Alpine grassland quality inversion method based on hyperspectral observation satellite remote sensing image

Abstract

The invention discloses an alpine grassland quality inversion method based on a hyperspectral observation satellite remote sensing image. The method comprises the following steps: preprocessing a purchased GF5B hyperspectral image; automatically extracting the reflectivity of a remote sensing image region of interest, and calling a continuous projection algorithm to automatically identify sensitive wavebands important for grassland quality index monitoring; constructing an inversion model of sensitive wave bands and grassland quality indexes (crude protein and crude fiber content, total digestible substance amount and total digestible protein amount) by combining part of measured values; and inverting the grassland quality of the whole area by using the model, and analyzing the spatial distribution of the grassland quality. The method is less in human assistance and high in calculation efficiency, can invert the alpine grassland quality index in a large-scale and near-real-time manner, further obtains the spatial distribution condition and growth condition of the forage grass quality, and provides auxiliary suggestions for alpine grassland quality evaluation and reasonable planning of the livestock capacity.

Description

Technical Field

[0001] The invention relates to the field of domestic satellite remote sensing applications and ecological environment monitoring, and specifically is a high-cold grassland quality inversion method based on hyperspectral observation satellite remote sensing images. Background Art

[0002] Natural grassland accounts for 14.7% of my country's land surface [1] , the total area reaches 4*10 8 hm 2 It is the largest terrestrial ecosystem in China. [2] It is not only the material basis for the survival and development of herders, but also an important barrier to stabilize the regional ecosystem and maintain the ecological functions of the plateau. [3] Grazing is the most common land use in alpine grasslands [4] Reasonable forage utilization is the key to the healthy development of grassland ecosystem [5] , and the monitoring of forage yield and quality indicators is the core indicator of rational use. However, the current monitoring of alpine grasslands pays more attention to forage yield indicators and ignores quality indicators, resulting in inaccurate assessment of stocking capacity and irrational grassland utilization.

[0003] Traditionally, the quality evaluation of alpine grasslands uses manual surveys, which involve sampling and surveying the crude protein (g / kg), crude fiber content (%), total digestible matter (kg / 100kg), and total digestible protein (kg / 100kg) of grasslands, and manually analyzing and judging the grassland quality based on the content differences of these indicators. This method relies too much on manual experience, is costly, and is inefficient, making it difficult to meet the requirements of near-real-time and high-efficiency monitoring of large-scale forage quality.

[0004] The development of satellite remote sensing technology has made it possible to quickly monitor the quality of alpine grasslands. However, the current methods of using satellites to monitor forage quality are mainly based on foreign satellites, such as MODIS, which have low spatial resolution (MODIS 250m) and a small number of optional bands, making it difficult to effectively characterize the subtle changes in forage quality. With the development of my country's satellite remote sensing technology, especially the launch of the hyperspectral observation satellite (GF5B), which includes 330 bands from ultraviolet to short-wave infrared, with a wide spectral range and high spectral resolution, it provides high spatial resolution (30m) rich spectral information for alpine grassland forage quality monitoring, which provides convenience for the use of domestic satellite hyperspectral remote sensing images to monitor alpine grassland quality indicators.

[0005] The references of background technology are as follows:

[0006] [1] Shen Haihua, Zhu Yankun, Zhao Xia, Geng Xiaoqing, Gao Shuqin, Fang Jingyun, "Analysis of the current status of grassland resources in China," Chinese Science Bulletin, vol. 61, no. 2, pp. 139-154, 2016.

[0007] [2] Yan Lingyan, Kong Lingqiao, Zhang Lu, Ouyang Zhiyun, Hu Jinming, "Concepts, methods and key issues of grassland ecosystem carrying capacity," Chinese Journal of Eco-Agriculture, vol. 30, no. 8, p. 10, 2022.

[0009] [3] Mo Xingguo, Liu Wen, Meng Chengcheng, Hu Shi, Liu Suxia, Lin Zhonghui, "Changes in grassland yield and grass-livestock balance on the Qinghai-Tibet Plateau," Chinese Journal of Applied Ecology, vol. 32, no. 7, pp. 2415–2425, 2021.

