification methods in shaded areas.

[0004] To mitigate the impact of Sentinel-2 shadowing, existing studies typically incorporate Sentinel-1 synthetic aperture radar (SAR) data. Sentinel-1 employs microwave imaging, offering all-weather, all-day observation capabilities, unaffected by lighting conditions, and capable of acquiring stable backscatter information even in shadowed areas. However, SAR images generally suffer from speckle noise, and their ability to distinguish land cover types is significantly influenced by terrain undulations and imaging geometry. Using Sentinel-1 alone as supplementary data is insufficient to obtain accurate classification results for mountainous areas.

[0005] Digital elevation models (DEMs) can characterize terrain relief features and are essential foundational data for terrain correction, shading analysis, and terrain factor extraction. In land cover classification in mountainous areas, introducing DEMs and their derived terrain factors such as slope and aspect helps improve the classification results for shaded areas. However, most existing DEMs are low to medium spatial resolution products, which cannot accurately describe subtle terrain changes in complex mountainous areas, limiting their role in shading compensation and terrain-aided classification.

[0006] In summary, existing land use classification and related technologies in mountainous areas have the following main shortcomings:

[0007] 1. Optical remote sensing images are severely affected by shadows cast by mountainous areas;

[0008] Classification methods based on optical images such as Sentinel-2 suffer from spectral information loss or distortion in shadowed areas of mountainous regions. Traditional spectral features are difficult to accurately reflect the true attributes of ground features, leading to a decrease in classification accuracy.

[0009] 2. Sentinel-1 data is difficult to effectively distinguish complex land cover types on its own;

[0010] Although Sentinel-1 data is unaffected by lighting and can be used to compensate for information in shadowed areas, its backscattering characteristics are affected by multiple factors such as terrain, angle of incidence, and surface roughness, and it is prone to noise interference and class confusion when used alone.

[0011] 3. The existing DEM has insufficient spatial resolution, making it difficult to support shadow compensation and fine classification of mountainous areas;

[0012] Commonly used large-area DEM products suffer from low spatial resolution and blurred terrain details, resulting in insufficient accuracy of derived factors such as slope and aspect, making it difficult to effectively serve shadow analysis and land cover classification.

[0013] 4. Multi-source data fusion classification methods rely insufficiently on terrain information;

[0014] Existing Sentinel-1 and Sentinel-2 fusion classification methods focus primarily on spectral and backscattering features, lacking effective modeling of fine terrain structures and making it difficult to fully utilize terrain information to constrain and compensate for shadowed areas.

[0015] Therefore, in order to obtain accurate land cover types in mountainous areas, how to improve the spatial resolution and topographic representation of mountain DEMs in the absence of high spatial resolution DEM data, and effectively integrate them with multi-source data such as Sentinel-1 and Sentinel-2 to alleviate the shadowing effect in mountainous areas and improve the accuracy of land use classification, has become an urgent technical problem to be solved. Summary of the Invention

[0016] To address the aforementioned technical problems, this invention provides a method for monitoring land use in mountainous areas based on multi-source data fusion. It utilizes the terrain constraint capability of DEM and the advantage of SAR being unaffected by illumination to effectively compensate for shadowed areas in mountainous regions. By fusing the different features of SAR imagery, optical remote sensing imagery, and DEM data, it improves the overall accuracy and stability of land use monitoring in mountainous areas.

[0017] This invention provides a method for monitoring land use in mountainous areas based on multi-source data fusion, employing the following technical solution:

[0018] A method for monitoring land use in mountainous areas based on multi-source data fusion includes the following steps:

[0019] S1. Acquisition and preprocessing of multi-source data;

[0020] Acquire multi-source data from different time phases, including SAR images, optical remote sensing images, and DEM data;

[0021] Multi-source data are preprocessed to obtain preprocessed SAR images, preprocessed optical remote sensing images, and preprocessed DEM data.

[0022] Construction of training samples for S2 and DEM fine-reconstruction;

[0023] The preprocessed DEM data is divided into multiple DEM sub-blocks according to a preset window size;

[0024] Downsampling multiple DEM sub-blocks yields downsampled DEM sub-blocks;

[0025] Construct training samples for refined DEM reconstruction, use preprocessed DEM data as the target output for self-supervised training, and use downsampled DEM sub-blocks as model input;

[0026] Construction and training of S3 and DEM refined reconstruction models;

[0027] S31. Construct an initial refined reconstruction model of the DEM;

[0028] S32. Construct the loss function;

[0029] S33. Train the initial DEM fine reconstruction model based on the DEM fine reconstruction training samples to obtain the target DEM fine reconstruction model.

