Geospatial analysis method based on remote sensing data

Abstract

The invention discloses a geographic space analysis method based on remote sensing data. Comprising the following steps: S1, acquiring remote sensing data; s2, transmitting remote sensing data; s3, performing feature extraction on the data; s4, the extracted features are processed; s5, performing fusion preparation on the radar data and the picture data; s6, visualization of the geographic space three-dimensional diagram is carried out; according to the method, the collected remote sensing data is processed through the edge cloud layer, preprocessing, image classification and feature extraction are performed on the remote sensing data, subsequent fusion of radar data and picture data is facilitated, and the definition and accuracy of fused spatial data are improved; processing the extracted three-dimensional features and two-dimensional features, and judging whether the features are the same feature point or not through a super-strong factor; weight values are distributed according to the regional coincidence degree values, radar data and picture data are fused through feature hierarchy fusion, and a three-dimensional diagram is displayed through a visualization module.

Description

Technical Field

[0001] The present invention belongs to the technical field of geospatial analysis, and more specifically, particularly relates to a geospatial analysis method based on remote sensing data. Background Art

[0002] Remote sensing technology refers to the collection of electromagnetic wave reflection or radiation information far away from the target object through sensors, and obtaining information about the target object or area by processing this data. Remote sensing data mainly includes satellite images, aerial photography, laser radar (LiDAR) data, etc. These data are usually multidimensional, including multiple dimensions such as space, spectrum, and time, and can provide comprehensive information about surface materials and environment.

[0003] Remote sensing data is data of various natural information such as the ground, atmosphere, and water bodies obtained by satellites, aircraft, or other sensors. It can be in the form of remote sensing images, remote sensing radar data, etc. The acquisition method of remote sensing data can realize the efficient detection and monitoring of surface features, providing an important source of information for GIS spatial analysis.

[0004] Geographic Information System (GIS) spatial analysis is the process of storing, querying, analyzing and displaying geographic spatial data using computer technology, mathematical methods, etc. Through GIS spatial analysis, the regularity, interrelationships and spatiotemporal variation characteristics of geographical phenomena can be revealed, providing a scientific basis for decision-making.

[0005] Currently, remote sensing data and GIS spatial analysis have been widely used in urban planning, environmental monitoring, resource survey and other fields, and have also played an important role in agriculture, meteorology, geology and other industries. The continuous deepening of related research and the continuous advancement of technology have provided more possibilities for better use of remote sensing data for GIS spatial analysis.

[0006] However, there are some problems with the existing technology: the existing remote sensing data can only be spatially analyzed through one type of data after collection, which is prone to data errors and cannot accurately display spatial information. In addition, for the fusion of data in different formats, the existing technology generally adopts direct fusion processing, resulting in serious data information loss. Therefore, we propose a geospatial analysis method based on remote sensing data. Summary of the invention

[0007] In view of the problems existing in the prior art, the purpose of the present invention is to provide a geospatial analysis method based on remote sensing data, which collects spatial data by adopting radar data and image data, and fuses the radar data and image data, so as to accurately display spatial information and prevent data loss.

[0008] To achieve the above object, the present invention provides the following technical solution: A geospatial analysis method based on remote sensing data, comprising the following steps:

[0009] S1. Acquisition of remote sensing data: remote sensing data is acquired through satellites, drones or aerial photography. Remote sensing data includes radar and image data types;

[0010] S2. Transmission of remote sensing data: Satellites, drones or aerial photography transmit remote sensing data to the edge cloud layer through the communication module, and the edge cloud layer pre-processes the remote sensing data to obtain processed data;

[0011] S3. Extract features from the data: The edge cloud layer performs image classification on the processed data again, extracts classification data of different types of objects from the radar data and the picture data; then performs feature extraction on the extracted classification data, and the features obtained from the radar data are three-dimensional features, and the features obtained from the picture data are two-dimensional features;

[0012] S4, processing the extracted features: analyzing and processing the obtained three-dimensional features and two-dimensional features to obtain multiple super-strong factors in the three-dimensional features and the two-dimensional features, and then classifying the super-strong factors respectively, determining the distances of the multiple super-strong factors by pixel values ​​and position values, performing regional classification processing, and then matching the multiple super-strong factors in the same regional classification to obtain a regional overlap value;

[0013] S5. Prepare the fusion of radar data and image data: according to the obtained regional overlap value, weight allocation is implemented for radar data and image data. A larger weight value is allocated to a region with a higher overlap value, and a smaller weight value is allocated to a region with a lower overlap value. Then, radar data and image data are fused through feature level fusion to form a three-dimensional geographic space map.