[0011] [4] Yu Hui, Wang Genxu, Yang Yan, and Lü Yaqiong, "The concept of grassland green carrying capacity and its application framework in national parks," Acta Ecologica Sinica, vol. 40, no. 20, pp. 7248-7254, 2020.

[0013] [5]Xu Minyun, He Jinsheng, "Progress in research on grassland carrying capacity: concepts, theories and models," Acta Prataculturae Sinica, vol. 23, no. 03, pp. 313-324, 2014. Summary of the invention

[0014] In view of the shortcomings of the current alpine grassland quality index inversion algorithm based on domestic satellites, the present invention provides an alpine grassland quality inversion method based on hyperspectral observation satellite remote sensing images, which can achieve large-scale and high-efficiency monitoring and inversion of alpine grassland forage quality indicators (crude protein, crude fiber content, total digestible matter, total digestible protein) with less human intervention, providing a new idea for quickly understanding the quality of alpine grassland.

[0015] The technical solution adopted by the present invention is: a method for inverting the quality of alpine grassland based on hyperspectral observation satellite remote sensing images, comprising the following steps:

[0016] Step 1: Purchase GF5B hyperspectral images of the hyperspectral observation satellite with cloud coverage less than 30% according to the scope of the proposed study area;

[0017] Step 2: Preprocess the purchased GF5B hyperspectral image, convert the original data of the GF5B hyperspectral remote sensing image into the spectral reflectance of each band, and eliminate the influence of the atmosphere on the satellite data. The preprocessing includes radiation correction and geometric correction;

[0018] Step 3: Extract accurate images of the study area and extract spectra of the ROI: Use the mask tool in the ArcMap toolbox to clip and obtain the precise range of the study area to obtain a clipped image; randomly select no less than 50 ROIs on the clipped image, with no less than 3 pixels for each ROI, and use the "Multi-value Extraction to Point" tool in the ArcMap Spatial Analysis Toolbox to extract the spectral information of limited measured points to obtain the spectral data of the ROI;

[0019] Step 4: Use an integrated genetic algorithm and an improved K-means clustering algorithm with repeated iterations to reduce the dimension of the spectral data of the ROI, automatically screen the spectral band range that can effectively characterize the characteristics of the study area, and use the spectral band range to extract the surface reflectance of the corresponding band of the measured point;

[0020] Step 5: constructing inversion models of alpine grassland quality indicators respectively: the alpine grassland quality indicators include crude protein (g / kg), crude fiber content (%), total digestible matter (kg / 100kg), and total digestible protein (kg / 100kg). The measured data are divided into a training set and a validation set. Combined with the feature selection results, each band selected on the training set is substituted into the partial least squares regression algorithm one by one to establish the model, and the estimated value is obtained on the validation set;

[0021] Step 6: Model evaluation and determination of the optimal inversion model: Calculate the root mean square error (RMSE) and determination coefficient (R) of the measured and estimated values ​​of the alpine grassland quality indicators on the validation set. 2 , according to the root mean square error RMSE and the coefficient of determination R 2 Determine the optimal inversion model for each alpine grassland quality index;

[0022] Step 7: Overall inversion of alpine grassland quality indicators in the study area: Use the spectral band range screened in step 4 to screen the spectral information of the entire study area. According to the optimal inversion model of each alpine grassland quality indicator, the forage quality parameters of the entire study area are inverted. The spatial distribution of each quality indicator of the alpine grassland is comprehensively analyzed to reveal the growth status of the alpine grassland and provide suggestions for the reasonable planning of livestock carrying capacity.

[0023] Furthermore, the step 1 comprises:

[0024] Step 1.1, use ArcMap vectorization to obtain the vector boundary of the proposed study area;

[0025] Step 1.2: On the Land Observation Satellite Data Service Platform, select the hyperspectral payload of GF5B, set the cloud coverage to less than 30%, and purchase appropriate satellite images based on vector boundaries and time.

[0026] Furthermore, the step 1.1 includes:

[0027] Step 1.1.1, specify the spatial coordinate system and create the surface element extracted by the vector range;

[0028] Step 1.1.2: According to the display of the study area in the GF5B image, vectorize along the boundary of the study area, complete the creation of surface elements, and derive the vector boundary of the proposed study area.