[0030] S4. Input the original low spatial resolution DEM data into the target DEM fine reconstruction model, and the target DEM fine reconstruction model outputs high resolution DEM data.

[0031] S5. Extraction of terrain factors;

[0032] Topographic factors are extracted from high-resolution DEM data, and these topographic factors are used to characterize the complex topographic structure features of mountainous areas.

[0033] S6. Construction and fusion of multi-source data;

[0034] S61. Extracting optical spectral features for characterizing land cover features based on preprocessed optical remote sensing images;

[0035] S62. Calculate the spectral index characteristics based on optical spectral features;

[0036] S63. Microwave remote sensing features are extracted based on SAR image data;

[0037] S64. Integrate optical spectral features, spectral index features, microwave remote sensing features, and topographic factors to construct a multi-source feature set for land use classification in mountainous areas;

[0038] S7. Construction of land use classification model in mountainous areas and application of target classification model;

[0039] A land use classification model for mountainous areas was constructed, and the model was trained based on a multi-source feature set to obtain a target classification model that distinguishes land cover types.

[0040] The multi-source data to be identified is input into the target classification model for classification, generating a distribution map of land use categories in mountainous areas.

[0041] Preferably, the DEM fine reconstruction model includes an input layer, a feature extraction module, a residual feature learning module, an upsampling reconstruction module, and an output layer.

[0042] Preferably, the loss function in S32 includes:

[0043] Reconstruction error loss term It is used to measure the numerical difference between the initial DEM fine reconstruction model and the reference digital elevation model, so as to constrain the elevation accuracy of the initial DEM fine reconstruction model;

[0044] Terrain slope constraint loss term Based on the gradient information of the preprocessed DEM data, the slope variation of the reconstruction results and the reference data is constrained to maintain the continuity of terrain edges and slopes;

[0045] Terrain curvature constraint loss term Based on the second-order variation information of preprocessed DEM data, the curvature characteristics of the reconstruction results and reference data are constrained to suppress unreal terrain abrupt changes and enhance the consistency of terrain structure.

[0046] The formula for calculating the self-supervised loss function is as follows:

[0047] ;

[0048] in, For self-supervised loss function, This represents the reconstruction error loss term. This represents the terrain slope constraint loss term. The terrain curvature constraint loss term is represented by α and β, respectively. and The weighting coefficients.

[0049] Preferably, the terrain factors include:

[0050] Elevation factor, used to characterize the absolute elevation information of the Earth's surface;

[0051] Slope factor, used to characterize the degree of inclination of the earth's surface;

[0052] The aspect factor, which represents the aspect by sine and cosine components, eliminates the discontinuity problem caused by the periodicity of the aspect angle.

[0053] The topographic relief factor is used to characterize the magnitude of elevation variation within a local area.

[0054] The terrain location index factor is used to characterize the relative positional relationship of pixels in a local terrain environment.

[0055] Preferably, the preset elevation factor is: , Represented as in pixels Elevation value at the location;

[0056] The formula for calculating the slope factor is:

[0057] ;

[0058] in, For slope factor, and These represent the first-order rates of change of elevation in the east-west and north-south directions, respectively.

[0059] The formula for calculating the aspect factor is as follows:

[0060] ;

[0061] in, For slope aspect factor, and These represent the first-order rates of change of elevation in the east-west and north-south directions, respectively.

[0062] The formula for calculating the aspect factor is decomposed into sine and cosine components, which are expressed as follows:

[0063] ;

[0064] ;

[0065] in, For the sinusoidal component, For cosine components;

[0066] The formula for calculating the terrain relief factor is as follows:

[0067] ;

[0068] in, For topographic relief factor, These represent the maximum and minimum elevation values ​​within the neighborhood window, respectively.

[0069] The formula for calculating the terrain location index factor is as follows:

[0070] ;

[0071] in, It is a topographic location index factor. Elevation factor Represented by pixels The average elevation value within the neighborhood window centered on the center.