[0014] S6. Visualization of geospatial three-dimensional map: Display the geospatial three-dimensional map through a visualization module, and integrate raster data into the geospatial three-dimensional map. The raster data is used to display and store geographic data of the geospatial three-dimensional map.

[0015] Specifically, the satellite in S2 uses high-frequency wireless signals to transmit remote sensing data, the drone or the aerial photography uses an Internet of Things network to transmit remote sensing data, and the edge cloud layer preprocesses the remote sensing data including noise removal, standardization, radiation correction, geometric correction, atmospheric correction, and cloud and fog removal.

[0016] Specifically, the noise removal adopts the Gaussian filtering algorithm, and the calculation formula is as follows:

[0017] ,

[0018] in, is the coordinate of the filtered image data information The pixel value at is the coordinate in the original image data information The pixel value of is the Gaussian kernel in The weight value at and The Gaussian kernel is and The radius in the direction, and , is the error coefficient generated by the Gaussian kernel during calculation;

[0019] Gaussian Kernel The calculation formula is as follows:

[0020] ,

[0021] The calculation formula is as follows:

[0022] ,in, and is the distance from the center of the Gaussian kernel, is the standard deviation of the Gaussian kernel;

[0023] The standardization adopts linear standardization, which maps all pixel values ​​to the range of [0,1], and its formula is as follows:

[0024] ,

[0025] in, is a pixel value in the image, is the minimum value of all pixels in the image, is the maximum value of all pixels in the image.

[0026] Specifically, the radiation correction is used to eliminate the image radiation value errors caused by sensor characteristics, atmospheric influences and other external factors, and restore the true radiation information of ground targets. The geometric correction is used to convert the remote sensing image from the sensor coordinate system to the map coordinate system to eliminate the geometric distortion in the image. The atmospheric correction is used to remove the influence of aerosols, water vapor, etc. in the atmosphere on the remote sensing image and restore the ground reflectivity. The declouding and defogging processing is used to remove the influence of clouds and haze in the image.

[0027] Specifically, the radiation correction is converted into radiation brightness or reflectivity through the calibration function of the sensor:

[0028] The radiance is calculated as follows:

[0029] ,

[0030] in, is the radiance, is the digital value of the image, is the gain factor, is the bias, i.e. the dark current of the sensor;

[0031] The calculation formula of reflectivity is:

[0032] ,

[0033] in, is the surface reflectivity, is the radiance, is the solar radiation intensity, is the sun’s zenith angle;

[0034] The calculation formula of the geometric correction is as follows:

[0035] ,

[0036] in, are the pixel coordinates in the original image, are the target map coordinates, are the transformation parameters calculated from the ground control points;

[0037] The atmospheric correction is calculated as follows:

[0038] Atmospheric influence formula:

[0039] ,

[0040] in, is the ground reflectivity radiance after removing the atmospheric influence, is the calculated radiance, is the atmospheric transmittance,

[0041] The calculation formula of reflectivity after removing the atmospheric influence is:

[0042] ,

[0043] in, is the reflectivity of the Earth’s surface;

[0044] The calculation formula for declouding and defogging is as follows:

[0045] Go to the cloud for processing:

[0046] ,

[0047] in, is the pixel value of the shortwave infrared band, is the set threshold, It is the cloud detection result, where 1 indicates cloud and 0 indicates non-cloud. To remove the cloud, the cloud area needs to be marked, and then the image information of other areas is used for interpolation and filling.

[0048] Dehazing Processing:

[0049] ,

[0050] in, is the pixel value of the observed image, is the pixel value of the clear image after defogging, that is, the actual ground reflectivity, is the transmittance, which describes the attenuation of light in the atmosphere, It is a panoramic illumination. In order to remove fog, the transmittance is optimized To restore the image ,Common dehazing methods include dark channel prior;

[0051] The minimum value of the image in the dark channel prior is used to estimate the transmittance:

[0052] ,

[0053] in, is a constant set to 0.95, is an image In the color channel The pixel value at It's pixels The small neighborhood around it is estimated by , restore the clear image after dehazing .

[0054] Specifically, the image classification in S3 adopts the K-means clustering algorithm, and the steps of the K-means clustering algorithm are as follows:

[0055] Initialization: Random selection The initial center point of each cluster;

[0056] Assignment phase: Assign each data point to the nearest cluster center, typically using the Euclidean distance metric:

[0057] ,

[0058] in, is a data point, It is The center of the cluster, is the dimension of the data, and They are data points and The cluster center is The weight of the dimension;

[0059] Update phase: According to the allocation results, recalculate the center of each cluster:

[0060] ,

[0061] in, Belongs to the cluster The number of data points, Belongs to the cluster data points;

[0062] Iteration: Repeat the assignment phase and the update phase until the cluster center no longer changes or the predetermined number of iterations is reached, so that the radar data and image data can extract classification data of different types of objects.

