Urban ecological environment remote sensing quantitative evaluation method based on space-air-ground cooperation

Abstract

The invention provides an urban ecological environment remote sensing quantitative evaluation method based on space-air-ground cooperation, and relates to the field of remote sensing technology and environmental science cross technology, and the method comprises the steps: carrying out the time-space registration and fusion of calibrated satellite remote sensing data, unmanned aerial vehicle remote sensing data and real-time environment monitoring data, extracting an urban vegetation coverage index, and carrying out the calculation of the urban vegetation coverage index; establishing a dynamic evaluation model to obtain an ecological quality level spatial distribution map of the target area; monitoring network configuration is dynamically optimized and a monitoring network optimization scheme is generated according to the identified ecological weak area in the ecological quality level spatial distribution diagram, and the monitoring network configuration comprises unmanned aerial vehicle monitoring frequency adjustment and ground sensor layout supplement. According to the invention, a full-chain technology system from space-air-ground cooperative monitoring, multi-source data fusion, intelligent dynamic evaluation and monitoring network closed-loop optimization is constructed, so that accurate diagnosis and targeted regulation and control of the urban ecological environment are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing technology and environmental science, in particular to a remote sensing quantitative evaluation method for urban ecological environment based on space-air-ground cooperation. BACKGROUND

[0002] The rapid development of urbanization has made the ecological environment problems of high-density urban areas increasingly serious, such as fragmentation of vegetation coverage. Precise and dynamic ecological evaluation is the prerequisite for effective regulation. Traditional monitoring methods relying on a single data source (such as ground surveys or satellite remote sensing) may have limitations in terms of scope, efficiency and accuracy. Space-air-ground cooperative monitoring technology integrates satellite (space), unmanned aerial vehicle (air) and ground sensor (ground) data, providing the possibility for multi-scale, multi-temporal and three-dimensional observation.

[0003] Taking the monthly ecological evaluation of a high-density core business district (building area exceeding 20 million square meters, building density 65%) and the surrounding area as an example, the existing technology usually uses satellite images to preliminarily identify vegetation coverage areas, then uses unmanned aerial vehicles for detailed exploration, and supplements with ground sensor verification. However, there are the following defects in data processing: first, there is a huge difference in scale between satellite data (such as 30-meter resolution) and unmanned aerial vehicle data (0.1-meter resolution). The existing method lacks effective scale conversion and precise registration algorithm, which may cause boundary errors when fusing at the pixel level, and cannot accurately establish the correction relationship between high and low resolution data, resulting in distortion of micro ecological information extraction; second, ground sensor data is mainly used for post-validation, and cannot be used for real-time and systematic reverse correction of remote sensing inversion models. For example, the temperature inversion model with fixed parameters cannot be dynamically optimized according to sensor data, resulting in fluctuation of precision. At the same time, the evaluation model is usually trained based on specific scene data, which may not be suitable for different seasons (such as leaf fall leading to sudden change of vegetation index) or complex situations across regions, and has poor generalization ability. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a remote sensing quantitative evaluation method for urban ecological environment based on space-air-ground cooperation. By constructing a full-chain technical system from space-air-ground cooperative monitoring, multi-source data fusion, intelligent dynamic evaluation to monitoring network closed-loop optimization, precise diagnosis and targeted regulation of urban ecological environment are realized.

[0005] To solve the above technical problems, the technical scheme of the present application is as follows: In a first aspect, a remote sensing quantitative evaluation method for urban ecological environment based on space-air-ground cooperation, the method comprising: obtaining ecological base data of a target area through a satellite remote sensing platform; performing inversion processing on the ecological base data to obtain satellite remote sensing inversion results of regional vegetation coverage and land surface temperature; identifying the satellite remote sensing inversion results, and determining ecological key areas based on a preset threshold; ecological data of the ecological key area through a UAV remote sensing platform, and processing the ecological data to obtain UAV remote sensing inversion results of vegetation species distribution, vegetation health degree, and building roof greening coverage rate; acquire real-time environmental monitoring data through a ground sensor network deployed in the target area; use the real-time environmental monitoring data to respectively calibrate the satellite remote sensing inversion results and the UAV remote sensing inversion results to obtain calibrated satellite remote sensing data and UAV remote sensing data; spatially register and fuse the calibrated satellite remote sensing data, the UAV remote sensing data, and the real-time environmental monitoring data, extract urban vegetation coverage indexes, establish a dynamic evaluation model, and obtain an ecological quality grade spatial distribution map of the target area; According to the ecological weak area identified in the ecological quality grade spatial distribution map, dynamically optimize the monitoring network configuration and generate a monitoring network optimization scheme, wherein the monitoring network configuration includes adjusting the UAV monitoring frequency and supplementing the ground sensor deployment.

[0006] In a second aspect, a city ecological environment remote sensing quantitative evaluation system based on space-air-ground cooperation includes: The cooperative monitoring module is configured to acquire ecological base data of a target area through a satellite remote sensing platform; perform inversion processing on the ecological base data to obtain satellite remote sensing inversion results of regional vegetation coverage and land surface temperature; identify the satellite remote sensing inversion results and determine ecological key areas based on a preset threshold; collect ecological data of the ecological key areas through a UAV remote sensing platform, and process the ecological data to obtain UAV remote sensing inversion results of vegetation species distribution, vegetation health degree, and building roof greening coverage rate; The calibration module is configured to acquire real-time environmental monitoring data through a ground sensor network deployed in the target area; use the real-time environmental monitoring data to respectively calibrate the satellite remote sensing inversion results and the UAV remote sensing inversion results to obtain calibrated satellite remote sensing data and UAV remote sensing data; The evaluation module is configured to spatially register and fuse the calibrated satellite remote sensing data, the UAV remote sensing data, and the real-time environmental monitoring data, extract urban vegetation coverage indexes, establish a dynamic evaluation model, and obtain an ecological quality grade spatial distribution map of the target area; The dynamic optimization module is configured to dynamically optimize the monitoring network configuration and generate a monitoring network optimization scheme according to the ecological weak area identified in the ecological quality grade spatial distribution map, wherein the monitoring network configuration includes adjusting the UAV monitoring frequency and supplementing the ground sensor deployment.

[0007] In a third aspect, a computing device includes: one or more processors; a memory device storing one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method.

[0008] In a fourth aspect, a computer-readable storage medium stores a program, which when executed by a processor, implements the method.

[0009] The above scheme of the present application at least includes the following beneficial effects: Rely on satellite remote sensing wide range of coverage to obtain the target area ecological base data; Through the inversion processing extracts the core parameters of regional vegetation coverage and land surface temperature; Based on the preset threshold value, the ecological key area is identified and determined, and the clear range guidance is provided for the subsequent focused key area to carry out fine monitoring; The unmanned aerial vehicle remote sensing data is collected in the ecological key area, the focus area is focused, and the data collection pertinence is improved; After processing the collected data, the micro ecological information such as vegetation species distribution, vegetation health degree and building roof greening coverage rate is obtained, and the information loss of satellite data in micro scale is supplemented; Real-time environmental monitoring data is obtained through the ground sensor network, and the measured basis is provided for the calibration of remote sensing data; The satellite and unmanned aerial vehicle remote sensing inversion results are calibrated respectively by using the data, the system deviation of the two types of remote sensing data is eliminated, and the fitting degree of data and actual environment is improved; The multi-source data is spatio-temporally registered, and the data space scale and time dimension are unified; The characteristics of multi-source data are integrated through fusion processing; The urban vegetation coverage index is extracted, the dynamic evaluation model is established, and the ecological quality grade spatial distribution map is generated, so that the system integration and visual presentation of multi-dimensional ecological information are realized; According to the ecological quality grade spatial distribution map, the ecological weak area is accurately identified; The unmanned aerial vehicle monitoring frequency is adjusted, the ground sensor layout is supplemented, and the monitoring network configuration is optimized; The monitoring resources are concentrated in the key weak area, and the pertinence and efficiency of the monitoring network are improved. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 is a flowchart of a city ecological environment remote sensing quantitative evaluation method based on space-air-ground cooperation provided by an embodiment of the present application.

[0011] Figure 2 is a city ecological environment remote sensing quantitative evaluation system based on space-air-ground cooperation provided by an embodiment of the present application. DETAILED DESCRIPTION

[0012] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it is understood that the present disclosure can be embodied in many forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the present disclosure to those skilled in the art.

[0013] As Figure 1 shown, the embodiment of the present application proposes a space-air-ground collaborative urban ecological environment remote sensing quantitative evaluation method, which comprises the following steps: Step 100, obtaining the ecological base data of the target area through a satellite remote sensing platform; performing inversion processing on the ecological base data to obtain satellite remote sensing inversion results of regional vegetation coverage and surface temperature; identifying the satellite remote sensing inversion results and determining the ecological key area based on a preset threshold; Step 200, collecting ecological data of the ecological key area through a UAV remote sensing platform, and processing the ecological data to obtain UAV remote sensing inversion results of vegetation species distribution, vegetation health degree and building roof greening coverage rate; Step 300, obtaining real-time environmental monitoring data through a ground sensor network arranged in the target area; using the real-time environmental monitoring data, respectively calibrating the satellite remote sensing inversion results and the UAV remote sensing inversion results to obtain calibrated satellite remote sensing data and UAV remote sensing data; Step 400, spatiotemporal registration and fusion of the calibrated satellite remote sensing data, UAV remote sensing data and real-time environmental monitoring data, extraction of urban vegetation coverage index, establishment of a dynamic evaluation model, and obtaining of an ecological quality grade spatial distribution map of the target area; Step 500, dynamically optimizing the monitoring network configuration and generating a monitoring network optimization scheme according to the ecological weak area identified in the ecological quality grade spatial distribution map, wherein the monitoring network configuration includes adjusting the UAV monitoring frequency and supplementing the ground sensor arrangement.

[0014] In the embodiment of the present application, the satellite remote sensing platform is used to obtain large-scale ecological base data of the target area, which can realize the overall coverage of the overall ecological condition of the target area. The ecological base data is processed by inversion, which can convert the original satellite data into vegetation coverage and surface temperature information that can directly reflect the regional vegetation growth condition and the surface thermal environment. The satellite remote sensing inversion result is identified and the ecological key area is determined based on the preset threshold, which can quickly focus on the area with representative ecological condition or needing attention from the large-scale data, and provide a clear direction for subsequent targeted monitoring. For the determined ecological key area, the ecological data is collected by the unmanned aerial vehicle remote sensing platform, which can realize the acquisition of close-range and high-resolution data of the key area. The collected ecological data is processed, which can further mine the specific distribution of vegetation species, the growth and health status of vegetation, and the coverage ratio of building roof greening, which can supplement the deficiency of satellite remote sensing in fine information acquisition, and make the ecological details of the ecological key area more clearly presented. The real-time environmental monitoring data is obtained by the ground sensor network arranged in the target area, which can provide real-time reference close to the actual environment for the satellite remote sensing inversion result and the unmanned aerial vehicle remote sensing inversion result. The real-time environmental monitoring data is used to calibrate the two kinds of remote sensing inversion results respectively, which can correct the local deviation of satellite remote sensing due to large-scale coverage and the measurement error of unmanned aerial vehicle remote sensing due to environmental factors, so that the calibrated satellite remote sensing data and unmanned aerial vehicle remote sensing data are more consistent with the actual ecological environment of the target area.

