Method for monitoring temporal changes of surface features based on gee cloud platform and remote sensing embedded vector

Abstract

This invention relates to the fields of satellite remote sensing technology and geographic information application technology, and discloses a method for monitoring temporal changes of ground features based on the GEE cloud platform and remote sensing embedding vectors. Based on annual synthetic data of satellite image embedding vectors stored on the GEE (Google Earth Engine) cloud platform, a ground feature extraction function based on sample points and a similarity threshold is defined using the JavaScript programming language. This function obtains a similarity map by calculating the dot product of the sample point embedding vector and the full map embedding vector, and then performs thresholding to obtain a binary mask for ground features. By executing this function on the set start and end years respectively, two ground feature masks are obtained. The two masks are compared and calculated to accurately identify areas of ground feature expansion and contraction, and finally a unified change map is generated and exported. This invention solves the problems of high dependence on spectral features, large sample requirements, and complex processes in traditional ground feature change monitoring, and provides an efficient and automated temporal change analysis scheme.

Description

Technical Field

[0001] This invention relates to the field of satellite remote sensing technology and its application technology, and in particular to a method for monitoring temporal changes of ground features based on the GEE cloud platform and remote sensing embedded vectors. Background Technology

[0002] Land cover change monitoring is one of the core applications in the field of remote sensing, and it is of great significance for urban expansion, deforestation, desertification, and crop yield estimation. Traditional monitoring methods, such as thresholding based on spectral indices (e.g., NDVI) or supervised classification methods (e.g., random forests, support vector machines), heavily rely on the spectral information of remote sensing images.

[0003] However, these methods have many drawbacks: 1) interference from the phenomena of different spectra of the same object and the same spectra of different objects is serious; 2) a large number of high-quality training samples are required, and manual sampling is costly; 3) the image preprocessing process is complex, including atmospheric correction, cloud removal, image stitching, etc., making it difficult to achieve large-scale, long-term automated processing.

[0004] Google Earth Engine (GEE) is a cloud-based computing platform specifically designed for processing satellite imagery, providing massive amounts of remote sensing data and powerful computing capabilities. In recent years, deep learning-based remote sensing image embedding vector models (such as Satellite Embedding) have been introduced into the GEE platform. This model compresses high-dimensional image data into low-dimensional feature vectors, enabling more robust representation of ground feature information.

[0005] However, there is currently a lack of an automated monitoring process that utilizes the GEE platform and image-embedded vector data to achieve large-scale, long-term land cover changes with only a small number of sample points. Summary of the Invention

[0006] The purpose of this invention is to provide a method for monitoring temporal changes of ground features based on the GEE cloud platform and remote sensing embedded vectors, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for monitoring temporal changes of ground features based on the GEE cloud platform and remote sensing embedded vectors includes the following steps:

[0009] Step 1: In the GEE cloud platform, use the JavaScript programming language to define the geometric boundaries of the study area, the vector data of sample points used to identify target features, the start year, the end year, and the similarity threshold;

[0010] Step 2: In the GEE cloud platform, define a ground feature mask acquisition function using the JavaScript programming language. This function takes the year, sample points, and similarity threshold as input parameters.

[0011] Step 3: In the ground feature mask acquisition function, load the annual satellite image embedding vector image set provided by the GEE cloud platform, and filter to obtain the embedding vector mosaic image for that year according to the input year parameter;

[0012] Step 4: In the ground feature mask acquisition function, the sample point data is used to sample the mosaic image of the embedding vector to obtain the embedding vector value of the sample area; by calculating the dot product between the sample embedding vector value and the mosaic image of the whole image, and taking the average value, a similarity map representing the similarity score is generated.

[0013] Step 5: In the feature mask acquisition function, the similarity map is compared with the similarity threshold to generate a binary mask map of the target feature for that year, where a pixel value of 1 represents the target feature and a pixel value of 0 represents a non-target feature.