[0063] Specifically, the calculation of feature extraction in S3 is as follows:

[0064] The two-dimensional features of the image data are calculated using SIFT feature extraction:

[0065] ,

[0066] in, is the Gaussian kernel, is an image, is the feature point;

[0067] The three-dimensional features of radar data are calculated using normal vectors:

[0068] First, calculate the center point of the neighborhood points:

[0069] ,

[0070] in, It is represented as the center point of the neighborhood point, that is, the average position of all neighborhood points. The position vectors of the points are summed and averaged; Represented as the first point in the neighborhood points, each point is a vector, usually represented as ,in is the dimension of the point; It is expressed as the number of neighborhood points, that is, the number of points considered;

[0071] Then, calculate the covariance matrix of the neighborhood points :

[0072] ,

[0073] in, It is expressed as a covariance matrix, which describes the distribution of neighborhood points in space and reflects the degree and direction of point dispersion; Represented as each neighborhood point With center point The difference between represents the deviation vector from the center point to each point; Representation is the transpose of the bias vector, which is multiplied by the bias vector itself to calculate the covariance of the point; It is expressed as the sum of the outer products of all deviation vectors, which is used to measure the distribution direction and degree of each point relative to the central point; It is expressed by dividing the sum of the outer products by , get the covariance matrix after mean;

[0074] The normal vector is the eigenvector corresponding to the smallest eigenvalue of the covariance matrix.

[0075] Specifically, the super strong factor in S4 is calculated as follows:

[0076] ,

[0077] in, is a constant used to adjust the intensity. is a two-dimensional feature point or a three-dimensional feature point, It is expressed as the calculated super strength factor;

[0078] All super factors are classified by calculating the distance using the Euclidean distance:

[0079] For any dimension Two points in and , the calculation formula of Euclidean distance is:

[0080] ,

[0081] in, Represented as a point and Point The Euclidean distance between two points represents the straight-line distance between them; Represents the sum index, indicating the sum from the first dimension to the dimension;

[0082] Represented as a point and Point In the Difference in dimension; Expressed as the square of the difference in each dimension; It is represented by summing the squares of all dimension differences and taking the square root to get the final Euclidean distance.

[0083] and When the value of is less than half the distance between the two nearest super strong factors, the point is included in the calculated super strong factor. If it exceeds half the distance between the two super strong factors, the point is not included in the calculated super strong factor.

[0084] Specifically, the steps for matching the multiple super factors in S4 are as follows:

[0085] Calculate the distance: For each feature point in the image data , calculate its distance to each feature point in the radar data ;

[0086] ,

[0087] Select K nearest neighbors: Select and query feature points The K target feature points closest to the descriptor;

[0088] Matching candidate points: By comparing the matching conditions of the K nearest neighbors, the most suitable matching point is finally selected;

[0089] ,in, Expressed as the first super factor matching distance, Expressed as the second super factor matching distance, ratio If it is less than 0.8, the match is considered good, otherwise it is considered wrong.

[0090] Specifically, the feature level fusion in S5 is as follows:

[0091] Set the two-dimensional feature to , the three-dimensional features are :

[0092] ,

[0093] in, is the fused feature vector, and are the feature vectors from different modes or sources, and is the weighting coefficient, indicating the relative importance of each feature, satisfying the constraints .

[0094] Technical effects and advantages of the present invention:

[0095] The present invention processes the collected remote sensing data through the edge cloud layer, performs preprocessing, image classification and feature extraction on the remote sensing data. Through the processing of the remote sensing data by the edge cloud layer, rapid analysis and processing of the remote sensing data can be achieved, and the operating pressure and complexity of the system can be reduced. The edge cloud layer extracts features from the radar data and the image data respectively, obtains the three-dimensional features of the radar data and the two-dimensional features of the image data, facilitates the subsequent fusion of the radar data and the image data, and improves the clarity and accuracy of the fused spatial data.

[0096] The extracted three-dimensional features and two-dimensional features are processed to obtain super-strong factors in the three-dimensional features and the two-dimensional features, that is, the most prominent points in the features, and the distances between multiple super-strong factors are determined according to the pixel value or position to achieve regional classification of the features, and then the three-dimensional features and the two-dimensional features are matched through the super-strong factors to determine the relationship between the three-dimensional features and the two-dimensional features, and to determine whether they are the same feature points;

[0097] The weight value is then assigned according to the regional overlap value. A larger weight value is assigned to a higher regional overlap value, and a smaller weight value is assigned to a lower regional overlap value. The radar data and the image data are then fused through feature-level fusion to form a three-dimensional geographic space map, that is, the features with large weight values ​​are fused to form a more refined three-dimensional map. For small weight values, direct fusion is used, and the feature requirements are not high, and they can be fused into a three-dimensional map. The three-dimensional map is then displayed through a visualization module, and raster data is fused into the three-dimensional geographic space map. The raster data is used to display and store geographic data in the three-dimensional geographic space map.