[0015] The calibrated satellite remote sensing data, unmanned aerial vehicle remote sensing data and real-time environmental monitoring data are time and space registered and fused, which can break the difference of data from different sources in time and space dimensions, and realize the collaborative correlation of multi-source data. The urban vegetation coverage index is extracted, which can efficiently mine the index reflecting the overall coverage characteristics of urban vegetation from the fused multi-source data. The dynamic evaluation model is established based on the extracted index, which can convert the multi-source data information into system evaluation logic, and finally generate the ecological quality grade spatial distribution map of the target area, which can intuitively present the ecological quality difference of different positions in the target area. The ecological weak area is identified according to the ecological quality grade spatial distribution map, which can accurately locate the area with poor ecological condition and needing to be improved from the intuitive spatial distribution result. Based on the identification result, the monitoring network configuration is dynamically optimized, the unmanned aerial vehicle monitoring frequency is adjusted, more monitoring resources can be invested in the ecological weak area, the ground sensor arrangement can enhance the real-time collection ability of environmental data in the weak area, and the subsequent monitoring is more targeted and effective. The monitoring network optimization scheme can convert the cognition of the ecological weak area into specific improvement direction, and provide practical operation guide for the ecological management of the target area.

[0016] In a preferred embodiment of the present application, the above step 100 comprises: Step 101, obtaining multi-spectral remote sensing images of the target area by a satellite remote sensing platform as the ecological base data, specifically including: selecting a satellite remote sensing platform covering the high-density core business district and the surrounding area, which needs to have multi-spectral data acquisition capability, such as Sentinel-2 series satellites or Landsat series satellites, obtaining multi-spectral remote sensing images of the target area at monthly frequency, ensuring that the images are not covered by large area cloud and snow to ensure data integrity, and the images obtained need to contain visible light band, near-infrared band and thermal infrared band. These band data will be used as large-scale ecological base data for subsequent ecological parameter inversion.

[0017] Step 102, pre-processing the large-scale ecological base data, including radiation calibration, atmospheric correction and geometric registration, to obtain surface reflectance data, specifically including: first, carrying out radiation calibration processing on the obtained large-scale ecological base data, converting the digital quantization value in the image to absolute radiance value according to the calibration coefficient provided by the satellite sensor, eliminating the error caused by the response difference of the sensor itself; then, atmospheric correction is carried out, using 6S atmospheric correction model, combining with the meteorological data at the imaging time of the target area, removing the interference of atmospheric scattering, absorption and other factors on the ground reflectance signal, and restoring the true reflection characteristics of the ground; then, geometric registration is implemented, taking the high-precision digital elevation model of the target area and the ground control points measured in the field as the reference, aligning the corrected image with the national 2000 geodetic coordinate system, ensuring the accuracy of the spatial position of the image. Through the above series of pre-processing operations, the surface reflectance data which can be used for subsequent inversion calculation is finally obtained.

[0018] Step 103, based on the surface reflectance data, calculating the normalized vegetation index, converting the normalized vegetation index to vegetation coverage, and obtaining the vegetation coverage inversion result, specifically including: based on the surface reflectance data obtained by pre-processing, extracting the near-infrared band reflectance and red light band reflectance, and calculating the normalized vegetation index of each pixel by band operation; combining the surface coverage characteristics of the high-density core business district, using the pixel bisection model as the vegetation coverage conversion model, which assumes that each pixel is composed of vegetation part and non-vegetation part. According to the calculated normalized vegetation index value, the proportion of vegetation and non-vegetation in the pixel is distinguished, and then the vegetation coverage value of each pixel is calculated. Integrating the vegetation coverage data of all pixels, the vegetation coverage inversion result of the whole target area is formed.

[0019] Step 104, based on the surface reflectivity data, calculating the radiation brightness value of the thermal infrared band; based on the radiation brightness value, extracting the surface emissivity; combining the atmospheric profile parameters, the radiation brightness value and the surface emissivity are inverted to obtain the surface temperature inversion result, specifically including: extracting the thermal infrared band information from the surface reflectivity data, combining the radiation brightness conversion relationship obtained in step 102, calculating the radiation brightness value of the thermal infrared band; according to the land use type map of the target area, the surface is divided into building land, vegetation cover area, road, water body and other categories, the sample data of the emissivity of different surface types is obtained by field measurement, the corresponding relationship between the surface type and the emissivity is established, and then the surface emissivity of each pixel in the image is extracted; through the meteorological department, the atmospheric profile parameters of the target area at the satellite imaging moment are obtained, including atmospheric temperature, atmospheric humidity, air pressure and other data; using the split window algorithm, that is, a surface temperature inversion model, the thermal infrared band radiation brightness value, the surface emissivity and the atmospheric profile parameters are substituted into the split window algorithm model, and the surface temperature value of each pixel is gradually calculated, the surface temperature data of all pixels are integrated, and the surface temperature inversion result of the whole target area is formed.

[0020] Step 105, spatially superimposing the vegetation coverage inversion result and the surface temperature inversion result to obtain the satellite remote sensing inversion result of the regional vegetation coverage and the surface temperature, specifically including: first, the vegetation coverage inversion result and the surface temperature inversion result are processed to unify the spatial resolution, and the two are adjusted to the same pixel size and spatial resolution to ensure that the data is consistent in spatial scale; second, taking the geographic coordinate system of the target area as the reference, the coordinate calibration is performed on the two inversion results to eliminate the spatial position deviation; then, using the spatial superposition function of the remote sensing image processing software, the vegetation coverage value of each pixel in the vegetation coverage inversion result is associated with the surface temperature value of the corresponding position in the surface temperature inversion result, so that each spatial pixel contains both vegetation coverage and surface temperature ecological parameter information; finally, the data integrity of the superimposed result is checked, and the invalid pixels caused by data loss or abnormality are removed to form the complete satellite remote sensing inversion result of the regional vegetation coverage and the surface temperature.

[0021] Step 106, the satellite remote sensing inversion results are statistically divided into zones, and the vegetation coverage values and land surface temperature values are extracted respectively, specifically including: according to the functional zoning of the high-density core business district, the target area is divided into statistical zones such as commercial building concentrated area, surrounding residential area, green park area, road traffic area, etc., and the boundary of each statistical zone is based on administrative planning boundary or actual ground features; for each statistical zone, the vegetation coverage values and land surface temperature values of all pixels in the zone are extracted, the average, maximum and minimum values of the vegetation coverage in each zone are calculated, and the area proportions of the three levels of vegetation coverage higher than 60%, 30% to 60% and lower than 30% are calculated; at the same time, the average, maximum and minimum values of the land surface temperature in each zone are calculated, and the area proportions of the three levels of land surface temperature higher than 35℃, 30 to 35℃ and lower than 30℃ are calculated; the statistical data of each zone are arranged into a table, and a spatial distribution map of the statistical results of the zones is generated to clearly show the differences in ecological parameters of different zones.

[0022] Step 107, the vegetation coverage values are compared with the first preset threshold value, and the areas lower than the first preset threshold value are identified and marked as sparse vegetation areas; the land surface temperature values are compared with the second preset threshold value, and the areas higher than the second preset threshold value are identified and marked as high-temperature land surface areas, specifically including: combining the ecological evaluation standard and historical monitoring data of high-density urban areas, the first preset threshold value is determined through expert demonstration and statistical analysis, which is set to 25%, i.e. the area with vegetation coverage lower than 25% is determined as sparse vegetation area, and this threshold value can accurately reflect the insufficient vegetation coverage in high-density urban areas; the vegetation coverage values of each pixel are compared with the first preset threshold value one by one, and the pixels with vegetation coverage values lower than 25% are screened out, and the spatial range of these pixels is marked as sparse vegetation area, and its boundary coordinates and area are recorded; referring to the average land surface temperature in summer of the target area and the comfort requirement of human living environment, the second preset threshold value is verified and determined through field monitoring data, which is set to 33℃, i.e. the area with land surface temperature higher than 33℃ is determined as high-temperature land surface area; the land surface temperature values of each pixel are compared with the second preset threshold value one by one, and the pixels with land surface temperature values higher than 33℃ are screened out, and the spatial range of these pixels is marked as high-temperature land surface area, and its boundary coordinates and area are recorded.

[0023] In step 108, the vegetation sparse area and the high-temperature area of the ground are spatially combined to obtain an ecological key area, specifically including: firstly, the spatial data of the marked vegetation sparse area and the high-temperature area of the ground are uniformly formatted, and the two are converted into the same vector data format, and the same geographic coordinate system is ensured, so as to avoid merging errors caused by data format or coordinate differences; then, the spatial merging function of the geographic information system software is used to integrate the spatial range of the vegetation sparse area and the spatial range of the high-temperature area of the ground, for the overlapping part of the two areas, the spatial information of the overlapping area is retained, and repeated calculation is not performed, for the part only belonging to the vegetation sparse area or only belonging to the high-temperature area of the ground, the spatial range is completely retained; then, the boundary of the combined area is smoothed to eliminate the jagged problem of the boundary caused by the dispersion of the pixels, so that the area boundary is more regular; then, the total area, the circumference and the area ratio of the ecological key area in the whole target area are calculated, and the distribution and area ratio of different land use types such as building roof, road and idle land in the ecological key area are also counted.

[0024] In the embodiment of the application, the multispectral remote sensing image of the target area is obtained by a satellite remote sensing platform as large-range ecological base data, and the spectral characteristics of different ground objects such as vegetation and ground in the target area can be comprehensively captured by means of the multiple band information contained in the multispectral image; the large-range ecological base data is preprocessed by radiation calibration, atmospheric correction and geometric registration, so as to convert the original remote sensing data into ground reflectivity data which can truly reflect the actual condition of the ground; the normalized vegetation index is calculated based on the ground reflectivity data, so as to effectively distinguish vegetation by making full use of the spectral response difference of vegetation in a specific band; the normalized vegetation index is converted into vegetation coverage, so as to convert the abstract spectral index into intuitive and quantifiable vegetation coverage information, and clearly present the distribution and coverage degree of the vegetation in the target area; the radiation brightness value of the thermal infrared band is calculated based on the ground reflectivity data, so as to accurately obtain the original observation information of the ground thermal radiation; the ground emissivity is extracted, and the ground temperature is retrieved in combination with the atmospheric profile parameter, so as to eliminate the influence of the atmosphere on the transmission of thermal radiation.

[0025] The vegetation coverage degree and the surface temperature are superimposed in space, the information of the two key ecological parameters is integrated in the spatial dimension, the spatial correlation of the two is intuitively presented, the comprehensive satellite remote sensing inversion result containing the information of the vegetation coverage and the surface temperature is formed, the satellite remote sensing inversion result is statistically divided into regions, and the vegetation coverage degree and the surface temperature are extracted, the spatialized inversion result is converted into specific and regional quantitative data, the areas with low vegetation coverage and the areas with high surface temperature are screened out, the areas with high surface temperature are accurately located, and the identification process of the two types of ecological condition key attention areas is more objective and targeted, the sparse vegetation area and the high surface temperature area are spatially combined to obtain the ecological key area, the ranges of the two types of key attention areas are integrated, and the key area covering the ecological problems of the vegetation and the thermal environment is formed.