[0014] Step 6: Call the ground feature mask acquisition function respectively, taking the start year and end year as parameters, to obtain the ground feature binary mask image for the start year and the ground feature binary mask image for the end year;

[0015] Step 7: In the GEE cloud platform, perform pixel-level comparison operations on the binary mask images of ground features for the start and end years to generate a ground feature change map.

[0016] Step 8: Call the Google Drive interface of the GEE cloud platform to export the land feature change map to the local machine for analysis, according to the specified spatial resolution and geographic coordinate system.

[0017] As a further improvement to this technical solution: the annual satellite image embedding vector image set mentioned in step three is the 'GOOGLE / SATELLITE_EMBEDDING / V1 / ANNUAL' dataset provided by the GEE platform.

[0018] As a further improvement to this technical solution: the process of calculating the average embedding similarity in step four specifically includes: sampling the satellite embedded image using the ground feature sample points to obtain the embedding vectors of all sample points; traversing the embedding vectors of all sample points, converting the embedding vector of each sample point into an image format, and calculating the dot product with the satellite embedded image to obtain the single-point similarity map of the sample point; calculating the pixel-level average value of all single-point similarity maps obtained in the step to obtain the final similarity map.

[0019] As a further improvement to this technical solution: the land feature change map described in step seven includes at least three change type codes: a) Stable: the pixel value is coded as 0, indicating that the pixel is either a non-land feature or a target land feature in both the start and end years; b) Expanding: the pixel value is coded as 1, defined as the region where the pixel value of the binary mask in the start year is 0 and the pixel value of the binary mask in the end year is 1; c) Shrinking: the pixel value is coded as 2, defined as the region where the pixel value of the binary mask in the start year is 1 and the pixel value of the binary mask in the end year is 0.

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] 1. This invention is highly efficient and robust: It uses pre-trained remote sensing embedding vectors to replace the original spectral bands, which has a stronger ability to represent ground objects and is less affected by changes in illumination, atmosphere and seasons. High-precision ground object extraction can be achieved with only a small number of sample points.

[0022] 2. Simplified calculation logic of the present invention: The present invention uses dot product to calculate similarity, replacing the complex classifier training process, which is fast, has clear logic, and is easy to implement.

[0023] 3. The invention has a high degree of automation: the entire process is implemented through JavaScript programming on the GEE cloud platform. From data loading, similarity calculation, change analysis to result export, it can run fully automatically and is suitable for large-scale, long-term monitoring tasks.

[0024] 4. The invention provides clear information on changes: This method not only monitors the extent of ground features, but also clearly distinguishes between two types of changes: "expansion" and "contraction," providing more detailed data support for decision-making.

[0025] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it according to the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Specific embodiments of the present invention are given in detail below with reference to the accompanying drawings. Attached Figure Description

[0026] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0027] Figure 1 This is a system flowchart of the method of the present invention;

[0028] Figure 2 This is a binary mask image of the target ground features in the starting year (2018) of the embodiment;

[0029] Figure 3 This is a binary mask image of the target ground features in the termination year (2022) of the embodiment;

[0030] Figure 4 This is the final (2018-2022) land cover change map generated in the example, where red indicates expansion and green indicates contraction. Detailed Implementation

[0031] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are for illustrative purposes only and are not intended to limit the scope of the invention. The invention is described more specifically in the following paragraphs by way of example with reference to the accompanying drawings. It should be noted that the drawings are in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of the present invention.

[0032] like Figure 1 As shown, this invention discloses a method for monitoring temporal changes of ground features based on the GEE cloud platform and remote sensing embedding vectors. This method quickly identifies expansion and contraction regions of target ground features by comparing the similarity of embedding vectors from images of two different years, and includes the following steps.

[0033] Before running this embodiment on the GEE platform, the operator must first import the data, including an ee.FeatureCollection of the study area (in this example, the area of ​​Chifeng City, filename chifeng) and an ee.FeatureCollection of ground feature sample points (in this example, sandy sample points, filename desert_points).