[0072] Preferably, the optical spectral characteristics include blue light band, green light band, red light band, and near-infrared band;

[0073] The spectral index features include the normalized vegetation index, the normalized water index, the enhanced vegetation index, and the soil-regulated vegetation index.

[0074] The microwave remote sensing features include backscattering features under different polarization modes.

[0075] Preferably, the formula for calculating the normalized vegetation index is:

[0076] ;

[0077] in, Normalized Difference Vegetation Index (NDVI) Indicates near-infrared reflectivity. Indicates the reflectivity in the red light band;

[0078] The formula for calculating the normalized water index is as follows:

[0079] ;

[0080] in, The normalized water index, Indicates the reflectivity in the green light band. Indicates near-infrared reflectance;

[0081] The formula for calculating the enhanced vegetation index is as follows:

[0082] ;

[0083] in, To enhance the vegetation index, Indicates the reflectivity of the blue light band. , , , This is an empirical coefficient;

[0084] The formula for calculating the soil-regulated vegetation index is as follows:

[0085] ;

[0086] in, To regulate the vegetation index in the soil, As a soil regulator, Indicates near-infrared reflectivity. This indicates the reflectivity in the red light band.

[0087] The beneficial effects of this invention are as follows:

[0088] 1. This invention utilizes the terrain constraint capability of DEM and the advantage of SAR being unaffected by illumination to effectively compensate for shadowed areas in mountainous regions. By fusing the different features of SAR imagery, optical remote sensing imagery, and DEM data, it improves the overall accuracy and stability of land use monitoring in mountainous areas.

[0089] 2. In the process of constructing the loss function, this invention combines reconstruction error loss, terrain slope constraint loss and terrain curvature constraint loss, which can improve elevation accuracy and spatial continuity in the DEM fine reconstruction model and effectively maintain the consistency of terrain structure, especially suitable for mountainous areas with complex terrain.

[0090] 3. By integrating optical spectral features, spectral index features, microwave remote sensing features, and topographic factor features extracted from high-resolution DEM data, this invention can enhance the ability to distinguish different land surface types in complex mountainous environments, and significantly improve the accuracy and robustness of land use classification in mountainous areas.

[0091] 4. This invention, through multi-source feature fusion and the introduction of topographic factors, not only improves the classification model's ability to identify water bodies, forests, grasslands, cultivated land, and artificial construction land, but also ensures the consistency between the classification results and the actual topographic structure. It can be widely applied in fields such as land use surveys in mountainous areas, resource management, and ecological environment monitoring. By providing high-precision, highly spatially continuous geographic information data, it provides reliable technical support for land planning and ecological protection. Attached Figure Description

[0092] Figure 1 This is a flowchart illustrating an embodiment of this application;

[0093] Figure 2 This is an example of DEM fine-scale reconstruction training samples for embodiments of this application;

[0094] Figure 3 This is a framework diagram of the refined DEM reconstruction model according to an embodiment of this application;

[0095] Figure 4This is a sample image of high-resolution DEM data output from the refined reconstruction model of the target DEM in an embodiment of this application.

[0096] Figure 5 This is a distribution map of land use categories in mountainous areas generated according to an embodiment of this application. Detailed Implementation

[0097] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0098] This embodiment discloses a method for monitoring land use in mountainous areas based on multi-source data fusion, which is illustrated using Sentinel-1 images (SAR images), Sentinel-2 remote sensing images (optical remote sensing images), and SRTM digital elevation model data (DEM data) of a certain region in the same year as examples.

[0099] Reference Figure 1 Specifically, this method for monitoring land use in mountainous areas based on multi-source data fusion includes the following steps:

[0100] S1. Acquisition and preprocessing of multi-source data;

[0101] Acquire multi-source data from different time phases, including SAR images, optical remote sensing images, and DEM data;

[0102] Multi-source data are preprocessed to obtain preprocessed SAR images, preprocessed optical remote sensing images, and preprocessed DEM data.

[0103] Specifically, the goal is to obtain SRTM digital elevation model data for a certain area. SRTM digital elevation model data is abbreviated as DEM data, which is data with a spatial resolution of 30 meters.