[0026] In a preferred embodiment of the present application, the step 200 comprises: The step 201 comprises: based on the ecological key area, planning a flight path of a UAV, collecting data by using a UAV remote sensing platform carrying a hyperspectral camera, and acquiring hyperspectral image data of the ecological key area, specifically comprising: extracting the geographical coordinate range of the ecological key area, such as 116.32° to 116.48° east longitude and 39.92° to 40.08° north latitude, determining that the area is about 12km 2 ; planning a flight path of the UAV based on the range, adopting a grid flight mode, setting the flight height to 120 meters, ensuring that the ground resolution of the hyperspectral camera reaches 0.1 meters, controlling the flight speed to be 8m / s, setting the adjacent flight line overlap rate to be 80% and the heading overlap rate to be 60%, avoiding data missing caused by insufficient overlap rate; selecting a multi-rotor UAV carrying a hyperspectral camera with a 600-1000nm wave band range and 256 continuous wave bands, checking the camera parameters before flight, setting the spectral sampling interval to be 1.5nm, and adjusting the exposure time to be 1 / 1000s; selecting weather without clouds and wind less than or equal to 3 levels, and collecting flight data from 9:30 to 11:00 in the local time in the morning, recording the longitude, latitude, altitude and flight attitude data of the UAV every 5 minutes during the collection process, synchronously acquiring the hyperspectral image data of the ecological key area, and ensuring that the image completely covers the sparse vegetation area, the high surface temperature area and the composite ecological weak area without data missing area.

[0027] Step 202, pre-processing the hyperspectral image data, including radiation correction, spectral correction and geometric correction, to obtain hyperspectral reflectivity data, specifically including: before flight, placing the hyperspectral camera and a standard white plate (reflectivity known and uniform, reflectivity value 99%) in the open area of the ecological key area, and shooting the standard white plate image under the same light condition; after flight, reading the original gray value of each pixel in the hyperspectral image, and reading the gray value of the corresponding waveband of the standard white plate image, taking the known reflectivity of the standard white plate as the basis to calculate the radiation correction coefficient of each waveband, and the coefficient calculation is based on the corrected reflectivity = (original gray value / white plate gray value) x white plate reflectivity; applying the coefficient to all hyperspectral image pixels to eliminate the influence of light intensity changes and camera sensor response differences on image gray values, and obtaining the preliminary radiation corrected image data.

[0028] Select high-purity mercury lamp as spectral standard light source, let the mercury lamp light pass through the hyperspectral camera optical system in a dark room environment, record the response position of each waveband of the camera to the characteristic spectral line of the mercury lamp; compare the wavelength of the known characteristic spectral line of the mercury lamp with the recorded response wavelength, calculate the wavelength offset of each waveband, and correct the wavelength of the waveband whose offset exceeds 0.5 nm, so that the actual detection wavelength of each waveband of the camera is consistent with the theoretical wavelength; at the same time, by shooting the standard spectral plate containing the standard reflectivity curve of 400 to 1000 nm waveband, verify the deviation of the reflectivity data of each waveband after correction and the reflectivity data of the standard spectral plate, and ensure that the deviation is less than or equal to 2%, and complete the spectral correction.

[0029] Uniformly select 20 ground control points in the ecological key area, and preferentially select fixed and easily identifiable points such as road intersections, high-rise building roof corners, and artificial marker tops as control points; use a high-precision GNSS receiver (positioning accuracy less than or equal to 0.5 meters) to measure the geographic coordinates (east longitude, north latitude) of each control point; load the hyperspectral image into the geographic information system software, find the pixel position corresponding to each control point in the image, and record its row and column number; using a polynomial fitting algorithm, taking the pixel row and column number and geographic coordinates of the control points as samples, calculate the coordinate conversion parameters; based on the algorithm, resample the hyperspectral image, and the resampling method uses bilinear interpolation method, so that the pixel position of the corrected image accurately corresponds to the actual geographic coordinates, and the coordinate residual of each control point after correction is verified to be less than or equal to 0.1 meters, to obtain the hyperspectral reflectivity data.

[0030] Step 203, extracting spectral features of different vegetation species based on the hyperspectral reflectance data; using spectral angle mapping method to classify the spectral features to obtain vegetation species distribution information, specifically including: determining common vegetation species in the high-density core business district through field investigation, including trees (camphor tree, plane tree), shrubs (red leaf photinia, gold sun female privet), herbaceous plants (admirable grass, dog tooth grass), etc. a total of 8 categories; selecting pure sample areas of each species in the field, with an area of 10m or more, without other species mixed, using a hyperspectral camera to shoot hyperspectral images of the sample area under the same lighting conditions, extracting standard spectral curves of each species, and taking the spectral average of 100 valid pixels in the sample area for each curve; extracting the spectral curve from the hyperspectral reflectance data, calculating the angle between the spectral curve of each pixel to be classified and the standard spectral curve of 8 species using spectral angle mapping method; setting the angle threshold to 0.1 radian, and classifying the pixels with an angle between the spectral curve of the pixel to be classified and the standard spectral curve less than 0.1 radian into the corresponding species category; marking the pixels with an angle greater than 0.1 radian as unidentified vegetation; after classification, randomly selecting 50 verification points in the field, comparing the field species type of the verification points with the classification results to ensure that the classification accuracy is greater than or equal to 90%, and obtaining the vegetation species distribution information.

[0031] Step 204, calculating the normalized vegetation index based on the hyperspectral reflectance data, dividing the normalized vegetation index according to the preset health degree threshold, and obtaining the vegetation health evaluation result, specifically including: extracting the reflectance values of the near-infrared band (850nm) and the red band (670nm) from the hyperspectral reflectance data; calculating the normalized vegetation index of each pixel based on the reflectance data of the two bands; combining the vegetation growth characteristics of the high-density core business district, setting the health degree threshold by analyzing the normalized vegetation index statistical data of healthy vegetation, sub-healthy vegetation and declining vegetation in the region in the past three years: when the normalized vegetation index is greater than or equal to 0.7, it is determined as healthy vegetation; when the normalized vegetation index is between 0.4 and 0.7, it is determined as sub-healthy vegetation; when the normalized vegetation index is less than 0.4, it is determined as declining vegetation; comparing the normalized vegetation index value of each pixel calculated with the above threshold to determine the health degree level of each pixel; taking 50m x 50m as a statistical unit, calculating the area proportion of different health degree levels of vegetation in each unit to obtain the vegetation health evaluation result.

[0032] Step 205, based on the hyperspectral reflectance data, identify the building roof contour, and extract the proportion of vegetation pixels within the contour, calculate the building roof green coverage rate, specifically including: loading the hyperspectral reflectance data to the object-oriented image analysis software, using multi-scale segmentation algorithm to segment the image, the segmentation parameters are set as: spectral weight 0.6, shape weight 0.4, compactness 0.5, the initial segmentation scale is set to 50, corresponding to the actual area of about 2500m 2 , according to the actual size of the building roof, the segmented objects are merged or split to ensure that each segmented object can fully reflect the boundary of the building roof or non-roof area; based on the spectral and shape features of the high-density core business district building roof, set the roof identification threshold: the average value of the hyperspectral reflectance of the roof area in the visible light band (450-650nm) is 0.3-0.5, the shape index (perimeter 2 / area) is greater than or equal to 10, and there is no obvious vegetation spectral feature, and the near-infrared reflectance is less than 0.3; according to the above threshold, the objects that meet the conditions are selected from the segmented objects and marked as building roof contour; for each building roof contour, the number of vegetation pixels within the contour and the total number of pixels within the contour are counted, the number of vegetation pixels within the contour is the vegetation pixel determination standard, which is the near-infrared reflectance greater than or equal to 0.5 and the red light reflectance less than or equal to 0.3, and the total number of pixels within the contour; calculate the ratio of the number of vegetation pixels to the total number of pixels, which is the green coverage rate of the corresponding building roof; for the building roof with an area less than 100m 2 , because the number of pixels is too small, it may cause calculation error, so its green coverage rate is marked as to be supplemented, finally the green coverage rate data of all qualified building roofs is obtained, the data is recorded in the form of table, including roof number, roof area, vegetation pixel number, total pixel number and green coverage rate.

[0033] Step 206, integrate the vegetation species distribution information, vegetation health assessment results and building roof greening coverage rate to obtain the UAV remote sensing inversion result, specifically including: converting the vegetation species distribution information, vegetation health assessment results (including health grade and corresponding spatial range data of each region), building roof greening coverage rate into WGS84 coordinate system to ensure consistent spatial reference; in the geographic information system software, the vegetation species distribution information and the vegetation health assessment results (including spatial range and health grade association data) are spatially overlaid to obtain vegetation spatial distribution data containing vegetation species type-health grade double attribute association, so that each vegetation distribution area is associated with the corresponding health grade information; then the overlaid data and building roof contour vector data (extracted from step 205 roof identification result) are overlaid, so that each building roof is associated with the corresponding greening coverage rate data; in the integration process, the attribute information of the overlapping area is associated and stored, for example, a certain area has both tree distribution and sub-health grade, and contains part of the building roof, then the attribute information of the area records the tree species, sub-health grade and the greening coverage rate of the corresponding roof; finally, the UAV remote sensing inversion result is generated, which contains comprehensive vector data (containing spatial association information of species distribution range, health grade, roof contour and corresponding greening coverage rate) and attribute data table (recording detailed data of species type, health proportion, roof greening coverage rate and the like of each spatial unit).

[0034] In the embodiment of the present application, the flight path of the unmanned aerial vehicle is planned based on the ecological key area, which can accurately focus on the monitoring area and improve the data collection efficiency; the unmanned aerial vehicle remote sensing platform carrying a hyperspectral camera can collect data, and the hyperspectral image data containing hundreds of continuous wave bands can be obtained; the hyperspectral image data is subjected to radiation correction, which can eliminate the influence of factors such as light intensity variation and sensor response difference on the image gray value; spectral correction can correct the spectral drift and band width deviation between different wave bands, and ensure the accuracy of the spectral information of each wave band; geometric precision correction can eliminate the image geometric deformation caused by the change of the flight attitude of the unmanned aerial vehicle and the terrain undulation; the hyperspectral reflectivity data obtained through the three types of correction processing can truly reflect the spectral characteristics of vegetation, buildings and other ground objects; the spectral characteristics of different vegetation species are extracted based on the hyperspectral reflectivity data, which can fully utilize the advantages of continuous wave bands and high resolution of hyperspectral data, and capture the unique spectral response of different species in a specific wave band (such as the difference in spectral absorption peak caused by the difference in leaf pigment content); through the spectral angle mapping method, the spectral curve of the vegetation to be identified can be matched with the standard spectral curve of the known species, and the influence of spectral intensity variation on the classification result can be reduced; the normalized difference vegetation index is calculated based on the hyperspectral reflectivity data, which can effectively integrate the spectral differences of vegetation in the near-infrared and red wave bands; the normalized difference vegetation index is divided according to the preset health degree threshold, which can establish a unified judgment standard for vegetation health degree evaluation; the object-oriented image analysis method is used to identify the building roof profile, which can take the object composed of adjacent pixels as the analysis unit, and identify the building roof profile by combining the shape, texture, spectrum and other comprehensive characteristics of the roof, so as to accurately define the spatial range of the building roof; the building roof green coverage rate is calculated by extracting the proportion of vegetation pixels in the profile, which can quantify the coverage degree of vegetation in the roof area; the vegetation species distribution information, vegetation health degree evaluation result and building roof green coverage rate are integrated, which can associate and summarize the ecological information of the three dimensions of the species composition, growth state and roof greening of the vegetation in the key area, and avoid the limitation of single information; the unmanned aerial vehicle remote sensing inversion result formed after integration can comprehensively present the refined ecological condition of the ecological key area.