[0029] Furthermore, the step 2 comprises:

[0030] Step 2.1, using the ENVI 5.6 radiometric correction tool, 330 bands of visible light and short-wave infrared of the GF5B hyperspectral image were radiometrically calibrated to obtain the surface reflectance;

[0031] Step 2.2: Use the ENVI 5.6 geometric correction tool to call the HDR header file information of GF5B and set the parameters to obtain the geometrically corrected GF5B image.

[0032] Furthermore, the step 3 comprises:

[0033] Step 3.1, import the corrected image into ArcMap, use the mask tool to extract the exact range of the study area, and obtain the cropped image;

[0034] Step 3.2, manually select no less than 50 ROIs from the cropped image in ENVI5.6, and export the vector range and spectrum of the ROI for later use;

[0035] Step 3.3, extract the longitude and latitude of the center point of each ROI vector range in ArcMap as the measured point coordinates of the ROI, and use the "Multi-value Extraction to Point" tool to export the spectral information of each measured point for future use.

[0036] Further, the step 4 comprises:

[0037] Step 4.1, standardization, scaling the ROI spectral data to a similar range to avoid the adverse effects of differences between features on the clustering results;

[0038] Step 4.2, randomly select a feature subset of the spectrum, and randomly select an initial feature subset from all spectral bands as the starting point;

[0039] Step 4.3, iterative optimization, adding or deleting certain bands to achieve feature subset screening;

[0040] Step 4.4, spectral band selection convergence condition, when the predetermined maximum number of iterations is reached or the clustering performance is no longer significantly improved, the iteration is stopped;

[0041] Step 4.5, analysis of spectral screening results, analysis of the spectral band range and number of bands of the optimal feature subset, determination of the spectral band range that can effectively characterize the characteristics of the study area based on the correlation between the band range and vegetation (grassland) parameters, and use of the determined spectral band range to extract the surface reflectance of the corresponding band of the measured point.

[0042] Further, the step 4.3 includes:

[0043] Step 4.3.1, feature subset evaluation, applying the K-means clustering algorithm to the currently selected feature subset, and calculating the evaluation index of the clustering result, wherein the evaluation index includes the overall intra-cluster sum of squares;

[0044] Step 4.3.2, feature subset update, calling the genetic algorithm, adding or removing a spectral band to update the feature subset, exploring a wider range of combinations, and evaluating the quality of each combination through the evaluation function;

[0045] Step 4.3.3, evaluation function, evaluates the performance of each feature subset based on the overall within-cluster sum of squares, a quality indicator of K-means clustering.

[0046] Further, the step 5 comprises:

[0047] Step 5.1, divide the measured grassland quality indicators into training set and validation set in a ratio of 8:2;

[0048] Step 5.2, on the training set, each selected band is substituted into the partial least squares regression algorithm one by one to establish the inversion model of crude protein, crude fiber content, total digestible matter, and total digestible protein;

[0049] Step 5.3, on the validation set, based on the training model, the estimated values ​​of the forage quality indicators crude protein, crude fiber content, total digestible matter, and total digestible protein are obtained.

[0050] Further, the step 6 comprises:

[0051] Step 6.1: Calculate the R of the estimated and measured values ​​of the quality indicators of alpine grassland, crude protein, crude fiber content, total digestible matter, and total digestible protein on the validation set. 2 and RMSE;

[0052] Step 6.2: Compare the R of each model 2 and RMSE, select R 2 The model with the highest and lower RMSE is taken as the optimal model for each quality index.

[0053] Further, the step 7 comprises:

[0054] Step 7.1, reduce the dimension of the spectrum of the entire study area according to the band range screened in step 4, and only retain the spectrum of the optimal subset;

[0055] Step 7.2, according to the optimal inversion model of each quality index determined in step 6, based on the spectral information of the optimal subset of the entire study area, invert the quality index of the entire study area;

[0056] Step 7.3: Analyze the growth status of alpine grassland forage based on the forage quality indicators of the entire study area and combined with geographical location information.