[0104] Simultaneously, Sentinel-1 and Sentinel-2 remote sensing images of a certain region for the same year were acquired. The Sentinel-1 image is a SAR image, and the Sentinel-2 remote sensing image is an optical remote sensing image. Both the SAR image and the optical remote sensing image are images with a spatial resolution of 10 meters.

[0105] Preprocessing was performed on DEM data, SAR images, and optical remote sensing images respectively. The preprocessing of DEM data included unifying the coordinate system, cropping, and checking for elevation anomalies. The preprocessed DEM data was obtained after preprocessing.

[0106] Preprocessing of SAR images includes cropping, filtering and denoising, missing data filling, median synthesis, etc., and the preprocessed SAR image is obtained after preprocessing.

[0107] Preprocessing of optical remote sensing images includes radiometric calibration, atmospheric correction, cloud removal, cropping, median composite, etc., and the resulting image is a preprocessed optical remote sensing image.

[0108] Construction of training samples for S2 and DEM fine-reconstruction;

[0109] The preprocessed DEM data is divided into multiple DEM sub-blocks according to a preset window size;

[0110] Mean degradation is performed on multiple DEM sub-blocks to obtain DEM sub-block data;

[0111] We construct training samples for refined DEM reconstruction, using preprocessed DEM data as the target output for self-supervised training and DEM sub-block data as the model input.

[0112] Reference Figure 2 Specifically, based on the DEM data in step S1, the DEM data with a spatial resolution of 30 meters is divided into multiple DEM sub-blocks with an overlap of 0.02. Each DEM sub-block is a block of 192×192 pixels.

[0113] The DEM sub-blocks are then downsampled by a factor of 3, i.e., mean degradation, to obtain downsampled DEM sub-blocks. The downsampled DEM sub-block data is 64×64 pixel data with a spatial resolution of 90 meters, which is used to construct low spatial resolution DEM data.

[0114] For DEM data, a spatial resolution of 30 meters is considered high resolution, while a spatial resolution of 90 meters is considered low resolution.

[0115] Using downsampled DEM sub-blocks as model input samples and preprocessed DEM data as the target output for self-supervised training, a refined DEM reconstruction training sample is constructed.

[0116] Construction and training of S3 and DEM refined reconstruction models;

[0117] The construction of the DEM fine-grained reconstruction model uses an image super-resolution reconstruction model as a tool, such as... Figure 3 As shown, it is a framework diagram of the refined reconstruction model of DEM.

[0118] Reference Figure 3 Specifically, step S3 includes the following steps:

[0119] S31. Construct an initial refined reconstruction model of the DEM;

[0120] An initial DEM fine reconstruction model is constructed based on the DEM fine reconstruction training samples obtained in step S2, and a single-image super-resolution reconstruction model based on residual learning structure, EDSR model, is selected as the basic model for digital elevation model reconstruction.

[0121] The DEM fine-scale reconstruction model includes:

[0122] 1) Input layer, used to receive low-resolution DEM data;

[0123] 2) Feature extraction module, used to perform preliminary feature mapping on the input low-resolution DEM data, mapping the original elevation information to a high-dimensional feature space;

[0124] 3) Residual feature learning module, which includes multiple residual units. Each residual unit extracts terrain spatial features through convolution operation and nonlinear activation function, and realizes the superposition of input features and output features through residual connection, so as to enhance the ability to express terrain details and maintain the stability of the overall terrain structure.

[0125] 4) Upsampling reconstruction module, used to map and reconstruct low-resolution features (90-meter spatial resolution in this embodiment) into DEM data of target resolution (30-meter spatial resolution in this embodiment). The upsampling reconstruction module uses pixel rearrangement to improve spatial resolution, so that the feature map is mapped from 64×64 pixels to 192×192 pixels.

[0126] 5) Output layer, used to generate DEM data with high spatial resolution (30-meter spatial resolution in this embodiment).

[0127] S32. Construct the loss function;

[0128] The construction of the loss function for the DEM fine-grained reconstruction model achieves high-quality reconstruction of the model without manual annotation, and constructs a self-supervised loss function based on terrain constraints.