[0035] In a preferred embodiment of the present application, the above step 300 comprises: Step 301, based on the ecological key area and the unmanned aerial vehicle remote sensing inversion result, a ground sensor network is arranged in the target area, the ground sensor network comprises air temperature and humidity sensors, particulate matter concentration sensors located in the green land and chlorophyll content sensors located in the vegetation canopy, specifically comprising: first, extract the ecological key area data, which includes sparse vegetation area, high temperature area and complex ecological weak area, with a total area of about 12km 2; simultaneously call the unmanned aerial vehicle remote sensing inversion result, determine the target area green space distribution range, vegetation species type and vegetation health degree level (healthy, sub-health, recession); based on the above data, plan the ground sensor network layout scheme: in the green space, according to every 2km 2 Lay out 1 monitoring point density, a total of 6 air temperature and humidity sensors and 6 particulate matter concentration sensors, the sensor selects air temperature and humidity sensor with precision of ±0.5℃ (temperature), ±5%RH (humidity), and particulate matter concentration sensor with precision of ±5μg / m 3 (PM2.5), when laying out, prefer to choose the green space where the sub-health and recession level vegetation is located in the unmanned aerial vehicle inversion result, to ensure that the vegetation growth environment of different health status is covered; at the vegetation canopy, for the trees such as camphor tree and plane tree and shrubs such as red leaf photinia identified by the unmanned aerial vehicle, select 3 plants of different health degree level for each species, lay out chlorophyll content sensors in the upper part of the plant canopy (height 2 to 3m), a total of 18, the sensor selects portable chlorophyll sensor with measurement range of 0 to 100SPAD and precision of ±1SPAD, and the fixing method is to fix it on the branch with adjustable buckle to avoid damaging the vegetation; after the layout is completed, use high-precision GNSS receiver (positioning accuracy less than or equal to 0.5m) to record the geographic coordinates (east longitude, north latitude) and layout height of each sensor, and establish the spatial correlation between the sensor position and the ecological key area and the green space distribution.

[0036] Step 302, collect real-time environmental monitoring data by the ground sensor network at a preset time frequency, obtain air temperature and humidity time series data, particulate matter concentration time series data and vegetation chlorophyll content time series data, specifically including: first, set the preset time frequency of sensor data collection, air temperature and humidity sensor and particulate matter concentration sensor adopt high frequency collection mode, collect data once every 10 minutes, the collection period covers 24 hours of the target area, to ensure that the diurnal environmental parameter change is captured; vegetation chlorophyll content sensor adopts medium-high frequency collection mode, collects data once every 2 hours, the collection period is limited to local time from 9:00 am to 17:00 pm, during which the photosynthesis of vegetation is active, and the chlorophyll content data can better reflect the actual health status of vegetation. Before collection, preheat and calibrate all sensors: air temperature and humidity sensor is calibrated by standard temperature and humidity calibration box (temperature range 0 to 50℃, humidity range 20% to 90%), to ensure that the measured value deviates from the standard value by less than or equal to 0.3℃ (temperature), less than or equal to 3%RH (humidity); the particulate matter concentration sensor is calibrated by the standard aerosol generator to generate aerosol with known concentration (50μg / m 3 , 100μg / m 3 , 200μg / m 3 ), to ensure that the measured value deviates from the standard value by less than or equal to 3μg / m 3The chlorophyll content sensor is calibrated by standard chlorophyll solution (SPAD values are 20, 40, 60, and 80 respectively) to ensure that the measured value deviates from the standard value by less than or equal to 0.5 SPAD; during the collection process, the sensor transmits data to the data management platform in real time through the 4G network, and the platform automatically records the collection time (accurate to seconds), sensor number and corresponding geographic coordinates of each data; real-time quality control is performed on the collected data, if a certain data exceeds the sensor measurement range (such as temperature less than negative 40℃ or greater than 85℃) or the fluctuation of the continuous 3 collection values exceeds 10%, it is marked as invalid data and triggers the sensor self-checking program, to ensure that the effective rate of the finally obtained air temperature and humidity time series data, particulate matter concentration time series data, and vegetation chlorophyll content time series data is greater than or equal to 95%.

[0037] Step 303, the real-time environmental monitoring data is matched in space-time with the ground temperature data in the satellite remote sensing inversion result, and is matched in space position with the vegetation health degree data in the unmanned aerial vehicle remote sensing inversion result, to obtain a matching result, specifically including: satellite remote sensing ground temperature data is called, the satellite overflight time (accurate to minutes) and the corresponding pixel geographic coordinates (pixel center point longitude and latitude) of each data are extracted; the temperature data in the air temperature and humidity time series data is screened out from the real-time environmental monitoring data, the sensor temperature data near the same geographic coordinates in the time window of ± 15 minutes of the satellite overflight time is matched, if the straight line distance between the satellite pixel center point coordinates and the sensor coordinates is less than or equal to 15 meters, the satellite pixel resolution is 30 meters, it is ensured that the sensor is located within the pixel range, and the difference between the sensor temperature data collection time and the satellite overflight time is less than or equal to 15 minutes, then the satellite ground temperature data and the sensor temperature data are paired; if the same satellite pixel corresponds to multiple sensor temperature data in the time window, the average value is taken as the matching value; if there is no corresponding sensor data or the distance exceeds 15 meters in the time window, the satellite pixel data is marked as a matching failure.

[0038] The unmanned aerial vehicle remote sensing vegetation health degree evaluation result is called, the geographic coordinate range and the corresponding health degree grade of each vegetation health degree evaluation unit (the smallest unit is a 10m*10m grid) are extracted; the vegetation chlorophyll content time series data is screened out from the real-time environmental monitoring data, the 10m*10m grid unit in which the sensor is located is determined according to the sensor geographic coordinates, if the sensor coordinates fall within a certain grid unit, the chlorophyll content data of the sensor is paired with the vegetation health degree grade of the grid unit; if multiple chlorophyll sensors are arranged in the same grid unit, the average value of all sensor data is taken as the matching value; if there is no chlorophyll sensor arranged in a certain grid unit, the health degree data of the grid unit is marked as a matching failure.

[0039] The successfully paired satellite ground surface temperature-sensor temperature data and unmanned aerial vehicle vegetation health-sensor chlorophyll data are summarized to form a matching result table, which includes data number, geographic coordinates and collection time of satellite / unmanned aerial vehicle data, collection time and values of sensor data, and matching type (satellite matching / unmanned aerial vehicle matching); the data that fails to match is recorded, and the failure reason is analyzed, such as that the time window does not meet the requirement or the spatial distance exceeds the range, so as to ensure that the matching success rate is greater than or equal to 80%.

[0040] In step 304, a linear regression relationship between temperature data in the air temperature and humidity time series data and satellite remote sensing ground surface temperature data is established based on the matching result; based on the linear regression relationship, the ground surface temperature inversion result is calibrated to obtain calibrated satellite remote sensing data, specifically including: from the matching result, the successfully paired satellite ground surface temperature data and sensor temperature data are selected, a total of 120 effective samples are selected, covering all satellite pixel types in the ecological key area, including building land, green land and bare land corresponding pixels; taking the sensor temperature data as the independent variable (x) and the satellite ground surface temperature data as the dependent variable (y), a linear regression relationship is established by using the least square method, and the coefficients a and constant term b in the regression equation y=ax+b are calculated, for example, a=0.92 and b=1.3 are calculated, and the significance of the regression relationship is verified by R 2 test (ensuring that R 2 is greater than or equal to 0.85); the bias correction method is used for calibration: first, the bias of the satellite ground surface temperature inversion value and the sensor measured temperature value of all samples is calculated, bias=inversion value-measured value, and the average bias is obtained, for example, the average bias=2.1℃; all satellite ground surface temperature inversion pixels obtained in step 104 are corrected one by one according to calibrated ground surface temperature=original inversion ground surface temperature-average bias; after calibration, 20 groups of satellite pixel data and corresponding sensor temperature data that do not participate in regression modeling are selected for verification, to ensure that the absolute error of the calibrated ground surface temperature and the sensor measured temperature is less than or equal to 1.0℃, and finally the satellite remote sensing data containing the calibrated ground surface temperature data is obtained.

[0041] Step 305, based on the matching result, a nonlinear mapping relationship between the vegetation chlorophyll content data and the vegetation health degree evaluation result is established; based on the nonlinear mapping relationship, the evaluation result is calibrated to obtain the calibrated unmanned aerial vehicle remote sensing data, specifically including: from the matching result, the successfully paired unmanned aerial vehicle vegetation health degree evaluation result and the sensor chlorophyll content data are selected, a total of 80 groups of effective samples are selected, covering the health, sub-health and recession three health degree grades, and the number of samples in each grade is greater than or equal to 25 groups; taking the vegetation chlorophyll content data as the independent variable (x) and the quantized value corresponding to the unmanned aerial vehicle vegetation health degree (health = 100, sub-health = 60, recession = 20) as the dependent variable (y), a nonlinear mapping relationship is established by using a cubic polynomial fitting, and the fitting equation is y = cx 3 +dx 2 +ex+f, the coefficients c, d, e and f are calculated by the least square method, such as c = -0.002, d = 0.15, e = 2.3 and f = 18, and the rationality of the mapping relationship is verified by residual analysis (to ensure that the residual error is less than or equal to 5); error compensation method is used for calibration: the error (error = quantized value-theoretical value) between the unmanned aerial vehicle health degree quantized value of each health degree grade sample and the theoretical value obtained based on the nonlinear mapping relationship is calculated to obtain the average error of each grade, such as health grade average error = 8, sub-health grade average error = 5, and recession grade average error = 6; for all vegetation health degree evaluation units obtained in step 204, the average error corresponding to the health degree is compensated, the calibration value of the health grade unit = original quantized value minus 8, the calibration value of the sub-health grade unit = original quantized value minus 5, and the calibration value of the recession grade unit = original quantized value minus 6; after calibration, 15 field vegetation quadrats are selected, each quadrat is 2m x 2m, the actual health degree grade is determined by artificial investigation, the corresponding calibrated unmanned aerial vehicle health degree data of the quadrat is compared, and it is ensured that the health degree grade determination accuracy is greater than or equal to 90%, and finally the unmanned aerial vehicle remote sensing data containing the calibrated vegetation health degree data is obtained.

[0042] In the embodiment of the present application, the ground sensor network is laid in combination with the ecological key area and the UAV remote sensing inversion result, so that the sensor distribution points can be focused on the areas that need to be calibrated, and invalid distribution points can be avoided. Meanwhile, the air temperature and humidity, particulate matter concentration and vegetation chlorophyll content sensors are selected, so that the core measured parameters required for satellite and UAV inversion result calibration can be obtained, and accurate ground reference basis for subsequent data calibration is provided. The real-time environmental monitoring data is collected at a preset time frequency, so that continuous air temperature and humidity, particulate matter concentration and vegetation chlorophyll content time series data can be obtained. The time series data can cover the time nodes of satellite and UAV data collection, so that the problem that single data cannot be matched due to time deviation is avoided, and continuous data support in time dimension for subsequent space-time matching and calibration is provided. The real-time environmental monitoring data and satellite land surface temperature data are matched in space-time, so that the differences in collection time and spatial position between the two are eliminated, and the ground temperature data is accurately associated with the satellite inversion data of the corresponding area and period. The ground chlorophyll data is matched with the UAV vegetation health data in spatial position, so that the corresponding UAV inversion area of the ground chlorophyll data is determined, and accurate data association relationship for subsequent calibration of the two types of remote sensing data is established. The linear regression relationship between the ground temperature data and the satellite land surface temperature is established, so that a quantitative association model between the two is constructed. The satellite inversion result is calibrated, the deviation value calculated according to the model is used to directly correct the satellite data, so that the calibration process has a clear quantitative basis, the error caused by subjective correction is avoided, and the fitting degree of satellite remote sensing data and actual environment is improved. The non-linear mapping is established according to the non-linear relationship between vegetation health and chlorophyll content, so that the actual association characteristics between the two can be more fitted, and the fitting deviation of linear relationship to non-linear data is avoided. The UAV vegetation health evaluation result is calibrated, so that the system error in the UAV inversion process can be targetedly made up, and the calibrated UAV data is more consistent with the actual health status of the vegetation.