[0034] Step 1: In the GEE cloud platform, use the JavaScript programming language to define the geometric boundaries of the study area, the vector data of sample points for identifying target features, the start year, the end year, and the similarity threshold. This step uses a global configuration object (CONFIG) to centrally manage all parameters. Simultaneously, configure the map environment to load and display the study area and sample points. Example code for specific configuration settings is shown below:

[0035] / / =================== Parameter Configuration (CONFIGURATION) ==================== / / --- / / --- 1. The following variables are imported above / / ​​--- (e.g., 'chifeng' and 'desert_points') var CONFIG = { / / --- Basic Settings --- year_start:2018, / / [Start] Year year_end: 2022, / / [End] Year scale: 10, / / Spatial resolution (m) threshold: 0.80, / / Similarity threshold (0-1, higher is more similar) / / Study area boundary variable name study_area:chifeng, / / Feature sample point variable name feature_points: desert_points, / / Feature type naming / / (e.g., "Desert", "Crop", "Water") feature_name:'Desert', / / Export folder settings exportFolder: 'GEE_Feature_Change_Exports';================================================================= / / --- 0. Environment Setup---var geometry = CONFIG.study_area.geometry();Map.centerObject(geometry, 8);Map.addLayer(geometry,{color: 'red'}, 'Study Area', true, 0.5);Map.addLayer(CONFIG.feature_points, {color: 'yellow'}, CONFIG.feature_name + 'Sample Points');Map.setOptions('SATELLITE');

[0036] Step 2: In the GEE cloud platform, define a feature mask acquisition function using the JavaScript programming language. This function takes the year, sample points, and similarity threshold as input parameters. This step defines a function named `getFeatureMask`. Step 3: In the feature mask acquisition function, load the annual satellite image embedding vector image set provided by the GEE cloud platform, and filter to obtain the mosaic image of that year based on the input year parameter. In this embodiment, the specific dataset loaded is GOOGLE / SATELLITE_EMBEDDING / V1 / ANNUAL. A time range is constructed using the input year parameter, and the images for that year are filtered and mosaicked. The specific function setting example code is shown below:

[0037] var getFeatureMask = function(year, sample_points, threshold, scale) { / / 1. Get the image var startDate = ee.Date.fromYMD(year, 1, 1); var endDate =startDate.advance(1, 'year'); var mosaic = ee.ImageCollection('GOOGLE / SATELLITE_EMBEDDING / V1 / ANNUAL') .filter(ee.Filter.date(startDate, endDate)) .mosaic();

[0038] Step 3: In the feature mask acquisition function, the sample point data is used to sample the mosaic image of the embedding vector to obtain the embedding vector value of the sample region; by calculating the dot product between the sample embedding vector value and the mosaic image of the whole image, and taking the average value, a similarity map representing the similarity score is generated; in this embodiment, sample regions are first used to sample the sample points; then, map is used to traverse all samples, and the dot product (dotProduct) of each sample vector with the whole image is calculated to obtain multiple single-point similarity maps; finally, the mean() method is used to calculate the pixel-level average value of all single-point similarity maps to obtain the final similarity map. The specific function setting example code is shown below:

[0039] / / 2. Calculate the similarity map var sampleEmbeddings = mosaic.sampleRegions({ collection:sample_points, scale: scale}); var sampleDistances = ee.ImageCollection(sampleEmbeddings.map(function (f) { var arrayImage = ee.Image(f.toArray(mosaic.bandNames())).arrayFlatten([mosaic.bandNames()]); var dotProduct =arrayImage.multiply(mosaic).reduce('sum').rename('similarity'); returndotProduct;})); var meanDistance = sampleDistances.mean();