[0129] Self-supervised loss functions include:

[0130] Reconstruction error loss term It is used to measure the numerical difference between the initial DEM fine reconstruction model and the reference digital elevation model, so as to constrain the elevation accuracy of the initial DEM fine reconstruction model;

[0131] Terrain slope constraint loss term Based on the gradient information of the preprocessed DEM data, the slope variation of the reconstruction results and the reference data is constrained to maintain the continuity of terrain edges and slopes;

[0132] Terrain curvature constraint loss term Based on the second-order variation information of preprocessed DEM data, the curvature characteristics of the reconstruction results and reference data are constrained to suppress unreal terrain abrupt changes and enhance the consistency of terrain structure.

[0133] The formula for calculating the self-supervised loss function is:

[0134] ;

[0135] in, For self-supervised loss function, This represents the reconstruction error loss term. This represents the terrain slope constraint loss term. The terrain curvature constraint loss term is represented by α and β, respectively. and The weighting coefficients.

[0136] S33. Train the initial DEM fine reconstruction model based on the DEM fine reconstruction training samples to obtain the target DEM fine reconstruction model.

[0137] The DEM fine reconstruction training samples are input into the initial DEM fine reconstruction model to train the initial DEM fine reconstruction model. After training the initial DEM fine reconstruction model, the target DEM fine reconstruction model is obtained.

[0138] Specifically, the training process is as follows:

[0139] 1) Training data organization method;

[0140] In this embodiment, training samples are input into the initial DEM fine reconstruction model in batches, with each training batch containing multiple DEM fine reconstruction training samples. Each low-resolution input sample has a pixel size of 64×64, corresponding to a spatial resolution of 90 meters; each high-resolution reference sample has a pixel size of 192×192, corresponding to a spatial resolution of 30 meters.

[0141] 2) Training parameter settings;

[0142] In this embodiment, the model training parameters are set as follows: training batch size is 8, and initial learning rate is set to 1×10. -4 The model was trained for 200 epochs, and an adaptive moment estimation optimization algorithm was used to update the parameters of the initial DEM refinement reconstruction model. These parameters can be adjusted based on the training data size and computational resources available.

[0143] 3) Loss function calculation and backpropagation.

[0144] In each training iteration, low-resolution digital DEM data is input into the initial DEM fine-reconstruction model to obtain the corresponding high-resolution reconstruction results.

[0145] Based on the self-supervised loss function constructed in step S32, the following losses are calculated respectively:

[0146] Reconstruction error loss is used to constrain the consistency of elevation values;

[0147] Terrain slope constraint loss is used to constrain the first-order spatial gradient change of the terrain;

[0148] Terrain curvature constraint loss is used to constrain the second-order spatial variation characteristics of terrain.

[0149] The reconstruction error loss, terrain slope constraint loss, and terrain curvature constraint loss are weighted and summed according to preset weights to obtain the total loss value of the current training iteration. The parameters of the initial DEM fine reconstruction model are then updated through backpropagation.

[0150] 4) Model convergence and saving.

[0151] After completing the preset number of training rounds, when the loss function of the initial DEM fine reconstruction model tends to stabilize, the model training is considered to have converged. At this point, the training is complete, and the target DEM fine reconstruction model is obtained.

[0152] In addition, during the training process, the parameters of the initial DEM fine reconstruction model are periodically saved so that they can be called later for DEM model reconstruction prediction.

[0153] S4. Input the original low spatial resolution DEM data into the target DEM fine reconstruction model, and the target DEM fine reconstruction model outputs high resolution DEM data.

[0154] Reference Figure 4 The original low spatial resolution DEM data is DEM data with a spatial resolution of 30 meters, and the high spatial resolution DEM data is DEM data with a spatial resolution of 10 meters.

[0155] S5. Extraction of terrain factors;

[0156] Because mountainous areas have large topographic relief, it is difficult to accurately distinguish different land surface types by relying solely on optical remote sensing images and SAR images. Therefore, topographic factors are introduced to enhance the classification model's adaptability to complex terrain conditions.

[0157] Topographic factors are extracted from high-resolution DEM data, and these topographic factors are used to characterize the complex topographic structure features of mountainous areas.

[0158] Topographic factors include:

[0159] Elevation factor, used to characterize the absolute elevation information of the Earth's surface;

[0160] Preset elevation factor is , Represented as in pixels Elevation value at the location;

[0161] Slope factor, used to characterize the degree of inclination of the earth's surface;

[0162] The formula for calculating the slope factor is:

[0163] ;

[0164] in, For slope factor, and These represent the first-order rates of change of elevation in the east-west and north-south directions, respectively.