[0043] In a preferred embodiment of the present application, the step 400 comprises: Step 401, the calibrated satellite remote sensing data and the calibrated unmanned aerial vehicle remote sensing data are processed in the same coordinate system to obtain remote sensing data after coordinate system unification, which specifically includes: first, determining the reference of coordinate system unification, selecting WGS84 / UTM projection coordinate system (consistent with the coordinate system of remote sensing data processing in step 206); loading the calibrated satellite remote sensing data (including ground temperature, vegetation coverage, resolution 30 meters) and the calibrated unmanned aerial vehicle remote sensing data (including vegetation species distribution, health degree, roof greening coverage rate, resolution 0.1 meter) into ArcGISPro geographic information software; through the projection and transformation tools of the software, the coordinates of the two types of data are converted respectively: if the original coordinate system of satellite data is other projection, such as Gauss-Kruger, the projection grid function is called to convert it to WGS84 / UTM coordinate system; if the unmanned aerial vehicle data exists local coordinate offset in the flight process, the spatial correction tool is selected to correct the coordinates by selecting more than 10 ground control points; after conversion, 20 random space points are selected, and the coordinate values of the two types of data at the point are read respectively to ensure that the coordinate deviation of the same point is less than or equal to 0.5 meters, and finally the remote sensing data after coordinate system unification is obtained.

[0044] Step 402, the remote sensing data after coordinate system unification is processed by spatial resampling, and the remote sensing data after spatial resampling is obtained by using bilinear interpolation algorithm to unify the remote sensing data of different resolutions to the same spatial resolution, which specifically includes: first, determining the unified target spatial resolution, combining the refined evaluation demand of high-density core business district and the data processing efficiency, selecting 10 meters as the standard resolution (taking into account the wide coverage of satellite data and the micro details of unmanned aerial vehicle data); loading the satellite and unmanned aerial vehicle remote sensing data after coordinate system unification, calling the resampling tool and selecting the bilinear interpolation algorithm, which can effectively preserve the spatial texture and spectral characteristics of the data by calculating the weighted average value of the four adjacent pixels around the sampling point to obtain the new pixel value; the satellite data (30 meters) is up-sampled, and the 30-meter pixel is decomposed into 10-meter pixel, and the vegetation coverage and ground temperature information of the intermediate pixel are supplemented by bilinear interpolation; the unmanned aerial vehicle data (0.1 meter) is down-sampled, and the 0.1-meter pixel is aggregated into 10-meter pixel, and the vegetation health degree and roof greening coverage rate information of the micro pixel are merged by bilinear interpolation; after resampling, the key features of the data before and after resampling are compared, such as the average vegetation coverage of a piece of green land, to ensure that the feature deviation is less than or equal to 3%, and the remote sensing data after spatial resampling is obtained.

[0045] Step 403, time alignment processing is performed on the spatially resampled remote sensing data to obtain time-aligned remote sensing data, which specifically includes: first, the collection time of the satellite remote sensing data with the same coordinate system and spatial resampling is extracted, and the time is taken as the reference for time alignment, such as the time when Landsat-9 satellite passes through the target area on June 15, 2025, 10:15; the actual collection time of the unmanned aerial vehicle remote sensing data is queried, if the unmanned aerial vehicle data is collected for three times (such as 9:30, 10:00, and 10:30 on June 15), a linear interpolation method is used to complete and align the time series of the unmanned aerial vehicle data, and the unmanned aerial vehicle data value at 10:15 is calculated by taking the satellite reference time as the center and selecting the unmanned aerial vehicle data within 30 minutes before and after the reference time (9:45 to 10:45), for example, the vegetation health degree of a certain area at 9:30 is 65, at 10:00 is 68, and the interpolation value at 10:15 is 71; after alignment, five key ecological regions are selected, such as sparse vegetation area and high ground temperature area, and the vegetation-related information (such as satellite vegetation coverage and unmanned aerial vehicle vegetation coverage) of the same region after alignment is compared to ensure that the time consistency deviation is less than or equal to 5%, and time-aligned remote sensing data is obtained.

[0046] Step 404, spatial position matching is performed on real-time environmental monitoring data to generate environmental monitoring data with the same grid as remote sensing data through spatial interpolation method; the time-aligned remote sensing data and the environmental monitoring data are superimposed in space and time to obtain space-time consistent registration data, which specifically includes: first, the real-time environmental monitoring data (air temperature and humidity, particulate matter concentration, chlorophyll content, and point distribution) is matched in spatial position, and the Kriging interpolation method (suitable for spatial interpolation of environmental monitoring data, which can reflect the spatial correlation of data) is used to generate environmental monitoring grid data covering the target area with a grid size of 10 meters (consistent with the resolution of remote sensing data), for example, based on the point data of 20 air temperature sensors, the air temperature value of each 10-meter grid is calculated through interpolation, and the cross-validation error of the interpolation result is ensured to be less than or equal to 0.3℃; then, the time-space superposition is performed: the interpolated environmental monitoring grid data is loaded into geographic information software, the satellite reference time (10:15) determined in step 403 is used for time alignment, the time series value of the environmental monitoring data is interpolated by using the adjacent interpolation method, and the environmental monitoring data at 10:15 is extracted; then, the time-aligned remote sensing data and the environmental monitoring grid data are superimposed in space, so that each 10-meter grid unit contains remote sensing data (vegetation coverage, ground temperature, vegetation health degree, etc.) and environmental monitoring data (air temperature and humidity, particulate matter concentration, etc.) at the same time, forming space-time consistent registration data; after superposition, the data source integrity of three grid units is verified to ensure that there is no data missing field.

[0047] Step 405, multi-source data fusion processing is performed on the registration data, and a comprehensive feature data set containing spectral features, texture features and time sequence features is obtained through a data layer fusion method. Specifically, a data layer fusion (pixel-level fusion) method is used to perform multi-source data fusion on the registration data. First, features of various types of data are extracted. From satellite remote sensing data, multispectral spectral features such as red band and near-infrared band reflectivity are extracted. From unmanned aerial vehicle remote sensing data, high-resolution texture features are extracted. Texture parameters such as contrast and correlation are calculated through a gray level co-occurrence matrix, and the window size is set to 3x3. From ground environment monitoring data, time sequence features such as the average, maximum and rate of change of air temperature in the past 24 hours are extracted. Then, a comprehensive feature data set is constructed. A 10-meter grid unit is used as a row, and spectral features (5 band reflectivity) + texture features (4 texture parameters) + time sequence features (3 environmental parameter statistics) are used as columns. All feature values of each grid unit are integrated into the same data matrix. During the integration process, missing feature values are filled with the mean value of the same column to ensure that the filling ratio is less than or equal to 5%, and all feature values are normalized. Finally, a comprehensive feature data set containing 12 feature dimensions is obtained.

[0048] Step 406, based on the comprehensive feature data set, an ecological index extraction model is constructed. Specifically, first, the composition and spatial attributes of the comprehensive feature data set are determined. This data set includes spectral features, texture features and time sequence features extracted from calibrated satellite remote sensing data, unmanned aerial vehicle remote sensing data and real-time environmental monitoring data, with a total of 12 feature dimensions. The data is spatially divided into 10-meter grid units in the target area, providing a structured input basis for the construction of the ecological index extraction model.

[0049] Based on the comprehensive feature data set, a U-Net deep learning model is selected as the preset algorithm to construct the ecological index extraction model. The reason for choosing this model is that it has a special encoder-decoder structure, which can efficiently process spatial grid type data. At the same time, it can achieve fine ecological index extraction through feature downsampling and upsampling, and is suitable for high-density urban areas such as core business districts with a building area of over 200 million square meters and a building density of 65%. It meets the needs of ecological index extraction accuracy and spatial resolution.

[0050] The encoder part sets three convolution blocks, each of which is composed of two 3x3 convolution layers and one 2x2 max pooling layer, wherein the convolution layer adopts ReLU as the activation function to enhance the model's nonlinear fitting ability for feature data, and the max pooling layer is set to a step of 2 to down-sample the input integrated feature data, gradually extract deep spatial features in the data, and capture the potential relationship between different features and urban vegetation coverage indicators; the decoder part is correspondingly set with three deconvolution blocks, each of which contains one 2x2 deconvolution layer and two 3x3 convolution layers, the deconvolution layer is set to a step of 2 to up-sample the deep features extracted by the encoder, map the features back to the 10-meter grid spatial resolution consistent with the original integrated feature data set, ensure accurate matching of feature information and spatial position, and the convolution layer also adopts ReLU as the activation function to further optimize the feature mapping effect; the output layer is set with one 1x1 convolution layer, adopts Sigmoid as the activation function to make the output result in a reasonable interval of 0 to 1, and finally realizes the extraction function of three core urban vegetation coverage indicators (vegetation coverage, vegetation health, and building roof greening coverage) for each 10-meter grid unit in the target area.

[0051] Meanwhile, according to the dimensions of the integrated feature data set and the number of extracted indicators, the input dimension of the model is set to 10x10x12, wherein 10x10 corresponds to the spatial range of the 10-meter grid, and 12 corresponds to the total dimension of the three types of features of spectrum, texture, and time sequence; the output dimension of the model is set to 10x10x3, wherein 3 corresponds to the three core urban vegetation coverage indicators, and the overall construction of the ecological index extraction model is completed.

[0052] Step 407, using the comprehensive feature dataset, the ecological index extraction model is trained to obtain the trained ecological index extraction model, specifically including: the comprehensive feature dataset is divided into a training set (70% grid cells) and a validation set (30% grid cells) according to a ratio of 7:3, and the division ensures that the training set and the validation set cover all ecological types of the target area, that is, sparse vegetation area, high surface temperature area, and complex ecological weak area; an Adam optimizer (learning rate set to 0.001) is used as a training algorithm, a mean square error (MSE) is used as a loss function (suitable for continuous value regression task of vegetation coverage index), and model training is performed; in the training process, a batch gradient descent method is used, 32 training samples are input each time (batch size is set to 32), and a total of 50 epochs are trained; every 1 epoch, the loss value of the validation set is calculated, when the loss value of the validation set does not decrease for 3 consecutive epochs, the training is stopped and the current optimal model is saved (to avoid overfitting); after the training is completed, the optimal model is loaded, the mean absolute error (MAE) of the validation set is calculated, and the MAE of the vegetation coverage index is ensured to be less than or equal to 2%, the MAE of the vegetation health index is ensured to be less than or equal to 0.3 (5-point system), and the MAE of the roof green coverage rate index is ensured to be less than or equal to 1%, thereby obtaining the trained ecological index extraction model.

[0053] Step 408, inputting the registered multi-source data into the trained ecological index extraction model to extract the urban vegetation coverage index of each spatial unit in the target area, specifically including: remodeling the spatiotemporal consistent registration data (containing 10-meter grid spectrum, texture, and time series features) obtained in step 404 into a 10x10x12 feature tensor according to the requirements of the model input dimension in step 406; inputting the feature tensor into the trained ecological index extraction model, and the model outputs three urban vegetation coverage index values of each 10-meter grid unit: vegetation coverage, vegetation health, and building roof green coverage rate through forward propagation calculation, the vegetation coverage is expressed by percentage, ranging from 0 to 100%, the vegetation health is expressed by 5-point system, 1 point being the worst and 5 points being the best, and the building roof green coverage rate is expressed by percentage, ranging from 0 to 100%; after the extraction is completed, 15 sample plots are randomly selected, each sample plot is 10x10 meters, the actual index value and the model extraction value of the sample plot are compared to ensure that the absolute error of the two meets the verification standard of step 407, and finally the urban vegetation coverage index data of each spatial unit in the target area is obtained.