[0098] Further features and advantages of the present invention will become apparent from the following detailed description of exemplary embodiments of the present invention with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] Figure 1 It is a schematic diagram of the step flow provided by the present invention. DETAILED DESCRIPTION

[0100] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0101] like Figure 1 As shown, the geospatial analysis method based on remote sensing data provided by the embodiment of the present invention includes the following steps:

[0102] S1. Acquisition of remote sensing data: remote sensing data is acquired through satellites, drones or aerial photography. Remote sensing data includes radar and image data types;

[0103] S2. Transmission of remote sensing data: Satellites, drones or aerial photography transmit remote sensing data to the edge cloud layer through the communication module, and the edge cloud layer pre-processes the remote sensing data to obtain processed data;

[0104] S3. Extract features from the data: The edge cloud layer performs image classification on the processed data again, extracts classification data of different types of objects from the radar data and the picture data; then performs feature extraction on the extracted classification data, and the features obtained from the radar data are three-dimensional features, and the features obtained from the picture data are two-dimensional features;

[0105] S4, processing the extracted features: analyzing and processing the obtained three-dimensional features and two-dimensional features to obtain multiple super-strong factors in the three-dimensional features and the two-dimensional features, and then classifying the super-strong factors respectively, determining the distances of the multiple super-strong factors by pixel values ​​and position values, performing regional classification processing, and then matching the multiple super-strong factors in the same regional classification to obtain a regional overlap value;

[0106] S5. Prepare the fusion of radar data and image data: according to the obtained regional overlap value, weight allocation is implemented for radar data and image data. A larger weight value is allocated to a region with a higher overlap value, and a smaller weight value is allocated to a region with a lower overlap value. Then, radar data and image data are fused through feature level fusion to form a three-dimensional geographic space map.

[0107] S6. Visualization of geospatial three-dimensional map: Display the geospatial three-dimensional map through a visualization module, and integrate raster data into the geospatial three-dimensional map. The raster data is used to display and store geographic data of the geospatial three-dimensional map.

[0108] In this embodiment, preferably, the satellite in S2 uses a high-frequency wireless signal to transmit remote sensing data, the drone or the aerial photography uses an Internet of Things network to transmit remote sensing data, and the edge cloud layer preprocesses the remote sensing data including noise removal, standardization, radiation correction, geometric correction, atmospheric correction, and cloud and fog removal;

[0109] It should be noted that different remote sensing devices use different transmission methods when transmitting data. Choosing an appropriate transmission method can improve transmission efficiency and reduce transmission costs. Remote sensing image data is often affected by factors such as noise, atmospheric influence, and sensor errors. The preprocessing process aims to improve data quality and provide clear basic data for subsequent analysis.

[0110] In this embodiment, preferably, the noise removal adopts a Gaussian filtering algorithm, and the calculation formula is as follows:

[0111] ,

[0112] in, is the coordinate of the filtered image data information The pixel value at is the coordinate in the original image data information The pixel value of is the Gaussian kernel in The weight value at and The Gaussian kernel is and The radius in the direction, and , is the error coefficient generated by the Gaussian kernel during calculation;

[0113] Gaussian Kernel The calculation formula is as follows:

[0114] ,

[0115] The calculation formula is as follows:

[0116] ,in, and is the distance from the center of the Gaussian kernel, is the standard deviation of the Gaussian kernel;

[0117] The standardization adopts linear standardization, which maps all pixel values ​​to the range of [0,1], and its formula is as follows:

[0118] ,

[0119] in, is a pixel value in the image, is the minimum value of all pixels in the image, is the maximum value of all pixels in the image;

[0120] It should be noted that by filtering out noise from image information through the Gaussian filtering algorithm, the clarity of the image can be improved and more accurate image information can be obtained. In addition, by using linear normalization processing to map all pixel values ​​to [0,1], the size of the image can be reduced, thereby improving the efficiency of subsequent image processing.

[0121] In this embodiment, preferably, the radiation correction is used to eliminate the image radiation value error caused by sensor characteristics, atmospheric influence and other external factors, and restore the real radiation information of the ground target; the geometric correction is used to convert the remote sensing image from the sensor coordinate system to the map coordinate system to eliminate the geometric distortion in the image; the atmospheric correction is used to remove the influence of aerosols, water vapor, etc. in the atmosphere on the remote sensing image and restore the ground reflectivity; the cloud removal and defogging processing is used to remove the influence of clouds and haze in the image;

[0122] It should be noted that by processing images through radiation correction, geometric correction, atmospheric correction, and cloud and haze removal, the influence of factors such as noise, atmospheric influence, and sensor error can be reduced. The preprocessing process aims to improve data quality and provide clear basic data for subsequent analysis.