[0165] The aspect factor, which represents the aspect by sine and cosine components, eliminates the discontinuity problem caused by the periodicity of the aspect angle.

[0166] The formula for calculating the aspect factor is:

[0167] ;

[0168] in, For slope aspect factor, and These represent the first-order rates of change of elevation in the east-west and north-south directions, respectively.

[0169] The formula for calculating the aspect factor is decomposed into sine and cosine components, which are expressed as follows:

[0170] ;

[0171] ;

[0172] in, For the sinusoidal component, For cosine components;

[0173] The topographic relief factor is used to characterize the magnitude of elevation variation within a local area.

[0174] The formula for calculating the terrain relief factor is:

[0175] ;

[0176] in, For topographic relief factor, These represent the maximum and minimum elevation values ​​within the neighborhood window, respectively.

[0177] The terrain location index factor is used to characterize the relative positional relationship of pixels in a local terrain environment.

[0178] The formula for calculating the terrain location index factor is:

[0179] ;

[0180] in, It is a topographic location index factor. Elevation factor Represented by pixels The average elevation value within the neighborhood window centered on the center.

[0181] S6. Construction and fusion of multi-source data;

[0182] Because mountainous areas have complex land cover types, relying on a single data source is insufficient to accurately reflect the differences in characteristics between different land cover types. Therefore, by integrating optical spectral features, spectral index features, microwave remote sensing features, and topographic factors, the adaptability and classification accuracy of the classification model to complex mountainous environments can be improved.

[0183] Specifically, S6 includes the following steps:

[0184] S61. Extracting optical spectral features for characterizing land cover features based on preprocessed optical remote sensing images;

[0185] Optical spectral characteristics are used to characterize the spectral reflectance properties of different land surface types, including the blue, green, red, and near-infrared bands.

[0186] Among them, the blue light band, green light band, and red light band are visible light bands.

[0187] S62. Calculate the spectral index characteristics based on optical spectral features;

[0188] To enhance the separability between different land cover types, multiple spectral index features were further constructed based on optical band characteristics.

[0189] Spectral index characteristics include normalized vegetation index, normalized water index, enhanced vegetation index, and soil-regulated vegetation index;

[0190] The normalized difference in vegetation index (NDVI) is used to characterize vegetation cover, and its calculation formula is as follows:

[0191] ;

[0192] in, Normalized Difference Vegetation Index (NDVI) Indicates near-infrared reflectivity. Indicates the reflectivity in the red light band;

[0193] The Normalized Difference Water Index (NDDI) is used to enhance water body information, and its calculation formula is as follows:

[0194] ;

[0195] in, The normalized water index, Indicates the reflectivity in the green light band. Indicates near-infrared reflectance;

[0196] The Enhanced Vegetation Index (EVI) is used to reduce the influence of atmospheric and soil background and improve the ability to identify areas with high vegetation cover. Its calculation formula is as follows:

[0197] ;

[0198] in, To enhance the vegetation index, Indicates the reflectivity of the blue light band. , , , This is an empirical coefficient;

[0199] The soil-modified vegetation index is used to reduce the impact of bare soil background on the vegetation index. Its calculation formula is as follows:

[0200] ;

[0201] in, To regulate the vegetation index in the soil, As a soil regulator, Indicates near-infrared reflectivity. This indicates the reflectivity in the red light band.

[0202] S63. Microwave remote sensing features are extracted based on SAR image data;

[0203] Backscattering coefficients are extracted from SAR image data as microwave remote sensing features. These features include backscattering characteristics under different polarization modes, specifically vertical-vertical (VV) polarization backscattering characteristics and vertical-horizontal (VH) polarization backscattering characteristics. Microwave remote sensing features are used to reflect the characteristics of the Earth's surface structure and its differences in scattering microwave signals, enabling stable acquisition of surface information under cloudy, foggy, and complex terrain conditions.

[0204] S64. The optical spectral features, spectral index features, microwave remote sensing features, and topographic factors obtained in step S5 are fused to construct a multi-source feature set for land use classification in mountainous areas.