[0054] Step 409, based on the urban vegetation coverage index, calculate the comprehensive ecological index of each spatial unit, establish a dynamic evaluation model, including: first, determine the core input of the dynamic evaluation model as the three urban vegetation coverage indexes (vegetation coverage, vegetation health, building roof greening coverage) extracted in step 408. These indexes take the 10-meter grid unit of the target area as the basic calculation unit, providing a quantitative basis for model construction.

[0055] The first step in building the model is to determine the weight of each index. Based on the ecological characteristics of the high-density core business district (building area exceeding 200 million square meters, building density 65%), and considering the synergistic improvement of ground vegetation and roof greening on regional ecology, the contribution of each index is scored, and the average value is finally determined as the weight. Among them, the weight of vegetation coverage is 40%, the weight of vegetation health is 30%, and the weight of building roof greening coverage is 30%; Then determine the calculation logic of the comprehensive ecological index. Considering that the original data of vegetation health is 5 points, it needs to be converted to percentage (multiplied by 20) to unify the index scale. The calculation method is: the comprehensive ecological index is equal to the vegetation coverage multiplied by 0.4 plus the vegetation health multiplied by 20 and then multiplied by 0.3 plus the building roof greening coverage multiplied by 0.3, so that the index range is controlled between 0 and 100 points, directly reflecting the ecological quality; Then realize the calculation function of the model, write the index calculation program, the program reads the three index values of each 10-meter grid unit, automatically operates according to the above weighted sum logic, outputs the comprehensive ecological index of all grid units, and ensures that the calculation process is repeatable and efficient.

[0056] After that, the module structure of the dynamic evaluation model is built, including data input module, index calculation module and result output module; The data input module is responsible for receiving the monthly updated calibrated remote sensing data, unmanned aerial vehicle data and real-time environmental monitoring data, and automatically importing the latest index data through the interface; The index calculation module has the above weighted sum logic built-in, and performs batch calculation on the newly input index data; The result output module outputs the comprehensive ecological index calculated in the form of grid data, and generates index statistical report at the same time.

[0057] After the model is built, its reliability needs to be verified. Select 20 typical grid units, covering building-intensive areas, green areas, and mixed areas. Compare the comprehensive ecological index calculated by artificial method with the output results of the model to ensure that the error is less than or equal to 5 points, and verify the accuracy of the calculation logic. Since the model is built based on clear weighted rules, there is no need for parameter training of traditional machine learning models, but its dynamic nature is reflected in the automatic triggering of the index calculation module according to the monthly updated data, and the comprehensive ecological index is regenerated to realize the continuous dynamic evaluation of the ecological quality of the target area, and the whole construction of the dynamic evaluation model is completed.

[0058] Step 410, using the dynamic evaluation model, the comprehensive ecological index of all space units in the target area is graded, and the ecological quality grade space distribution map is obtained, which specifically includes: referring to the "Urban Ecological Environment Quality Evaluation Standard" and the ecological planning target of the high-density core business district, the grade division threshold of the comprehensive ecological index is determined: excellent is 80 to 100 points, the ecological condition is good, the vegetation coverage is sufficient and healthy, good is 60 to 79 points, the ecological condition is better, the vegetation coverage basically meets the standard, medium is 40 to 59 points, the ecological condition is general, there are local sparse vegetation problems, and poor is 0 to 39 points, the ecological condition is poor, the vegetation is sparse and the health degree is low; in the ArcGISPro software, load the comprehensive ecological index grid data (10 meter resolution), call the reclassification tool, assign the grid units of different score segments to the corresponding grade, excellent = 4, good = 3, medium = 2, poor = 1; then symbolize the reclassified grid data, use gradient color to distinguish the grades: excellent uses dark green, good uses light green, medium uses yellow, and poor uses red, to generate the ecological quality grade space distribution map of the target area; finally, the area and proportion of each grade area are counted, for example, the total area and proportion of the poor grade area in the core business district are calculated and marked in the distribution map legend; at the same time, the vector format (shp file) and raster format (tif file) of the distribution map are output.

[0059] In the embodiment of the present application, the two types of calibrated remote sensing data are processed in the coordinate system, which can eliminate the spatial position deviation of satellite and unmanned aerial vehicle data caused by the difference of original coordinate system, so that all remote sensing data are based on the same spatial reference for subsequent processing; the bilinear interpolation algorithm is used for spatial resampling, which can smooth and retain the spatial detail characteristics of the data in the process of unifying the resolution, unify the data of different resolutions to the same spatial resolution, realize the data matching at the pixel level, and provide consistent spatial scale support for subsequent multi-source data fusion; the time alignment processing of remote sensing data can eliminate the time sequence difference of satellite and unmanned aerial vehicle data caused by different collection time, and ensure that all remote sensing data reflect the ecological condition of the target area at the same time; the point sensor data is generated into grid data through spatial interpolation, which can supplement the spatial coverage of scattered sensor data and match the spatial format of remote sensing data; time and space superposition can integrate the time and space information of remote sensing data and environmental monitoring data, and form the registration data with consistent time and space dimensions; the data layer fusion method is used for multi-source data fusion, which can comprehensively integrate the spectral, texture features of remote sensing data and the time sequence features of environmental monitoring data, and avoid the limitation of single data feature; the comprehensive feature data set containing multiple features is generated, which can provide rich and comprehensive feature information support for subsequent ecological index extraction.

[0060] The ecological index extraction model is constructed, the deep learning can be used to process complex multi-feature data, and key information related to the ecological index in the comprehensive feature data can be automatically mined; the model is trained by using the comprehensive feature data set, the characteristics of the ecological data of the target region can be fully learned by the model, the unique properties of the vegetation cover and the environment in the region are adapted, the extraction adaptability of the model to the ecological index of the region is improved, and the extraction ability of the model that fits the actual region is ensured; after the multi-source data after registration is input into the trained model, the processing advantage of the model for multi-feature data can be used to accurately extract the urban vegetation cover index of each spatial unit; the extraction mode covering all spatial units of the target region can avoid incomplete index extraction caused by local data loss; the comprehensive ecological index is calculated, the weights can be allocated according to the contribution of each vegetation cover index to the ecological quality, and the ecological influence of different indexes is quantified; the dynamic evaluation model is established based on the index, the dispersed index data can be converted into systematic ecological evaluation logic, and the quantitative evaluation of the regional ecological condition is realized; the comprehensive ecological index is graded and a spatial distribution map is generated, the abstract index data can be converted into intuitive spatial grade information, and the ecological quality difference of different positions in the target region is clearly presented.

[0061] In a preferred embodiment of the present application, the step 405 comprises: Step 4051, the multispectral satellite remote sensing data in the registration data is subjected to spectral feature extraction processing to obtain spectral feature components, specifically comprising: first, the multispectral satellite remote sensing data in the registration data is calibrated in step 304, and 10 effective multispectral bands containing visible light, near-infrared and thermal infrared are retained, and the data spatial resolution has been unified to 10-meter grid; the multispectral data is preprocessed, the reflectivity data of all bands are first converted into a matrix format, the number of grid units is the row and column, and the dimension is the number of bands, and then each band data is subjected to standardization processing to eliminate the influence of different band reflectivity value range, such as visible light band reflectivity of 0.1 to 0.4 and near-infrared band reflectivity of 0.3 to 0.8.

[0062] First, the covariance matrix of the standardized multispectral data is calculated, and the eigenvalue and eigenvector are obtained through matrix operation; according to the size of the eigenvalue, the first three principal components with a cumulative contribution rate greater than or equal to 90% are selected, wherein the first principal component mainly reflects the spectral difference between vegetation and non-vegetation ground objects, that is, the spectral distinction between vegetation and buildings and roads in the high-density core business district, the second principal component reflects the spectral response of healthy and sub-healthy vegetation, and the third principal component reflects the subtle difference of ground cover types, such as the spectral distinction between building roofs and bare land, and the selected three principal components are used as spectral feature components.

[0063] Step 4052, texture feature calculation is performed on the high-resolution unmanned aerial vehicle remote sensing data in the registration data, and texture features are extracted, specifically including: the high-resolution unmanned aerial vehicle remote sensing data in the registration data is the 0.1-meter resolution hyperspectral image obtained in step 201, which has been calibrated in step 305 and registered to a 10-meter grid with satellite data (when resampling in step 402, the 0.1-meter data is aggregated into 10-meter unit data through bilinear interpolation, while the texture details within the unit are retained).

[0064] First, the unmanned aerial vehicle hyperspectral image is converted into a gray image, and the near-infrared band (such as the 850 nm band) most sensitive to the texture differences between vegetation and buildings is selected. The band reflectivity data is mapped to a gray value of 0 to 255 through gray stretching, ensuring that the gray differences of different ground objects (such as tree canopies, shrubs, concrete roofs, and asphalt pavements) are obvious. The matrix calculation parameters are set, the distance factor is 1, the texture correlation between adjacent pixels, the angle factor is 0°, 45°, 90°, and 135°, covering texture information in four directions, and the gray level is 256, consistent with the gray image. For each 10-meter grid unit, the gray image area contains 100x100 original 0.1-meter pixels, and four core statistics of the gray co-occurrence matrix under four angles are calculated, including contrast, correlation, energy, and entropy. The contrast reflects the difference between light and dark textures, such as the high contrast of building roof textures compared to vegetation canopies. The correlation reflects the consistency of texture direction, such as the high correlation of road textures compared to green spaces. The energy reflects the uniformity of texture, such as the low energy of bare land textures compared to tree canopies. The entropy reflects the complexity of texture, such as the high entropy of mixed vegetation areas compared to single building areas. For each 10-meter grid unit, the average value of the same statistic under four angles is taken as the final texture feature value, and four texture features are obtained, namely contrast, correlation, energy, and entropy.

[0065] Step 4053, time series feature analysis is performed on the real-time environmental monitoring data in the registration data, and the time series variation characteristics of temperature, humidity, and particulate matter concentration are extracted through the sliding window method, specifically including: the real-time environmental monitoring data in the registration data is the air temperature and humidity time series data (once every 10 minutes) and the particulate matter concentration time series data (once every 10 minutes) obtained in step 302, and the data has been spatio-temporally matched with the remote sensing data. Each 10-meter grid unit corresponds to 72 hours of continuous time series data from a nearby sensor, and 3 days of continuous monitoring data are selected for monthly evaluation, totaling 432 data points.

[0066] The window size is set to 1 hour, containing 6 consecutive data points, and the window step is 30 minutes, ensuring the continuity of the time sequence characteristics. The time sequence data of each sensor is processed by sliding window. For temperature data, the mean and variance of each window are calculated, the mean reflects the average temperature in the period, and the variance reflects the temperature fluctuation in the period. For humidity data, the mean and range of each window are calculated, the range reflects the humidity change amplitude in the period. For particulate matter concentration data, the mean and peak value of each window are calculated, the peak value reflects the pollution peak level in the period.

[0067] For each 10-meter grid unit, select the window features within 24 hours closest to the remote sensing data collection time, a total of 48 windows corresponding to 24 hours, calculate the average value of each feature, such as the 24-hour average of the temperature window mean, the 24-hour average of the particulate matter concentration window peak, and finally obtain 6 time sequence characteristics, namely temperature mean, temperature variance, humidity mean, humidity range, particulate matter mean, and particulate matter peak.