[0123] In this embodiment, preferably, the radiation correction is converted into radiation brightness or reflectivity through a calibration function of the sensor:

[0124] The radiance is calculated as follows:

[0125] ,

[0126] in, is the radiance, is the digital value of the image, is the gain factor, is the bias, i.e. the dark current of the sensor;

[0127] The calculation formula of reflectivity is:

[0128] ,

[0129] in, is the surface reflectivity, is the radiance, is the solar radiation intensity, is the sun’s zenith angle;

[0130] The calculation formula of the geometric correction is as follows:

[0131] ,

[0132] in, are the pixel coordinates in the original image, are the target map coordinates, are the transformation parameters calculated from the ground control points;

[0133] The atmospheric correction is calculated as follows:

[0134] Atmospheric influence formula:

[0135] ,

[0136] in, is the ground reflectivity radiance after removing the atmospheric influence, is the calculated radiance, is the atmospheric transmittance,

[0137] The calculation formula of reflectivity after removing the atmospheric influence is:

[0138] ,

[0139] in, is the reflectivity of the Earth’s surface;

[0140] The calculation formula for declouding and defogging is as follows:

[0141] Go to the cloud for processing:

[0142] ,

[0143] in, is the pixel value of the shortwave infrared band, is the set threshold, It is the cloud detection result, where 1 indicates cloud and 0 indicates non-cloud. To remove the cloud, the cloud area needs to be marked, and then the image information of other areas is used for interpolation and filling.

[0144] Dehazing Processing:

[0145] ,

[0146] in, is the pixel value of the observed image, is the pixel value of the clear image after defogging, that is, the actual ground reflectivity, is the transmittance, which describes the attenuation of light in the atmosphere, It is a panoramic illumination. In order to remove fog, the transmittance is optimized To restore the image ,Common dehazing methods include dark channel prior;

[0147] The minimum value of the image in the dark channel prior is used to estimate the transmittance:

[0148] ,

[0149] in, is a constant set to 0.95, is an image In the color channel The pixel value at It's pixels The small neighborhood around it is estimated by , restore the clear image after dehazing ;

[0150] It should be noted that the image processing realized by the above calculation formula can not only eliminate the radiation error of the sensor, correct the geometric distortion caused by different satellite orbits and viewing angles, reduce the influence of the atmosphere on remote sensing signals, and remove the influence of clouds and haze in the image; it can also improve the clarity of the image.

[0151] In this embodiment, preferably, the image classification in S3 adopts a K-means clustering algorithm, and the steps of the K-means clustering algorithm are as follows:

[0152] Initialization: Random selection The initial center point of each cluster;

[0153] Assignment phase: Assign each data point to the nearest cluster center, typically using the Euclidean distance metric:

[0154] ,

[0155] in, is a data point, It is The center of the cluster, is the dimension of the data, and They are data points and The cluster center is The weight of the dimension;

[0156] Update phase: According to the allocation results, recalculate the center of each cluster:

[0157] ,

[0158] in, Belongs to the cluster The number of data points, Belongs to the cluster data points;

[0159] Iteration: Repeat the allocation phase and the update phase until the cluster center no longer changes or the predetermined number of iterations is reached, so that the radar data and image data can extract classification data of different types of objects;

[0160] It should be noted that through image classification processing, different types of land objects in the image can be extracted, and through specific land object types, the image feature processing can be quickly realized, the image processing can be realized efficiently, and the land object types can be classified and clarified.

[0161] In this embodiment, preferably, the calculation of the feature extraction in S3 is as follows:

[0162] The two-dimensional features of the image data are calculated using SIFT feature extraction:

[0163] ,

[0164] in, is the Gaussian kernel, is an image, is the feature point;

[0165] The three-dimensional features of radar data are calculated using normal vectors:

[0166] First, calculate the center point of the neighborhood points:

[0167] ,

[0168] in, It is represented as the center point of the neighborhood point, that is, the average position of all neighborhood points. The position vectors of the points are summed and averaged; Represented as the first point in the neighborhood points, each point is a vector, usually represented as ,in is the dimension of the point; It is expressed as the number of neighborhood points, that is, the number of points considered;

[0169] Then, calculate the covariance matrix of the neighborhood points :

[0170] ,

[0171] in, It is expressed as a covariance matrix, which describes the distribution of neighborhood points in space and reflects the degree and direction of point dispersion; Represented as each neighborhood point With center point The difference between represents the deviation vector from the center point to each point; Representation is the transpose of the bias vector, which is multiplied by the bias vector itself to calculate the covariance of the point; It is expressed as the sum of the outer products of all deviation vectors, which is used to measure the distribution direction and degree of each point relative to the central point; It is expressed by dividing the sum of the outer products by , get the covariance matrix after mean;

[0172] The normal vector is the eigenvector corresponding to the smallest eigenvalue of the covariance matrix;

[0173] It should be noted that by extracting features of the types of objects after processing the image data and radar data, it is convenient to obtain feature information of the image data and radar data, and to facilitate subsequent fusion processing between the image data and radar data.