[0057] The present invention proposes a method for inverting the quality of alpine grassland based on hyperspectral observation satellite remote sensing images. The method uses the hyperspectral images of the domestic hyperspectral observation satellite (GF5B), obtains accurate images of the study area through preprocessing, and comprehensively uses ArcMap and ENVI5.6 toolkits to extract the spectrum of the area of ​​interest, uses the improved K-means clustering algorithm to screen effective spectral information to achieve spectral dimension reduction, and uses the unsupervised algorithm to establish a model relationship between the screened spectrum and limited measured data, and then inverts to obtain the spatial distribution overview of large-scale forage quality indicators of alpine grassland. This method uses fewer bands, has low manual intervention, and has high monitoring efficiency, and provides a new idea for quickly obtaining the growth status of large-scale alpine grassland using GF5B satellite images. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A schematic diagram of purchasing hyperspectral images of a hyperspectral observation satellite according to the present invention;

[0059] Figure 2 This is a schematic diagram of hyperspectral image preprocessing of a hyperspectral observation satellite of the present invention;

[0060] Figure 3 It is a schematic diagram of extracting point and surface spectral information of remote sensing images according to the present invention;

[0061] Figure 4 A schematic diagram of screening spectral information using the improved K-means clustering algorithm of the present invention;

[0062] Figure 5 It is a schematic diagram of the construction and evaluation of the alpine grassland quality inversion model of the present invention;

[0063] Figure 6 This is a schematic diagram of the spatial distribution analysis of the alpine grassland quality indicators of the present invention. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] The embodiment of the present invention provides a method for inverting the quality of alpine grassland based on hyperspectral observation satellite remote sensing images, comprising the following steps:

[0066] Step 1: Purchase hyperspectral images from a hyperspectral observation satellite (GF5B) with cloud coverage less than 30% based on the proposed study area. Figure 1 The detailed implementation steps are described as follows:

[0067] Step 1.1, use ArcMap vectorization to obtain the vector range of the study; Step 1.1 specifically includes:

[0068] Step 1.1.1, specify the spatial coordinate system and create vector surface features of the study area;

[0069] Step 1.1.2: according to the display of the study area in the GF5B image, vectorize along the boundary of the study area, complete the creation of surface elements, and derive the vector boundary of the proposed study area;

[0070] Step 1.2, on the Land Observation Satellite Data Service Platform, select the hyperspectral payload of GF5B, set the cloud coverage to less than 30%, and purchase appropriate satellite images based on vector boundaries and time;

[0071] Step 2: Perform radiation correction, atmospheric correction and other pre-processing on the purchased GF5B to convert the original observation value into surface reflectivity and eliminate the influence of the atmosphere on the satellite data. Figure 2 The detailed implementation steps are described as follows:

[0072] Step 2.1, using the ENVI 5.6 radiometric correction tool, 330 bands of visible light and short-wave infrared of the GF5B hyperspectral image were radiometrically calibrated to obtain the surface reflectance;

[0073] Step 2.2, use ENVI 5.6 geometric correction tool to call the hdr header file information of GF5B, set parameters to obtain the geometrically corrected GF5B image;

[0074] Step 3: Extract the precise image of the study area and the spectrum of the region of interest (ROI). According to the precise vector range of the study area, use the mask tool (Extract by Mask) in the ArcMap toolbox to clip and obtain the precise range of the study area; randomly select no less than 50 ROIs on the clipped image, and each ROI has no less than 3 pixels; use the "Multi-value Extract to Point" tool in the ArcMap spatial analysis toolbox to extract the spectral information of limited measured points. Figure 3 The detailed implementation steps are described as follows:

[0075] Step 3.1, import the corrected image into ArcMap, use the mask tool to extract the exact range of the study area, and obtain the cropped image;

[0076] Step 3.2, manually select no less than 50 ROIs from the cropped image in ENVI5.6, and export the vector range and spectrum of the ROI for later use;

[0077] Step 3.3, extract the longitude and latitude of the center point of each ROI vector range in ArcMap as the measured point coordinates of the ROI, and use the "Multi-value Extraction to Point" tool to export the spectral information of each measured point for future use.