[0068] Step 4054, fuse the spectral feature components, texture features, and time sequence change rule features to obtain the fused multi-dimensional feature data set, which specifically includes: first determine the spatial reference for feature layer fusion as the 10-meter grid unit, consistent with the spatial resolution of the registered data, each grid unit corresponds to 3 spectral feature components of step 4051, 4 texture features of step 4052, and 6 time sequence features of step 4053, a total of 13 original feature dimensions.

[0069] Take the geographic coordinates (latitude and longitude) of the grid unit as the index, and sequentially splice the spectral feature vector (3-dimensional), texture feature vector (4-dimensional), and time sequence feature vector (6-dimensional) of each grid unit to form a 13-dimensional feature vector; aggregate the feature vectors of all grid units to construct a multi-dimensional feature matrix, with the rows corresponding to the number of grid units and the columns corresponding to the 13 feature dimensions.

[0070] For each grid unit, check whether the feature values meet the expected range, such as the spectral feature component value range of -2 to 2, the texture contrast value range of 0 to 1000, and the temperature mean value range of 25℃ to 35℃. If a feature value exceeds the reasonable range, replace it with the feature mean value of the grid within a 3x3 range around the grid to ensure the integrity and reasonableness of the multi-dimensional feature data; the final fused multi-dimensional feature data set obtained by each grid unit contains spectral, texture, and time sequence features, and fully covers the spatial, spectral, and time dimension information required for high-density urban ecological evaluation.

[0071] At step 4055, the fused multi-dimensional feature data set is standardized to eliminate dimensional differences, and a comprehensive feature data set is obtained, specifically including: first, analyzing the dimensional differences of the fused multi-dimensional feature data set, the spectral feature component is the standardized principal component value (dimensionless, value range-2 to 2), the texture feature is a statistical quantity (such as the unit of contrast is gray value square, the value range is 0 to 1000), the time sequence feature includes temperature (unit ℃, value 25 to 35), humidity (unit %RH, value 40 to 70), particulate matter concentration (unit μg / m 3 , value 20 to 80), the numerical range of different features and the dimensional difference are significantly different, and need to be eliminated by standardization.

[0072] The min-max standardization method is used to process each feature dimension separately. For each feature (a total of 13), the minimum and maximum values of the feature values of all grid cells are calculated. According to the logic of standardized value=(original value-minimum value) / (maximum value-minimum value), the original value of each grid cell of the feature is converted to a standardized value of 0 to 1. If the maximum and minimum values of a certain feature are equal (such as the particulate matter concentration of all grids in a certain area is the same), the standardized value of all features is set to 0.5 to avoid the problem of zero denominator in the division operation. After the standardization is completed, the 13 standardized feature values of each grid cell are summarized to form a comprehensive feature data set. All features of the data set are in the same range of 0 to 1, and there is no dimensional difference.

[0073] In the embodiment of the present application, the multispectral satellite remote sensing data in the registration data is subjected to spectral feature extraction processing to retain the key information reflecting the vegetation and ground conditions in the spectral data; the texture features are extracted from the high-resolution unmanned aerial remote sensing data, which can capture the micro spatial distribution rules of the vegetation canopy structure, building roof texture and other micro spatial distribution rules in the image, supplement the spatial detail information that the spectral features cannot reflect, and make the extracted features more in line with the actual conditions of the complex underlying surface of the high-density urban area; the real-time environmental monitoring data is analyzed by the sliding window method to analyze the time sequence features, which can effectively capture the change trend and fluctuation rule of temperature, humidity and particulate matter concentration with time, convert the discrete monitoring data into features with time correlation, and supplement the time dimension information lacking in the spatial features; the spectral feature components, texture features and time sequence change rule features are fused to form a multi-dimensional feature data set covering the spatial, spectral and time dimensions, which can integrate the advantages of the respective features of the multi-source data, break the limitations of single type features, and provide more comprehensive feature support for subsequent extraction of ecological indexes; the multi-dimensional feature data set is standardized to eliminate the numerical range deviation caused by the dimensional differences of different features (such as spectral value, texture statistical quantity and concentration value).

[0074] In a preferred embodiment of the present application, step 5 comprises: Step 501, based on the ecological quality grade spatial distribution map, identify the area with the lowest ecological quality grade, determine the ecological weak area, specifically including: first, obtain the ecological quality grade spatial distribution map, which divides the target area (high-density core business district and surrounding area, building area exceeding 2000 square meters, building density 65%) into four grades of excellent, good, medium and poor, with the poor grade being the lowest; load the spatial distribution map in ArcGIS geographic information system software, filter out all grid units with poor grade field through attribute query function, and merge these grid units to form continuous or sporadic distribution areas; extract the geographic coordinate range of the merged areas, such as east longitude 116.33° to 116.38°, north latitude 39.94° to 39.97°, etc., and calculate the area of each region, such as the continuous area of about 2.5km 2 , the area of a single block of sporadic area is 0.1 to 0.3km 2 , and at the same time, combined with the land feature type (building intensive area, bare land, sporadic green land) to supplement the label, finally determine these labeled poor grade areas as ecological weak areas.

[0075] Step 502, according to the spatial distribution characteristics of the ecological weak area, adjust the monitoring scheme of the unmanned aerial vehicle remote sensing platform to obtain the unmanned aerial vehicle monitoring optimization scheme, the unmanned aerial vehicle monitoring optimization scheme includes increasing the monitoring frequency of the continuous distribution ecological weak area, and optimizing the flight path of the sporadic distribution ecological weak area, specifically including: from the ecological weak area, analyze its spatial distribution characteristics: divide the area greater than or equal to 1km 2 and spatially continuous into continuous distribution ecological weak areas, such as the 2.5km 2 area in the northeast of the core business district with dense buildings and sparse vegetation, divide the area less than 1km 2 and scattered in more than three different locations into sporadic distribution ecological weak areas, such as five scattered 0.1 to 0.2km 2bare land; adjust the UAV monitoring scheme for continuous distribution areas: the original monitoring frequency is once a month, and it is adjusted to twice a month, and the flight parameters remain consistent with step 201 (flight height 120 meters, ground resolution 0.1 meters, adjacent flight line overlap rate 80%), the newly added monitoring period is selected in the middle of each month, and the original monitoring in the first of each month forms an interval, which ensures timely capture of ecological changes; for sporadic distribution areas, optimize the flight path: abandon the original grid-shaped general path, use a radial path, take the center position of the area as the take-off and landing point, such as longitude 116.42°, latitude 39.98°, plan 5 radial flight lines to cover 5 sporadic areas, control the flight speed of each flight line at 8 m / s, and reduce the heading overlap rate to 50%, which shortens the flight time by about 30 minutes compared with the original scheme, while ensuring that all sporadic areas are within the UAV image coverage, and there is no data missing, and finally integrate to form a UAV monitoring optimization scheme including monitoring frequency, flight path, and parameter setting.

[0076] Step 503, according to the environmental characteristics of the ecological weak area, combined with the vegetation health evaluation result in the UAV remote sensing inversion result, to obtain the sensor layout optimization scheme, specifically including: for each ecological weak area determined in step 501, combined with high-resolution UAV images (0.1 meter resolution) to judge its ground object composition, such as some areas dominated by building roofs + sporadic declining vegetation, some areas dominated by bare land + road edges, and some areas dominated by sub-healthy green land; At the same time, call the UAV remote sensing inversion result, extract the vegetation health level (mainly declining and sub-healthy) in each ecological weak area; according to the environmental characteristics and vegetation health, formulate the sensor layout optimization scheme: for building roof + sporadic declining vegetation area, supplement the layout of vegetation chlorophyll content sensor, 2 in each declining vegetation concentration area (area greater than or equal to 0.2km 2 ) is laid out, fixed in the upper part of the vegetation canopy, i.e. height 2 to 2.5 meters, to monitor the change of vegetation health state; for bare land + road edge area, supplement the layout of air temperature and humidity sensor and particulate matter concentration sensor, 1 per 1km 2 is laid out, placed on a fixed support 1.5 meters high beside the road to monitor the impact of bare land dust on the environment; for sub-healthy green area, increase the layout density based on the original sensor, adjust the air temperature and humidity sensor from 1 every 2km 2 to 1 every 1.5km 2 , ensure that all sub-healthy vegetation growth areas in the green land are covered; record the specific location of each supplementary sensor (such as longitude 116.35°, latitude 39.95°, located in the west green land of XX Mansion), sensor type and layout height, and form a sensor layout optimization scheme.

[0077] Step 504, the unmanned aerial vehicle monitoring optimization scheme and the sensor layout optimization scheme are integrated to form a monitoring network optimization scheme, specifically including: first, the unmanned aerial vehicle monitoring optimization scheme is sorted out, and the contents include: the monitoring frequency of the continuous distribution ecological weak area (2 times per month), the monitoring time period (early and middle of the month), the flight parameters (height 120 meters, overlap rate 80%), the optimized flight path of the sporadic distribution ecological weak area (radial flight line, take-off and landing point coordinates, 5 flight line coverage range), and the flight time (30 minutes shorter than the original scheme); second, the sensor layout optimization scheme is sorted out, and the contents include: the type of the supplementary sensor (chlorophyll content sensor, air temperature and humidity sensor, particulate matter concentration sensor), the specific geographic coordinates and layout height of each sensor, the environmental characteristics (building roof area, bare area, sub-health green land) corresponding to the layout area, and the data acquisition frequency; the contents of the two types of schemes are integrated into the same document, the document includes three parts of the optimization target (focusing on the ecological weak area and improving the monitoring pertinence), the unmanned aerial vehicle monitoring optimization details and the sensor layout optimization details, and the spatial distribution map of the ecological weak area is attached, in which the optimized unmanned aerial vehicle flight line and the position of the added sensor are marked; finally, a scheme implementation schedule is formulated, the first optimization monitoring time of the unmanned aerial vehicle is determined (next month early), and the sensor installation completion time is determined (this week), to form a complete monitoring network optimization scheme.

[0078] In the embodiment of the present application, the lowest grade area is identified as the ecological weak area based on the spatial distribution map of the ecological quality grade, which can directly rely on the existing visualized grade division result to accurately locate the area with the worst ecological condition; the unmanned aerial vehicle monitoring scheme is adjusted according to the difference between the continuous and sporadic spatial distribution characteristics of the ecological weak area, the monitoring frequency of the continuous area is increased to strengthen the dynamic tracking of the regional ecological change, the flight path of the sporadic area is optimized to reduce invalid flight and improve the coverage efficiency, the resource waste of the unified monitoring scheme is avoided, and the unmanned aerial vehicle monitoring is more suitable for the regional distribution characteristics; the sensor layout scheme is formulated in combination with the environmental characteristics of the ecological weak area and the vegetation health degree evaluation result of the unmanned aerial vehicle, which can take into account the regional environmental differences (such as building density and green land distribution) and the vegetation growth condition; the sensor layout position and type are more suitable for the actual monitoring demand, blind layout is avoided, and the pertinence and effectiveness of the ground monitoring data are improved; the unmanned aerial vehicle monitoring optimization scheme and the sensor layout optimization scheme are integrated, a complete monitoring network optimization scheme covering the air and the ground can be formed; the limitation of single monitoring method optimization is avoided, the cooperation of the dynamic tracking in the air and the fixed-point monitoring on the ground is realized, and the overall coverage and monitoring capability of the monitoring network on the ecological weak area are improved.