[0040] Step four: In the feature mask acquisition function, the similarity map is compared with the similarity threshold to generate a binary mask map of the target feature for that year, where a pixel value of 1 represents a target feature and a pixel value of 0 represents a non-target feature. In this embodiment, the greater than operator is used to compare the meanDistance map with the configured threshold to generate a binary mask map similarPixels. The specific function settings are shown in the following example code:

[0041] / / 3. Visualize the binarized results var similarPixels = meanDistance.gt(threshold); / / 4. Return all results return { mosaic: mosaic, / / Original image similarity: meanDistance, / / Similarity map mask: similarPixels / / Binary map (1=target feature, 0=non-feature)};};

[0042] Step 5: Call the feature mask acquisition function, taking the start year and end year as input parameters, to obtain the binary feature mask image for the start year and the binary feature mask image for the end year. This step calls the getFeatureMask function twice, passing CONFIG.year_start and CONFIG.year_end respectively, and extracts the binary mask images mask_start and mask_end from the returned objects. The specific function settings are illustrated in the code below:

[0043] / / --- 2. Calculate data for two years --- print('Calculating ' + CONFIG.year_start + 'year' + CONFIG.feature_name + 'data...'); var result_start = getFeatureMask(CONFIG.year_start, CONFIG.feature_points, CONFIG.threshold, CONFIG.scale); print('Calculating ' + CONFIG.year_end + 'year' + CONFIG.feature_name + 'data...'); var result_end = getFeatureMask(CONFIG.year_end, CONFIG.feature_points, CONFIG.threshold, CONFIG.scale); / / Extract binary images for two years var mask_start = result_start.mask; / / Features for the start year (1=yes, 0=no) var mask_end = result_end.mask; / / Features for the end year (1=yes, 0=no)

[0044] Step Six: In the GEE cloud platform, perform pixel-level comparison operations on the binary mask images of the starting and ending years to generate a change map. In this embodiment, the change map (changeMap) contains three types of encoding: a) Expansion (encoding 1): identified by mask_end.gt(mask_start) (i.e., 0 -> 1). b) Contraction (encoding 2): identified by mask_start.gt(mask_end) (i.e., 1 -> 0). c) Stability (encoding 0): Pixels that have not undergone the above two changes remain at 0 (i.e., 0 -> 0 or 1 -> 1). This embodiment also includes code for visualizing the change results on a map. The specific function settings are shown in the following example code:

[0045] / / --- 3. Calculate Changes--- / / Logic: / / Expansion (newly added features): 0 at the start, 1 at the end / / Contraction (disappeared features): 1 at the start, 0 at the end / / 1 = Expansion (newly added features), 0 = other var expansion = mask_end.gt(mask_start); / / 1 = Contraction (disappeared features), 0 = other var contraction = mask_start.gt(mask_end); / / Create a uniform change layer: / / 0: Stable (non-feature -> non-feature or feature -> feature) / / 1: Expansion (non-feature -> target feature) / / 2: Contraction (target feature -> non-feature) var changeMap = ee.Image(0) / / Default value 0 (stable).where(expansion, 1) / / Mark expansion area as 1.where(contraction, 2); / / Mark contraction area as 2 / / --- 4. Visualize Changes--- var year_range = CONFIG.year_start + '-' + CONFIG.year_end; / / Visualize features over two years var binaryVisStart = {palette: ['#00000000', 'blue']}; / / Start year (blue) var binaryVisEnd = {palette: ['#00000000', 'cyan']}; / / End year (cyan) Map.addLayer(mask_start.selfMask().clip(geometry),binaryVisStart, CONFIG.feature_name + 'region(' + CONFIG.year_start + ')', false); Map.addLayer(mask_end.selfMask().clip(geometry),binaryVisEnd, CONFIG.feature_name + 'region(' + CONFIG.year_end +')', false); / / Visualize the change layer var changeVisParams = { min: 1,max: 2, palette: ['red', 'green'] / / 1: Expand (red), 2: Shrink (green)};Map.addLayer(changeMap).selfMask().clip(geometry), changeVisParams,CONFIG.feature_name + 'Change(' + year_range + ') [Red=Expansion, Green=Contraction]');

[0046] Step 7: Call the Google Drive interface of the GEE cloud platform to export the land feature change map to the local machine for analysis, based on the specified spatial resolution and geographic coordinate system. In this embodiment, the Export.image.toDrive interface is called to export the land feature change map (changeMap, converted to Byte type) generated in Step 7 to Google Drive. Additionally, this embodiment also exports the desert extent for the start and end years as a reference base map.