[0078] Step 4: Spectral feature selection: Use an improved K-means clustering algorithm to reduce the dimension of the spectral data of the ROI, automatically screen the spectral band range that can effectively characterize the characteristics of the study area; use these band ranges to extract the spectral information corresponding to the measured points. Figure 4 The detailed implementation steps are described as follows:

[0079] Step 4.1, standardization, scaling the spectral reflectance data of all bands in the ROI area to a similar range to avoid the adverse effects of differences between features on the clustering results;

[0080] Step 4.2: Randomly select a feature subset of the spectrum. Randomly select an initial feature subset (k i ) as the starting point, and the remaining bands are subset k j ;

[0081] Step 4.3, iterative optimization, starting from k j Add or delete some bands into k i Get a new feature subset Implement feature subset screening; step 4.3 specifically includes:

[0082] Step 4.3.1, feature subset evaluation, for the currently selected feature subset Apply the improved K-means clustering algorithm to calculate the evaluation index of clustering results - the overall intra-cluster sum of squares;

[0083] Step 4.4.2, feature subset update, calling the genetic algorithm, adding or removing a spectral band to update the feature subset, exploring a wider range of combinations, and evaluating the performance of each feature subset through the quality indicator overall cluster sum of squares;

[0084] Step 4.4, spectral band selection convergence condition, compare the overall intra-cluster sum of squares of performance indicators and determine whether the number of iterations reaches the predetermined maximum number of iterations (1000 times), and stop the iteration;

[0085] Step 4.5, analysis of spectral screening results, analysis of the spectral band range and number of bands of the optimal feature subset, determination of the spectral band range that can effectively characterize the characteristics of the study area based on the correlation between the band range and vegetation (grassland) parameters, and use of the determined spectral band range to extract the surface reflectance of the corresponding band of the measured point.

[0086] The spectral bands for screening the final quality indicators (crude protein, crude fiber content, total digestible matter, and total digestible protein) are shown in Table 1.

[0087] Table 1 The optimal feature subsets screened by the improved K-means clustering algorithm

[0088]

[0089]

[0090] R q Indicates the spectral reflectance in the band qnm

[0091] Step 5: The inversion models of alpine grassland quality indicators (crude protein, crude fiber content, total digestible matter, total digestible protein) were constructed separately. The measured data were divided into a training set and a validation set. Combined with the feature selection results, the selected bands were substituted into the partial least squares regression algorithm on the training set to build the model, and the estimated values ​​were obtained on the validation set. Figure 5 The detailed implementation steps are described as follows:

[0092] Step 5.1, divide the measured grassland quality indicators into training set and validation set in a ratio of 8:2;

[0093] Step 5.2, on the training set, use the selected bands to train the models for inverting crude protein (g / kg), crude fiber content (%), total digestible matter (kg / 100kg), and total digestible protein (kg / 100kg);

[0094] Step 5.3, on the validation set, based on the training model, the estimated values ​​of the forage quality indicators crude protein, crude fiber content, total digestible matter, and total digestible protein are obtained respectively;

[0095] Step 6: Model evaluation and optimal inversion model determination, calculate the root mean square error RMSE and determination coefficient R of the measured and estimated values ​​of crude protein, crude fiber content, total digestible matter, and total digestible protein on the validation set 2 , R 2 The model with the highest value and low RMSE is determined as the optimal model for each quality index. Figure 5 The detailed implementation steps are described as follows:

[0096] The optimal inversion model obtained by each quality index is:

[0097] y 粗蛋白 =10.37R 760 +38.12 (1)

[0098] y 粗纤维含量 =-0.16R 695 +0.27 (2)

[0099] y 可消化物质总量 =21.51R 615 +20.31 (3)

[0100] y 可消化蛋白总量 =3.78R 705 +4.36 (4)

[0101] Step 6.1: Calculate the R of the estimated and measured values ​​of the quality indicators of alpine grassland, crude protein, crude fiber content, total digestible matter, and total digestible protein on the validation set. 2 and RMSE;

[0102] Step 6.2: Compare the R of each model 2 and RMSE, select R 2 The model with the highest and lower RMSE is taken as the optimal model for each quality index;

[0103] Step 7: Overall inversion of the quality indicators of alpine grassland in the study area. Use the spectral band range selected in step 5 to screen the spectral information of the entire study area. According to the optimal inversion model of each quality indicator, the forage quality parameters of the entire study area are inverted. The spatial distribution of each quality indicator of alpine grassland is comprehensively analyzed to reveal the growth status of alpine grassland and provide suggestions for the reasonable planning of stocking capacity. Figure 6 The detailed implementation steps are described as follows:

[0104] Step 7.1, reduce the spectrum of the entire study area according to the band range selected in step 4, and only retain the spectrum of the optimal subset;

[0105] Step 7.2, according to the optimal inversion model of each quality index determined in step 6, based on the spectral information of the optimal subset of the entire study area, invert the quality index of the entire study area;

[0106] Step 7.3: Analyze the growth status of alpine grassland forage based on the forage quality indicators of the entire study area and combined with geographical location information.

[0107] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for inverting alpine grassland quality based on hyperspectral observation satellite remote sensing images, comprising the following steps: Step 1: Purchase GF5B hyperspectral images of the hyperspectral observation satellite with cloud coverage less than 30% according to the scope of the proposed study area; Step 2: Preprocess the purchased GF5B hyperspectral image, convert the original data of the GF5B hyperspectral remote sensing image into the spectral reflectance of each band, and eliminate the influence of the atmosphere on the satellite data. The preprocessing includes radiation correction and geometric correction; Step 3: Extract accurate images of the study area and extract spectra of the ROI: Use the mask tool in the ArcMap toolbox to clip and obtain the exact range of the study area to obtain a clipped image; randomly select no less than 50 ROIs on the clipped image, each with no less than 3 pixels, and use the "Multi-value Extraction to Point" tool in the ArcMap Spatial Analysis Toolbox to extract the spectral information of limited measured points to obtain the spectral data of the ROI; Step 4: Use an integrated genetic algorithm and an improved K-means clustering algorithm with repeated iterations to reduce the dimension of the spectral data of the ROI, automatically screen the spectral band range that can effectively characterize the characteristics of the study area, and use the spectral band range to extract the surface reflectance of the corresponding band of the measured point; Step 5: construct inversion models of alpine grassland quality indicators respectively: the alpine grassland quality indicators include crude protein, crude fiber content, total digestible matter, and total digestible protein. The measured data are divided into a training set and a validation set. Combined with the feature selection results, each band selected on the training set is substituted into the partial least squares regression algorithm one by one to establish the model, and the estimated value is obtained on the validation set; Step 6: Model evaluation and determination of the optimal inversion model: Calculate the root mean square error (RMSE) and determination coefficient (R) of the measured and estimated values ​​of the alpine grassland quality indicators on the validation set. 2 , according to the root mean square error RMSE and the coefficient of determination R 2 Determine the optimal inversion model for each alpine grassland quality index; Step 7: Overall inversion of alpine grassland quality indicators in the study area: Use the spectral band range screened in step 4 to screen the spectral information of the entire study area. According to the optimal inversion model of each alpine grassland quality indicator, the forage quality parameters of the entire study area are inverted. The spatial distribution of each quality indicator of the alpine grassland is comprehensively analyzed to reveal the growth status of the alpine grassland and provide suggestions for the reasonable planning of livestock carrying capacity.

2. The method for inverting alpine grassland quality based on hyperspectral observation satellite remote sensing images according to claim 1 is characterized in that: The step 1 comprises: Step 1.1, use ArcMap vectorization to obtain the vector boundary of the proposed study area; Step 1.2: On the Land Observation Satellite Data Service Platform, select the hyperspectral payload of GF5B, set the cloud coverage to less than 30%, and purchase appropriate satellite images based on vector boundaries and time.

3. According to the method for inverting alpine grassland quality based on hyperspectral observation satellite remote sensing images according to claim 2, the step 1.1 comprises: Step 1.1.1, specify the spatial coordinate system and create the surface element extracted by the vector range; Step 1.1.2: According to the display of the study area in the GF5B image, vectorize along the boundary of the study area, complete the creation of surface elements, and derive the vector boundary of the proposed study area.

4. The method for inverting alpine grassland quality based on hyperspectral observation satellite remote sensing images according to claim 1 is characterized in that: The step 2 comprises: Step 2.1, using the ENVI 5.6 radiometric correction tool, 330 bands of visible light and short-wave infrared of the GF5B hyperspectral image were radiometrically calibrated to obtain the surface reflectance; Step 2.2: Use the ENVI 5.6 geometric correction tool to call the HDR header file information of GF5B and set the parameters to obtain the geometrically corrected GF5B image.