[0174] In this embodiment, preferably, the super strong factor in S4 is calculated as follows:

[0175] ,

[0176] in, is a constant used to adjust the intensity. is a two-dimensional feature point or a three-dimensional feature point, It is expressed as the calculated super strength factor;

[0177] All super factors are classified by calculating the distance using the Euclidean distance:

[0178] For any dimension Two points in and , the calculation formula of Euclidean distance is:

[0179] ,

[0180] in, Represented as a point and Point The Euclidean distance between two points represents the straight-line distance between them; Represents the sum index, indicating the sum from the first dimension to the dimension;

[0181] Represented as a point and Point In the Difference in dimension; Expressed as the square of the difference in each dimension; It is represented by summing the squares of all dimension differences and taking the square root to get the final Euclidean distance.

[0182] and When the value of is less than half the distance between the two nearest super factors, the point is included in the calculated super factor. If it exceeds half the distance between the two super factors, the point is not included in the calculated super factor.

[0183] When it is necessary to explain, the calculation of the super-strong factor is convenient for obtaining the most prominent feature points in the feature, and partitioning is performed according to the distance or pixel value between the feature points, so as to calculate whether the features match through the super-strong factor, and to serve as the reference point and fusion detail point when the image data and radar data are subsequently fused.

[0184] In this embodiment, preferably, the steps of matching the multiple super factors in S4 are as follows:

[0185] Calculate the distance: For each feature point in the image data , calculate its distance to each feature point in the radar data ;

[0186] ,

[0187] Select K nearest neighbors: Select and query feature points The K target feature points closest to the descriptor;

[0188] Matching candidate points: By comparing the matching conditions of the K nearest neighbors, the most suitable matching point is finally selected;

[0189] ,in, Expressed as the first super factor matching distance, Expressed as the second super factor matching distance, ratio If it is less than 0.8, the match is considered good, otherwise it is considered wrong;

[0190] When it is necessary to explain, the distance between the super factors obtained by distance calculation can be used to partition other feature points, so as to improve the accuracy of the destination of the feature points and the accuracy of the three-dimensional map after the subsequent fusion of the image data and the radar data.

[0191] In this embodiment, preferably, the feature level fusion in S5 is as follows:

[0192] Set the two-dimensional feature to , the three-dimensional features are :

[0193] ,

[0194] in, is the fused feature vector, and are the feature vectors from different modes or sources, and is the weighting coefficient, indicating the relative importance of each feature, satisfying the constraints ;

[0195] It should be noted that the fusion of three-dimensional features and two-dimensional features is achieved through feature level fusion, which makes it easier to obtain a more accurate three-dimensional map and facilitate subsequent geographic analysis and processing.

[0196] The specific operation process of the present invention is as follows:

[0197] Step 1: Obtain remote sensing data: Obtain remote sensing data through satellites, drones or aerial photography. Remote sensing data includes radar and image data types;

[0198] Step 2: Transmission of remote sensing data: Satellites, drones or aerial photography transmit remote sensing data to the edge cloud layer through the communication module, and the edge cloud layer pre-processes the remote sensing data to obtain processed data;

[0199] Step 3: Extract features from the data: The edge cloud layer performs image classification on the processed data again, extracts classification data of different types of objects from the radar data and the picture data; then performs feature extraction on the extracted classification data, and the features obtained from the radar data are three-dimensional features, and the features obtained from the picture data are two-dimensional features;

[0200] Step 4: Process the extracted features: Analyze and process the obtained three-dimensional features and two-dimensional features to obtain multiple super-strong factors in the three-dimensional features and two-dimensional features, and then classify the super-strong factors respectively, determine the distances of multiple super-strong factors by pixel values ​​and position values, perform regional classification processing, and then match multiple super-strong factors in the same regional classification to obtain the regional overlap value;

[0201] Step 5: Prepare the fusion of radar data and image data: According to the obtained regional overlap value, weights are assigned to radar data and image data. A larger weight value is assigned to a region with a higher overlap value, and a smaller weight value is assigned to a region with a lower overlap value. Then, radar data and image data are fused through feature level fusion to form a three-dimensional geographic space map.