[0079] As shown in Figure 2 The embodiment of the present application also provides a city ecological environment remote sensing quantitative evaluation system based on space-air-ground cooperation, which comprises: The cooperative monitoring module is configured to acquire ecological base data of a target region through a satellite remote sensing platform; perform inversion processing on the ecological base data to obtain satellite remote sensing inversion results of regional vegetation coverage and ground surface temperature; identify the satellite remote sensing inversion results, and determine an ecological key region based on a preset threshold; collect ecological data of the ecological key region through a UAV remote sensing platform, and process the ecological data to obtain UAV remote sensing inversion results of vegetation species distribution, vegetation health degree, and building roof greening coverage rate; The calibration module is configured to acquire real-time environmental monitoring data through a ground sensor network arranged in the target region; use the real-time environmental monitoring data to perform data calibration on the satellite remote sensing inversion results and the UAV remote sensing inversion results, respectively, to obtain calibrated satellite remote sensing data and UAV remote sensing data; The evaluation module is configured to perform spatio-temporal registration and fusion on the calibrated satellite remote sensing data, the UAV remote sensing data, and the real-time environmental monitoring data, extract urban vegetation coverage indexes, establish a dynamic evaluation model, and obtain an ecological quality grade spatial distribution map of the target region. The dynamic optimization module is configured to dynamically optimize monitoring network configuration and generate a monitoring network optimization scheme according to an ecological weak region identified in the ecological quality grade spatial distribution map, wherein the monitoring network configuration includes adjusting a UAV monitoring frequency and supplementing ground sensor arrangement.

[0080] It should be noted that the system corresponds to the method described above, and all implementation manners in the method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0081] The above describes the preferred embodiments of the present application. It should be noted that, for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which should also be considered within the scope of protection of the present application.

Claims

1. A method for remote sensing quantitative evaluation of urban ecological environment based on space-air-ground cooperation, characterized in that, The method comprises: Step 100, obtaining ecological base data of a target region by a satellite remote sensing platform; performing inversion processing on the ecological base data to obtain satellite remote sensing inversion results of regional vegetation coverage and ground surface temperature; identifying the satellite remote sensing inversion results and determining ecological key regions based on a preset threshold value; Step 200, collecting ecological data of the ecological key regions by a UAV remote sensing platform, and processing the ecological data to obtain UAV remote sensing inversion results of vegetation species distribution, vegetation health degree and building roof greening coverage rate; Step 300, obtaining real-time environmental monitoring data by a ground sensor network arranged in the target region; using the real-time environmental monitoring data, respectively calibrating the satellite remote sensing inversion results and the UAV remote sensing inversion results to obtain calibrated satellite remote sensing data and UAV remote sensing data; Step 400, performing space-time registration and fusion on the calibrated satellite remote sensing data, the UAV remote sensing data and the real-time environmental monitoring data, extracting urban vegetation coverage indexes, establishing a dynamic evaluation model, and obtaining an ecological quality grade spatial distribution map of the target region; Step 500, dynamically optimizing monitoring network configuration and generating a monitoring network optimization scheme according to the ecological weak regions identified in the ecological quality grade spatial distribution map, wherein the monitoring network configuration comprises adjusting the UAV monitoring frequency and supplementing the ground sensor arrangement.

2. The method according to claim 1, wherein, Step 100 comprises: Obtaining multispectral remote sensing images of a target region by a satellite remote sensing platform as ecological base data; Preprocessing the large-range ecological base data, including radiation calibration, atmospheric correction and geometric registration, to obtain ground surface reflectivity data; Based on the ground surface reflectivity data, calculating a normalized vegetation index, converting the normalized vegetation index into vegetation coverage, and obtaining vegetation coverage inversion results; Based on the ground surface reflectivity data, calculating a radiation brightness value of a thermal infrared band; based on the radiation brightness value, extracting a ground surface emissivity; combining atmospheric profile parameters, performing inversion processing on the radiation brightness value and the ground surface emissivity to obtain a ground surface temperature inversion result; Spatially superimposing the vegetation coverage inversion results and the ground surface temperature inversion results to obtain satellite remote sensing inversion results of regional vegetation coverage and ground surface temperature; Statistically analyzing the satellite remote sensing inversion results to respectively extract vegetation coverage values and ground surface temperature values; Comparing the vegetation coverage values with a first preset threshold value to identify regions lower than the first preset threshold value and mark them as sparse vegetation regions; comparing the ground surface temperature values with a second preset threshold value to identify regions higher than the second preset threshold value and mark them as high ground surface temperature regions; Spatially merging the sparse vegetation regions and the high ground surface temperature regions to obtain ecological key regions.

3. The method according to claim 2, wherein, Step 200 comprises: Based on the ecological key regions, planning a UAV flight path, and collecting data by a UAV remote sensing platform carrying a hyperspectral camera to obtain hyperspectral image data of the ecological key regions; Preprocessing the hyperspectral image data, including radiation correction, spectral correction and geometric fine correction, to obtain hyperspectral reflectivity data; Extracting spectral features of different vegetation species based on the hyperspectral reflectivity data; and classifying the spectral features by using a spectral angle mapping method to obtain vegetation species distribution information; Based on the hyperspectral reflectivity data, calculating a normalized vegetation index, and dividing the normalized vegetation index according to a preset health degree threshold to obtain a vegetation health degree evaluation result; Based on the hyperspectral reflectivity data, identifying a building roof contour, extracting a vegetation pixel proportion within the contour, and calculating a building roof green coverage rate; Integrating the vegetation species distribution information, the vegetation health degree evaluation result and the building roof green coverage rate to obtain a UAV remote sensing inversion result.

4. The method according to claim 3, wherein, Step 300 includes: Based on the ecological key area and the UAV remote sensing inversion result, arranging a ground sensor network in a target area, wherein the ground sensor network includes air temperature and humidity sensors located in green land, particulate matter concentration sensors, and chlorophyll content sensors located in vegetation canopies; Collecting real-time environmental monitoring data at a preset time frequency through the ground sensor network to obtain air temperature and humidity time series data, particulate matter concentration time series data and vegetation chlorophyll content time series data; Spatiotemporally matching the real-time environmental monitoring data with ground temperature data in satellite remote sensing inversion results, and spatially and positionally matching the real-time environmental monitoring data with vegetation health degree data in the UAV remote sensing inversion result to obtain a matching result; Based on the matching result, establishing a linear regression relationship between temperature data in the air temperature and humidity time series data and satellite remote sensing ground temperature data; and based on the linear regression relationship, calibrating a ground temperature inversion result to obtain calibrated satellite remote sensing data; Based on the matching result, establishing a nonlinear mapping relationship between vegetation chlorophyll content data and vegetation health degree evaluation results; and based on the nonlinear mapping relationship, calibrating the evaluation results to obtain calibrated UAV remote sensing data.

5. The method according to claim 4, wherein, Step 400 includes: Performing coordinate system unification processing on the calibrated satellite remote sensing data and the calibrated UAV remote sensing data to obtain coordinate system unified remote sensing data; Performing spatial resampling processing on the coordinate system unified remote sensing data, and uniformly converting remote sensing data of different resolutions to the same spatial resolution by using a bilinear interpolation algorithm to obtain spatially resampled remote sensing data; Performing time alignment processing on the spatially resampled remote sensing data to obtain time-aligned remote sensing data; Performing spatial position matching processing on the real-time environmental monitoring data, and generating environmental monitoring data of the same grid as the remote sensing data by spatial interpolation method from point-distributed sensor monitoring data; and temporally and spatially superimposing the time-aligned remote sensing data and the environmental monitoring data to obtain spatiotemporally consistent registration data; Performing multi-source data fusion processing on the registration data to obtain a comprehensive feature data set containing spectral features, texture features and time series features by a data layer fusion method.

6. The space-air-ground collaborative based urban ecological environment remote sensing quantitative evaluation method according to claim 5, characterized in that, Step 400 further includes: Based on the comprehensive feature data set, constructing an ecological index extraction model; The comprehensive feature dataset is used for training the ecological index extraction model, so as to obtain a trained ecological index extraction model; The registered multi-source data is input into the trained ecological index extraction model, so as to extract the urban vegetation coverage index of each spatial unit in the target region; Based on the urban vegetation coverage index, the comprehensive ecological index of each spatial unit is calculated, and a dynamic evaluation model is established; The comprehensive ecological index of all spatial units in the target region is classified by using the dynamic evaluation model, so as to obtain an ecological quality level spatial distribution map.

7. The space-air-ground collaborative based urban ecological environment remote sensing quantitative evaluation method according to claim 6, characterized in that, The registered data is subjected to multi-source data fusion processing, and a comprehensive feature dataset containing spectral features, texture features and time sequence features is obtained by a data layer fusion method, including: The multi-spectral satellite remote sensing data in the registered data is subjected to spectral feature extraction processing, so as to obtain spectral feature components; The high-resolution unmanned aerial vehicle remote sensing data in the registered data is subjected to texture feature calculation, so as to extract texture features; The real-time environmental monitoring data in the registered data is subjected to time sequence feature analysis, and the time sequence variation law features of temperature, humidity and particulate matter concentration are extracted by a sliding window method; The spectral feature components, texture features and time sequence variation law features are subjected to feature layer fusion, so as to obtain a fused multi-dimensional feature dataset; The fused multi-dimensional feature dataset is subjected to standardization processing, so as to eliminate dimensional differences and obtain a comprehensive feature dataset. 8.The method according to claim 7, characterized in that, Step 500 includes: Based on the ecological quality level spatial distribution map, the region with the lowest ecological quality level is identified as an ecological weak region; According to the spatial distribution characteristics of the ecological weak region, the monitoring scheme of the unmanned aerial vehicle remote sensing platform is adjusted, so as to obtain an unmanned aerial vehicle monitoring optimization scheme, which includes increasing the monitoring frequency of the continuously distributed ecological weak region and optimizing the flight path of the sporadically distributed ecological weak region; According to the environmental characteristics of the ecological weak region, the sensor arrangement optimization scheme is obtained by combining the vegetation health degree evaluation result in the unmanned aerial vehicle remote sensing inversion result; The unmanned aerial vehicle monitoring optimization scheme and the sensor arrangement optimization scheme are summarized to form a monitoring network optimization scheme.

9. A space-air-ground collaborative-based urban ecological environment remote sensing quantitative evaluation system, the system implements the method of any one of claims 1 to 8, characterized in that, It includes: A cooperative monitoring module is configured to acquire ecological base data of a target region through a satellite remote sensing platform; The ecological base data is subjected to inversion processing, so as to obtain satellite remote sensing inversion results of regional vegetation coverage and land surface temperature; The satellite remote sensing inversion results are identified, and an ecological key region is determined based on a preset threshold; Ecological data of the ecological key region are collected by an unmanned aerial vehicle remote sensing platform, and the ecological data are processed, so as to obtain unmanned aerial vehicle remote sensing inversion results of vegetation species distribution, vegetation health degree and building roof greening coverage rate; A calibration module is configured to acquire real-time environmental monitoring data through a ground sensor network arranged in the target region; The real-time environmental monitoring data are used for calibrating the satellite remote sensing inversion results and the unmanned aerial vehicle remote sensing inversion results respectively, so as to obtain calibrated satellite remote sensing data and unmanned aerial vehicle remote sensing data; An evaluation module is configured to perform spatio-temporal registration and fusion on the calibrated satellite remote sensing data, the unmanned aerial vehicle remote sensing data and the real-time environment monitoring data, extract urban vegetation coverage indexes, establish a dynamic evaluation model, and obtain a spatial distribution map of ecological quality grades of the target region; A dynamic optimization module is configured to dynamically optimize a monitoring network configuration and generate a monitoring network optimization scheme according to an ecological weak region identified in the spatial distribution map of ecological quality grades, wherein the monitoring network configuration comprises adjusting an unmanned aerial vehicle monitoring frequency and supplementing ground sensor arrangement.

10. A computing device, comprising: comprising: one or more processors; a memory device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the method according to any one of claims 1 to 7.

Images