5. The method for inverting alpine grassland quality based on hyperspectral observation satellite remote sensing images according to claim 1 is characterized in that: The step 3 comprises: Step 3.1, import the corrected image into ArcMap, use the mask tool to extract the exact range of the study area, and obtain the cropped image; Step 3.2, manually select no less than 50 ROIs from the cropped image in ENVI5.6, and export the vector range and spectrum of the ROI for later use; Step 3.3, extract the longitude and latitude of the center point of each ROI vector range in ArcMap as the measured point coordinates of the ROI, and use the "Multi-value Extraction to Point" tool to export the spectral information of each measured point for future use.

6. The method for inverting alpine grassland quality based on hyperspectral observation satellite remote sensing images according to claim 1 is characterized in that: The step 4 comprises: Step 4.1, standardization, scaling the ROI spectral data to a similar range to avoid the adverse effects of differences between features on the clustering results; Step 4.2, randomly select a feature subset of the spectrum, and randomly select an initial feature subset from all spectral bands as the starting point; Step 4.3, iterative optimization, adding or deleting certain bands to achieve feature subset screening; Step 4.4, spectral band selection convergence condition, when the predetermined maximum number of iterations is reached or the clustering performance is no longer significantly improved, the iteration is stopped; Step 4.5, analysis of spectral screening results, analysis of the spectral band range and number of bands of the optimal feature subset, determination of the spectral band range that can effectively characterize the characteristics of the study area based on the correlation between the band range and vegetation parameters, and use of the determined spectral band range to extract the surface reflectance of the corresponding band of the measured point.

7. The method for inverting alpine grassland quality based on hyperspectral observation satellite remote sensing images according to claim 6 is characterized in that: The step 4.3 comprises: Step 4.3.1, feature subset evaluation, applying the K-means clustering algorithm to the currently selected feature subset, and calculating the evaluation index of the clustering result, wherein the evaluation index includes the overall intra-cluster sum of squares; Step 4.3.2, feature subset update, calling the genetic algorithm, adding or removing a spectral band to update the feature subset, exploring a wider range of combinations, and evaluating the quality of each combination through the evaluation function; Step 4.3.3, evaluation function, evaluates the performance of each feature subset based on the overall within-cluster sum of squares, a quality indicator of K-means clustering.

8. The method for inverting alpine grassland quality based on hyperspectral observation satellite remote sensing images according to claim 1 is characterized in that: The step 5 comprises: Step 5.1, divide the measured grassland quality indicators into training set and validation set in a ratio of 8:2; Step 5.2, on the training set, each selected band is substituted into the partial least squares regression algorithm one by one to establish the inversion model of crude protein, crude fiber content, total digestible matter, and total digestible protein; Step 5.3, on the validation set, based on the training model, the estimated values ​​of the forage quality indicators crude protein, crude fiber content, total digestible matter, and total digestible protein are obtained.

9. The method for inverting alpine grassland quality based on hyperspectral observation satellite remote sensing images according to claim 1 is characterized in that: The step 6 comprises: Step 6.1: Calculate the R of the estimated and measured values ​​of the quality indicators of alpine grassland, crude protein, crude fiber content, total digestible matter, and total digestible protein on the validation set. 2 and RMSE; Step 6.2: Compare the R of each model 2 and RMSE, select R 2 The model with the highest and lower RMSE is taken as the optimal model for each quality index.

10. The method for inverting alpine grassland quality based on hyperspectral observation satellite remote sensing images according to claim 1, characterized in that: The step 7 comprises: Step 7.1, reduce the dimension of the spectrum of the entire study area according to the band range screened in step 4, and only retain the spectrum of the optimal subset; Step 7.2, according to the optimal inversion model of each quality index determined in step 6, based on the spectral information of the optimal subset of the entire study area, invert the quality index of the entire study area; Step 7.3: Analyze the growth status of alpine grassland forage based on the forage quality indicators of the entire study area and combined with geographical location information.

Images