[0205] S7. Construction of land use classification model in mountainous areas and application of target classification model;

[0206] A land use classification model for mountainous areas was constructed, and the model was trained based on a multi-source feature set to obtain a target classification model that distinguishes land cover types.

[0207] The multi-source data to be identified is input into the target classification model for classification, generating a distribution map of land use categories in mountainous areas.

[0208] Reference Figure 5 Specifically, step S7 includes the following steps:

[0209] S71, Construction of training samples for classification. In this embodiment, based on manual interpretation and existing land use data, sample data of multiple land cover types in a certain area are obtained. The land cover types include water bodies, forest land, shrubland, grassland, cultivated land and artificial construction land. The above land cover type samples are merged to construct a training sample set, and each training sample in the training sample set is assigned a corresponding land use category label.

[0210] S72, Feature Sampling and Sample Set Construction: Based on the multi-source feature set constructed in step S6, extract the corresponding multi-source feature values ​​at the training sample locations to construct a labeled training sample set with category labels.

[0211] S73, Construction of the Mountainous Land Use Classification Model: In this embodiment, a random forest classification algorithm based on decision tree ensemble is used to construct the mountainous land use classification model. The random forest classification model constructs multiple decision trees and integrates and votes on the classification results of each decision tree to obtain the final classification result. In a specific implementation, the parameters of the random forest classification model are set as follows: the number of decision trees is 200; a random seed is set to ensure the stability and repeatability of the model training process. The above parameters can be adjusted according to the size of the study area and the number of samples.

[0212] S74, Training of the Mountainous Land Use Classification Model: Input the label training sample set constructed in step S72 into the random forest classification model to train the random forest classification model, so that the random forest classification model learns the mapping relationship between multi-source features and land use types. After training is completed, the target classification model is obtained.

[0213] S75, Application of the target classification model: Input the multi-source feature set constructed in step S6 into the target classification model to perform pixel-by-pixel classification of a certain area and generate a distribution map of land use categories in mountainous areas.

[0214] The land use classification results are output in raster form, with different pixels corresponding to different land use types. It can be widely used in application scenarios such as land use surveys in mountainous areas, ecological environment monitoring, land resource management and planning.

[0215] Although the invention has been described herein with reference to several illustrative embodiments, it should be understood that many other modifications and implementations can be devised by those skilled in the art, which will fall within the scope and spirit of the principles disclosed herein. More specifically, various variations and modifications can be made to the components and / or layout of the subject matter arrangement within the scope of the disclosure, drawings, and claims. Besides variations and modifications to the components and / or layout, other uses will be apparent to those skilled in the art.

Claims

1. A method for monitoring land use in mountainous areas based on multi-source data fusion, characterized in that, Includes the following steps: S1. Acquisition and preprocessing of multi-source data; Acquire multi-source data from different time phases, including SAR images, optical remote sensing images, and DEM data; Multi-source data are preprocessed to obtain preprocessed SAR images, preprocessed optical remote sensing images, and preprocessed DEM data. Construction of training samples for S2 and DEM fine-reconstruction; The preprocessed DEM data is divided into multiple DEM sub-blocks according to a preset window size; Downsampling multiple DEM sub-blocks yields downsampled DEM sub-blocks; Construct training samples for refined DEM reconstruction, use preprocessed DEM data as the target output for self-supervised training, and use downsampled DEM sub-blocks as model input; Construction and training of S3 and DEM refined reconstruction models; S31. Construct an initial refined reconstruction model of the DEM; S32. Construct the loss function; S33. Train the initial DEM fine reconstruction model based on the DEM fine reconstruction training samples to obtain the target DEM fine reconstruction model. S4. Input the original low spatial resolution DEM data into the target DEM fine reconstruction model, and the target DEM fine reconstruction model outputs high resolution DEM data. S5. Extraction of terrain factors; Topographic factors are extracted from high-resolution DEM data, and these topographic factors are used to characterize the complex topographic structure features of mountainous areas. S6. Construction and fusion of multi-source data; S61. Extracting optical spectral features for characterizing land cover features based on preprocessed optical remote sensing images; S62. Calculate the spectral index characteristics based on optical spectral features; S63. Microwave remote sensing features are extracted based on SAR image data; S64. Integrate optical spectral features, spectral index features, microwave remote sensing features, and topographic factors to construct a multi-source feature set for land use classification in mountainous areas; S7. Construction of land use classification model in mountainous areas and application of target classification model; A land use classification model for mountainous areas was constructed, and the model was trained based on a multi-source feature set to obtain a target classification model that distinguishes land cover types. The multi-source data to be identified is input into the target classification model for classification, generating a distribution map of land use categories in mountainous areas.