[0202] Step 6: Visualization of the geospatial three-dimensional map: Display the geospatial three-dimensional map through a visualization module, and integrate raster data into the geospatial three-dimensional map. The raster data is used to display and store geographic data in the geospatial three-dimensional map.

[0203] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A geospatial analysis method based on remote sensing data, characterized in that: The steps include: S1. Acquisition of remote sensing data: remote sensing data is acquired through satellites, drones or aerial photography. Remote sensing data includes radar and image data types; S2. Transmission of remote sensing data: Satellites, drones or aerial photography transmit remote sensing data to the edge cloud layer through the communication module, and the edge cloud layer pre-processes the remote sensing data to obtain processed data; S3. Extract features from the data: The edge cloud layer performs image classification on the processed data again, extracts classification data of different types of objects from the radar data and the picture data; then performs feature extraction on the extracted classification data, and the features obtained from the radar data are three-dimensional features, and the features obtained from the picture data are two-dimensional features; S4, processing the extracted features: analyzing and processing the obtained three-dimensional features and two-dimensional features to obtain multiple super-strong factors in the three-dimensional features and the two-dimensional features, and then classifying the super-strong factors respectively, determining the distances of the multiple super-strong factors by pixel values ​​and position values, performing regional classification processing, and then matching the multiple super-strong factors in the same regional classification to obtain a regional overlap value; S5. Prepare the fusion of radar data and image data: according to the obtained regional overlap value, weight allocation is implemented for radar data and image data. A larger weight value is allocated to a region with a higher overlap value, and a smaller weight value is allocated to a region with a lower overlap value. Then, radar data and image data are fused through feature level fusion to form a three-dimensional geographic space map. S6. Visualization of geospatial three-dimensional map: Display the geospatial three-dimensional map through a visualization module, and integrate raster data into the geospatial three-dimensional map. The raster data is used to display and store geographic data of the geospatial three-dimensional map.

2. The method for geospatial analysis based on remote sensing data according to claim 1, characterized in that: The satellite in S2 uses high-frequency wireless signals to transmit remote sensing data, the drone or the aerial photography uses an Internet of Things network to transmit remote sensing data, and the edge cloud layer preprocesses the remote sensing data including noise removal, standardization, radiation correction, geometric correction, atmospheric correction, and cloud and fog removal.

3. The method for geospatial analysis based on remote sensing data according to claim 2, characterized in that: The noise removal adopts the Gaussian filtering algorithm, and the calculation formula is as follows: ， in, is the coordinate of the filtered image data information The pixel value at is the coordinate in the original image data information The pixel value of is the Gaussian kernel in The weight value at and The Gaussian kernel is and The radius in the direction, and , is the error coefficient generated by the Gaussian kernel during calculation; Gaussian Kernel The calculation formula is as follows: ， The calculation formula is as follows: ,in, and is the distance from the center of the Gaussian kernel, is the standard deviation of the Gaussian kernel; The standardization adopts linear standardization, which maps all pixel values ​​to the range of [0,1], and its formula is as follows: ， in, is a pixel value in the image, is the minimum value of all pixels in the image, is the maximum value of all pixels in the image.

4. The method for geospatial analysis based on remote sensing data according to claim 2, characterized in that: The radiation correction is used to eliminate the image radiation value errors caused by sensor characteristics, atmospheric influences and other external factors, and restore the true radiation information of ground targets. The geometric correction is used to convert the remote sensing image from the sensor coordinate system to the map coordinate system to eliminate the geometric distortion in the image. The atmospheric correction is used to remove the influence of aerosols, water vapor, etc. in the atmosphere on the remote sensing image and restore the ground reflectivity. The declouding and defogging processing is used to remove the influence of clouds and haze in the image.

5. The method for geospatial analysis based on remote sensing data according to claim 4, characterized in that: The radiation correction is converted into radiance or reflectivity by the sensor calibration function: The radiance is calculated as follows: ， in, is the radiance, is the digital value of the image, is the gain factor, is the bias, i.e. the dark current of the sensor; The calculation formula of reflectivity is: ， in, is the surface reflectivity, is the radiance, is the solar radiation intensity, is the sun’s zenith angle; The calculation formula of the geometric correction is as follows: ， in, are the pixel coordinates in the original image, are the target map coordinates, are the transformation parameters calculated from the ground control points; The atmospheric correction is calculated as follows: Atmospheric influence formula: ， in, is the ground reflectivity radiance after removing the atmospheric influence, is the calculated radiance, is the atmospheric transmittance, The calculation formula of reflectivity after removing the atmospheric influence is: ， in, is the reflectivity of the Earth’s surface; The calculation formula for declouding and defogging is as follows: Go to the cloud for processing: ， in, is the pixel value of the shortwave infrared band, is the set threshold, It is the cloud detection result, where 1 indicates cloud and 0 indicates non-cloud. To remove the cloud, the cloud area needs to be marked, and then the image information of other areas is used for interpolation and filling. Dehazing Processing: ， in, is the pixel value of the observed image, is the pixel value of the clear image after defogging, that is, the actual ground reflectivity, is the transmittance, which describes the attenuation of light in the atmosphere, It is a panoramic illumination. In order to remove fog, the transmittance is optimized To restore the image ,Common dehazing methods include dark channel prior; The minimum value of the image in the dark channel prior is used to estimate the transmittance: ， in, is a constant set to 0.95, is an image In the color channel The pixel value at It's pixels The small neighborhood around it is estimated by , restore the clear image after dehazing .