[0047] / / 4. Execute export var file_prefix_change = year_range + '_' +CONFIG.feature_name.replace( / \s / g, '_'); / / For example, "2018-2022_Desert" / / Task 1: Export raw image data containing 3 specified bands (end year) Export.image.toDrive({ image: bands_to_export, description: file_prefix_change + '_rgb_end_year', / / [!!] Modified: variable name folder:CONFIG.exportFolder, fileNamePrefix: file_prefix_change + '_rgb_end_year', / / [!!] Modified: variable name region: geometry, scale:CONFIG.scale, fileFormat: 'GeoTIFF', maxPixels: 1e13}); / / Task 2: Export change label image (0=stable, 1=expanded, 2=shrink)Export.image.toDrive({image: label_to_export, description: file_prefix_change + '_change_label', / / [!!] Modified: variable name folder: CONFIG.exportFolder, fileNamePrefix: file_prefix_change + '_change_label', / / [!!] Modified: variable name region: geometry, scale: CONFIG.scale, fileFormat: 'GeoTIFF', maxPixels: 1e13}); / / Task 3: Export the desert region for the starting year (2018) (1=desert, 0=non-desert)var file_prefix_start = CONFIG.year_start +'_' + CONFIG.feature_name.replace( / \s / g, '_');Export.image.toDrive({ image: mask_start.to_export, description ... folder: CONFIG.exportFolder, fileNamePrefix: file_prefix_change + '_change_label', / / [!!] Modified: variable name region: geometry, scale: CONFIG.scale, fileFormat: 'GeoTIFF', maxPixels: 1e13}); / / Task 3: Export the desert region for the starting year (2018) (1=desert, 0=non-destoByte(), / / mask_start is the binary image of 2018 description: file_prefix_start + '_mask', folder:CONFIG.exportFolder, fileNamePrefix: file_prefix_start + '_mask', region: geometry, scale: CONFIG.scale, fileFormat: 'GeoTIFF', maxPixels: 1e13}); / / Task 4: Export the desert region of the end year (2022) (1=desert, 0=non-desert) var file_prefix_end = CONFIG.year_end + '_' +CONFIG.feature_name.replace( / \s / g, '_'); Export.image.toDrive({image: mask_end.toByte(), / / mask_end is the binary image of 2022 description: file_prefix_end + '_mask', folder: CONFIG.exportFolder, fileNamePrefix: file_prefix_end + '_mask', region: geometry,scale: CONFIG.scale, fileFormat: 'GeoTIFF', maxPixels: 1e13});.

[0048] Then, by performing the above steps through the Google Earth Engine cloud platform, a map of ground feature changes can be automatically generated, and the results can be exported locally for further analysis and mapping using software such as Python or ArcGIS.

[0049] Figure 2 and Figure 3The examples show the binary mask images of sandy land in Chifeng City for the start year (2018) and end year (2022). Figure 4 The image shows the spatial mapping results of sandy land changes in Chifeng City from 2018 to 2022 in this embodiment. It can be seen that during the monitoring period, the reduction of sandy land in the study area of ​​Chifeng City was quite significant, mainly concentrated in the northern part of the Horqin Sandy Land and most of the Hunshandake Sandy Land. The monitoring results of sandy land shrinkage are consistent with the implementation of the "Horqin and Hunshandake Sandy Land Elimination Campaign" in the Three-North Shelterbelt Project in this region, reflecting the significant achievements Chifeng City has made in sandy land management through the ecological governance project of the Three-North Shelterbelt Project, and the substantial improvement of the ecological environment within Chifeng City.