2. The method for monitoring land use in mountainous areas based on multi-source data fusion according to claim 1, characterized in that, The refined DEM reconstruction model includes an input layer, a feature extraction module, a residual feature learning module, an upsampling reconstruction module, and an output layer.

3. The method for monitoring land use in mountainous areas based on multi-source data fusion according to claim 1, characterized in that, The loss function in S32 includes: Reconstruction error loss term It is used to measure the numerical difference between the initial DEM fine reconstruction model and the reference digital elevation model, so as to constrain the elevation accuracy of the initial DEM fine reconstruction model; Terrain slope constraint loss term Based on the gradient information of the preprocessed DEM data, the slope variation of the reconstruction results and the reference data is constrained to maintain the continuity of terrain edges and slopes; Terrain curvature constraint loss term Based on the second-order variation information of preprocessed DEM data, the curvature characteristics of the reconstruction results and reference data are constrained to suppress unreal terrain abrupt changes and enhance the consistency of terrain structure. The formula for calculating the self-supervised loss function is as follows: ； in, For self-supervised loss function, This represents the reconstruction error loss term. This represents the terrain slope constraint loss term. The terrain curvature constraint loss term is represented by α and β, respectively. and The weighting coefficients.

4. The method for monitoring land use in mountainous areas based on multi-source data fusion according to claim 1, characterized in that, The terrain factors include: Elevation factor, used to characterize the absolute elevation information of the Earth's surface; Slope factor, used to characterize the degree of inclination of the earth's surface; The aspect factor, which represents the aspect by sine and cosine components, eliminates the discontinuity problem caused by the periodicity of the aspect angle. The topographic relief factor is used to characterize the magnitude of elevation variation within a local area. The terrain location index factor is used to characterize the relative positional relationship of pixels in a local terrain environment.

5. A method for monitoring land use in mountainous areas based on multi-source data fusion according to claim 4, characterized in that, Preset elevation factor is , Represented as in pixels Elevation value at the location; The formula for calculating the slope factor is: ； in, For slope factor, and These represent the first-order rates of change of elevation in the east-west and north-south directions, respectively. The formula for calculating the aspect factor is as follows: ； in, For slope aspect factor, and These represent the first-order rates of change of elevation in the east-west and north-south directions, respectively. The formula for calculating the aspect factor is decomposed into sine and cosine components, which are expressed as follows: ； ； in, For the sinusoidal component, For cosine components; The formula for calculating the terrain relief factor is as follows: ； in, For topographic relief factor, These represent the maximum and minimum elevation values ​​within the neighborhood window, respectively. The formula for calculating the terrain location index factor is as follows: ； in, It is a topographic location index factor. Elevation factor Represented by pixels The average elevation value within the neighborhood window centered on the center.

6. A method for monitoring land use in mountainous areas based on multi-source data fusion according to claim 1, characterized in that, The optical spectral characteristics include blue light band, green light band, red light band, and near-infrared band; The spectral index features include the normalized vegetation index, the normalized water index, the enhanced vegetation index, and the soil-regulated vegetation index. The microwave remote sensing features include backscattering features under different polarization modes.

7. A method for monitoring land use in mountainous areas based on multi-source data fusion according to claim 6, characterized in that, The formula for calculating the normalized vegetation index is as follows: ； in, Normalized Difference Vegetation Index (NDVI) Indicates near-infrared reflectivity. Indicates the reflectivity in the red light band; The formula for calculating the normalized water index is as follows: ； in, The normalized water index, Indicates the reflectivity in the green light band. Indicates near-infrared reflectance; The formula for calculating the enhanced vegetation index is as follows: ； in, To enhance the vegetation index, Indicates the reflectivity of the blue light band. , , , This is an empirical coefficient; The formula for calculating the soil-regulated vegetation index is as follows: ； in, To regulate the vegetation index in the soil, As a soil regulator, Indicates near-infrared reflectivity. This indicates the reflectivity in the red light band.

Images