6. The method for geospatial analysis based on remote sensing data according to claim 1, characterized in that: The image classification in S3 adopts the K-means clustering algorithm, and the steps of the K-means clustering algorithm are as follows: Initialization: Random selection The initial center point of each cluster; Assignment phase: Assign each data point to the nearest cluster center, typically using the Euclidean distance metric: ， in, is a data point, It is The center of the cluster, is the dimension of the data, and They are data points and The cluster center is The weight of the dimension; Update phase: According to the allocation results, recalculate the center of each cluster: ， in, Belongs to the cluster The number of data points, Belongs to the cluster data points; Iteration: Repeat the assignment phase and the update phase until the cluster center no longer changes or the predetermined number of iterations is reached, so that the radar data and image data can extract classification data of different types of objects.

7. The method for geospatial analysis based on remote sensing data according to claim 1, characterized in that: The feature extraction in S3 is calculated as follows: The two-dimensional features of the image data are calculated using SIFT feature extraction: ， in, is the Gaussian kernel, is an image, is the feature point; The three-dimensional features of radar data are calculated using normal vectors: First, calculate the center point of the neighborhood points: ， in, It is represented as the center point of the neighborhood point, that is, the average position of all neighborhood points. The position vectors of the points are summed and averaged; Represented as the first point in the neighborhood points, each point is a vector, usually represented as ,in is the dimension of the point; It is expressed as the number of neighborhood points, that is, the number of points considered; Then, calculate the covariance matrix of the neighborhood points : ， in, It is expressed as a covariance matrix, which describes the distribution of neighborhood points in space and reflects the degree and direction of point dispersion; Represented as each neighborhood point With center point The difference between represents the deviation vector from the center point to each point; Representation is the transpose of the bias vector, which is multiplied by the bias vector itself to calculate the covariance of the point; It is expressed as the sum of the outer products of all deviation vectors, which is used to measure the distribution direction and degree of each point relative to the central point; It is expressed by dividing the sum of the outer products by , get the covariance matrix after mean; The normal vector is the eigenvector corresponding to the smallest eigenvalue of the covariance matrix.

8. The method for geospatial analysis based on remote sensing data according to claim 1, characterized in that: The super strength factor in S4 is calculated as follows: ， in, is a constant used to adjust the intensity. is a two-dimensional feature point or a three-dimensional feature point, It is expressed as the calculated super strength factor; All super factors are classified by calculating the distance using the Euclidean distance: For any dimension Two points in and , the calculation formula of Euclidean distance is: ， in, Represented as a point and Point The Euclidean distance between two points represents the straight-line distance between them; Represents the sum index, indicating the sum from the first dimension to the dimension; Represented as a point and Point In the Difference in dimension; Expressed as the square of the difference in each dimension; It is represented by summing the squares of all dimension differences and taking the square root to get the final Euclidean distance. and When the value of is less than half the distance between the two nearest super strong factors, the point is included in the calculated super strong factor. If it exceeds half the distance between the two super strong factors, the point is not included in the calculated super strong factor.

9. The method for geospatial analysis based on remote sensing data according to claim 1, characterized in that: The steps for matching the multiple super strong factors in S4 are as follows: Calculate the distance: For each feature point in the image data , calculate its distance to each feature point in the radar data ; ， Select K nearest neighbors: Select and query feature points The K target feature points closest to the descriptor; Matching candidate points: By comparing the matching conditions of the K nearest neighbors, the most suitable matching point is finally selected; ,in, Expressed as the first super factor matching distance, Expressed as the second super factor matching distance, ratio If it is less than 0.8, the match is considered good, otherwise it is considered wrong.

10. The method for geospatial analysis based on remote sensing data according to claim 1, characterized in that: The feature level fusion in S5 is as follows: Set the 2D feature to , the three-dimensional features are : ， in, is the fused feature vector, and are the feature vectors from different modes or sources, and is the weighting coefficient, indicating the relative importance of each feature, satisfying the constraints .

Images