[0050] Meanwhile, sand dune expansion was also detected in some areas, mainly scattered in the core area of ​​the Horqin Sandy Land. This may reflect the situation during shrub coppicing. To avoid the problem of shrubs weakening their growth and reducing their windbreak and sand-fixing capabilities after a certain number of years, the reserve uses brush cutters to coppic the shrubs to promote new shoot growth, help the shrubs grow larger, and restore their growth advantage. This situation is likely to result in sand dune expansion detected in remote sensing data.

[0051] From a dynamic evolution perspective, comparing the overall extent of sand dune expansion and contraction, the trend of sand dune contraction was significantly stronger than that of sand dune expansion during this period. This result is consistent with the trend of sand dune ecological restoration under the Three-North Shelterbelt Project, indicating that the land cover temporal change monitoring method based on the GEE cloud platform and remote sensing embedded vectors provided in this invention can accurately and reasonably monitor large-scale land cover temporal changes.

[0052] This completes the rapid, automated dynamic monitoring and analysis of sandy land features within the monitoring area from 2018 to 2022, as described in this embodiment.

[0053] The key and protected aspects of this invention lie in: by utilizing the satellite embedded image dataset (GOOGLE / SATELLITE_EMBEDDING / V1 / ANNUAL) of the Google Earth Engine (GEE) cloud platform, combined with a sample point-driven semantic similarity calculation method, high-precision, long-term dynamic monitoring and change detection of specific land features (such as deserts, bare soil, etc.) over large-scale areas can be achieved. This method abandons the limitations of traditional methods relying on single spectral indices or supervised classification models, instead utilizing the spatial semantic consistency of deep learning embedded features to construct an end-to-end technical workflow for land feature identification—change analysis—results derivation, providing reliable data support for regional ecological degradation or restoration assessments, land cover change studies, and other related research.

[0054] The advantages or beneficial effects of this invention are as follows:

[0055] 1. Existing studies mostly rely on traditional remote sensing indices such as the Normalized Difference in Surface Exposure Index (NDPI) and the Normalized Difference in Vegetation Index (NDVEI) or supervised classification methods based on a limited number of bands for land cover identification. This makes it difficult to accurately distinguish between spectrally similar but semantically different land cover types (such as bare soil versus desert, bare rock versus buildings). This invention, for the first time, introduces satellite embedding on the GEE platform. Utilizing its high-dimensional semantic feature space, it achieves accurate extraction of target land cover through the dot product similarity between sample points and pixels, significantly improving the robustness and generalization ability of the identification, and is particularly suitable for bare surface types with complex spectral features and blurred boundaries.

[0056] 2. To address issues such as temporal inconsistency, cloud contamination, and mosaic color difference in remote sensing images over large-scale regions, this invention directly utilizes annual composite mosaic image products provided by GEE, avoiding registration errors and radiometric inconsistencies caused by the manual stitching of multiple images in traditional methods. Furthermore, ground feature identification based on annual composite data effectively smooths out short-term noise interference, improving the temporal comparability and result stability of long-term time-series analysis.

[0057] 3. This invention constructs a standardized logical framework for detecting land cover changes: by comparing the binary masks of land cover from two years, it automatically identifies two key change types: "expansion" (new land cover) and "contraction" (land cover disappearance), and generates a uniformly coded change map (0=stable, 1=expansion, 2=contraction). This method is logically clear, highly interpretable, and facilitates subsequent mapping, statistical analysis, and ecological process modeling, avoiding the "false change" misjudgment problem that often occurs in traditional change detection.

[0058] 4. This invention is entirely based on the GEE cloud architecture. All calculations (including sample feature extraction, similarity calculation, mask generation, and change analysis) are completed on the server side, which greatly reduces the local computing resource requirements. At the same time, it supports one-click export of high-resolution change label maps and corresponding year embedded image bands to Google Drive, which makes it convenient for users to conduct in-depth analysis, model training, or result visualization locally, realizing efficient collaboration between "cloud processing and local application".

[0059] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0060] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Those skilled in the art can readily implement the present invention based on the accompanying drawings and the above description. However, any modifications, alterations, or variations made by those skilled in the art without departing from the scope of the present invention, using the technical content disclosed above, are equivalent embodiments of the present invention. Furthermore, any modifications, alterations, or variations made to the above embodiments based on the essential technology of the present invention are still within the protection scope of the present invention.

Claims

1. A method for monitoring temporal changes of ground features based on the GEE cloud platform and remote sensing embedded vectors, characterized in that, Includes the following steps: Step 1: In the GEE cloud platform, use the JavaScript programming language to define the geometric boundaries of the study area, the vector data of sample points used to identify target features, the start year, the end year, and the similarity threshold in the form of spatial boundary coordinates. Step 2: In the GEE cloud platform, define a ground feature mask acquisition function using the JavaScript programming language. This function takes the year, sample points, and similarity threshold as input parameters. Step 3: In the ground feature mask acquisition function, load the annual satellite image embedding vector image set provided by the GEE cloud platform, and filter to obtain the embedding vector mosaic image for that year according to the input year parameter; Step 4: In the ground feature mask acquisition function, the sample point data is used to sample the mosaic image of the embedding vector to obtain the embedding vector value of the sample area; by calculating the dot product between the sample embedding vector value and the mosaic image of the whole image, and taking the average value, a similarity map representing the similarity score is generated. Step 5: In the feature mask acquisition function, the similarity map is compared with the similarity threshold to generate a binary mask map of the target feature for that year, where a pixel value of 1 represents the target feature and a pixel value of 0 represents a non-target feature. Step 6: Call the ground feature mask acquisition function respectively, taking the start year and end year as parameters, to obtain the ground feature binary mask image for the start year and the ground feature binary mask image for the end year; Step 7: In the GEE cloud platform, perform pixel-level comparison operations on the binary mask images of ground features for the start and end years to generate a ground feature change map. Step 8: Call the Google Drive interface of the GEE cloud platform to export the land feature change map to the local machine for analysis, according to the specified spatial resolution and geographic coordinate system.

2. The method for monitoring temporal changes of ground features based on the GEE cloud platform and remote sensing embedded vectors according to claim 1, characterized in that, The annual satellite imagery embedding vector imagery set mentioned in step three is the 'GOOGLE / SATELLITE_EMBEDDING / V1 / ANNUAL' dataset provided by the GEE platform.

3. The method for monitoring temporal changes of ground features based on the GEE cloud platform and remote sensing embedded vectors according to claim 1, characterized in that, The process of calculating the average embedding similarity in step four specifically includes: sampling the satellite embedded image using the ground feature sample points to obtain the embedding vectors of all sample points; traversing the embedding vectors of all sample points, converting the embedding vector of each sample point into an image format, and calculating the dot product with the satellite embedded image to obtain the single-point similarity map of that sample point; calculating the pixel-level average value of all single-point similarity maps obtained in the step to obtain the final similarity map.

4. The method for monitoring temporal changes of ground features based on the GEE cloud platform and remote sensing embedded vectors according to claim 1, characterized in that, The land feature change map described in step seven contains at least three change type codes: a) Stable: A cell value of 0 indicates that the cell is either a non-landscape or a target landscape in both the start and end years; b) Expansion: The pixel value is encoded as 1, defined as the region where the pixel value of the starting year binary mask is 0 and the pixel value of the ending year binary mask is 1. c) Shrinkage: The pixel value is encoded as 2, and is defined as the region where the pixel value of the starting year binary mask is 1 and the pixel value of the ending year binary mask is 0.

Images