Forest grass vegetation coverage refined inversion method and system based on multi-scale remote sensing

Abstract

The invention discloses a forest grass vegetation coverage refined inversion method and system based on multi-scale remote sensing, and the method comprises the steps: firstly providing end member types and end member abundance information through the field observation data of an unmanned plane, inputting the prior knowledge into a linear mixed pixel decomposition model, and carrying out the back calculation to obtain end member spectrum information of the corresponding end member types; by taking unmanned aerial vehicle over-resolution observation data as priori knowledge, the end member spectrum extraction precision is improved, and a technical support is provided for improving the inversion precision of the total coverage of forest and grass vegetation; the textural features of high-resolution satellite observation data are fully utilized, the spectral features of medium resolution and the textural features of high resolution are combined, and the optimal response relation between different remote sensing feature combinations and woody vegetation coverage measured data is constructed by utilizing a random forest model. Therefore, high-precision inversion of the woody vegetation coverage in a large-area range is realized, and data support is provided for further estimation of the herbaceous vegetation coverage.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image processing, and in particular to a method and system for fine-grained inversion of forest and grass vegetation coverage based on multi-scale remote sensing. Background Art

[0002] Vegetation coverage refers to the ratio of the area of ​​vegetation vertically projected onto the ground to the total area per unit area. Vegetation coverage is an important basic data and key parameter in the study of surface processes such as eco-hydrology and soil erosion. Therefore, the accuracy and quality of vegetation coverage data have always been the focus and attention of many experts and scholars. Traditional vegetation distribution surveys are mainly based on ground observations. This method has certain limitations, and the cost of manpower and material resources is relatively large. There is also a large uncertainty in the extrapolation based on field measured samples. With the continuous development of remote sensing earth observation technology, remote sensing inversion of vegetation coverage in large areas has become possible. Vegetation coverage inversion research based on remote sensing data has always been a hot topic in the field of quantitative remote sensing of vegetation.

[0003] At present, the vegetation coverage inversion methods based on remote sensing technology mainly include: empirical model, physical model and mixed pixel decomposition model. The empirical model mainly inverts the regional vegetation coverage by establishing the statistical relationship between remote sensing characteristics such as vegetation index or band reflectivity and measured vegetation coverage; the physical model mainly uses the vegetation radiation transmission mechanism model to quantitatively invert the vegetation canopy leaf area index from the remote sensing observation surface reflectance signal, and then uses the canopy porosity formula to convert the leaf area index into canopy vegetation coverage; the mixed pixel decomposition method is also widely used in the estimation of vegetation coverage in semi-arid areas. This method uses the pixel decomposition model to decompose the brightness information contribution of each end member in the pixel, and obtains the end member vegetation abundance to estimate the coverage. In arid / semi-arid areas, sparse woody vegetation, herbaceous vegetation and dry vegetation are the main components of forest and grassland vegetation. However, the vegetation cover inversion algorithm that has been widely used cannot finely identify the coverage of woody vegetation, herbaceous vegetation and dry vegetation, which seriously affects the inversion accuracy of the total coverage of forest and grassland vegetation.

[0004] In addition, the algorithm of existing MODIS and Landsat woody vegetation coverage (Tree Cover) remote sensing products is to use machine learning algorithms to directly construct a relationship model between the measured sample plots of woody vegetation coverage and the medium and low resolution remote sensing spectral characteristics, and apply it to large areas. However, the application accuracy of this inversion method is low in mixed scenarios of sparse woody vegetation, herbaceous vegetation and dry vegetation in arid and semi-arid areas, because the spectral characteristics of herbaceous vegetation and woody vegetation are very similar, and the spectral characteristics of dry vegetation and bare land are very similar, which seriously affects the inversion accuracy. By using the measured woody vegetation coverage data in semi-arid areas to verify the published Tree Cover products, it was found that the existing Tree Cover products seriously underestimated the woody vegetation coverage in semi-arid areas and could not meet the application requirements. At present, there are few studies on the remote sensing inversion of fine coverage of forest and grass vegetation under mixed surface cover types and complex terrain. The few studies that have been reported mainly use the "two-step method" to separate woody vegetation and herbaceous vegetation, which first identifies the woody vegetation canopy based on high-resolution remote sensing images and then calculates the coverage. However, the inversion accuracy of this method is affected by the classification accuracy of the woody vegetation canopy, and it is not suitable for large-scale regional expansion applications. A mixed pixel decomposition model is used to separate green vegetation, dry vegetation, and bare land, but the inversion accuracy of this method is affected by the end-member spectral accuracy of green vegetation, dry vegetation, and bare land. Therefore, it is urgent to propose a practical remote sensing inversion method for fine coverage of forest and grass vegetation in large areas.

[0005] In summary, the current vegetation coverage inversion methods based on remote sensing technology have the following main defects:

[0006] (1) The empirical model method refers to directly establishing a response relationship model between the vegetation index or band reflectance and the measured vegetation coverage, and using this model to extrapolate the vegetation coverage of a large area. The mathematical method for constructing the response relationship generally uses statistical regression and machine learning algorithms. However, when this method is applied to invert woody vegetation coverage, the medium and low resolution remote sensing observation spectral signals include mixed reflection signals of woody vegetation, herbaceous vegetation, dry vegetation and bare land, and it is impossible to obtain the refined coverage of different components in forest and grass vegetation, and the empirical model parameters cannot be promoted for use.

[0007] (2) The physical model method also quantitatively inverts the canopy leaf area index of vegetation through spectral reflectance, but the spectral signals of medium and low resolution remote sensing observations include the reflectance of woody vegetation and herbaceous vegetation. Therefore, the total leaf area index of forest and grass vegetation can only be quantitatively inverted through the vegetation radiation transmission model, and the leaf area index of woody vegetation and herbaceous vegetation cannot be separated. Therefore, it is also impossible to separate the coverage of woody vegetation and herbaceous vegetation through the canopy porosity formula. Therefore, this method can only invert the total coverage of forest and grass vegetation, and does not consider the coverage of dry vegetation.

[0008] (3) The principle of the mixed pixel decomposition model method is that a pixel in an image may actually be composed of multiple components, and each component contributes to the information observed by the remote sensing sensor. Therefore, the remote sensing information can be decomposed to establish a pixel decomposition model, and this model can be used to estimate vegetation coverage. The pixel binary model is the simplest model in the linear pixel decomposition model method. The pixel binary model assumes that the pixel is composed of only two parts, the vegetation-covered surface and the non-vegetation-covered surface. The mixed spectral information is linearly synthesized by only these two component factors. The ratio of their respective areas in the pixel is the weight of each factor. The percentage of the vegetation-covered surface in the pixel is the vegetation coverage of the pixel. The formula is expressed as:

[0009] F cover =(NDVI-NDVI min ) / (NDVI max -NDVI min )

[0010] In the formula, NDVI min is the minimum NDVI value of the surface without vegetation coverage, NDVI max is the maximum NDVI value of the surface covered by vegetation.

[0011] However, in areas where woody vegetation and herbaceous vegetation are mixed, the spectral reflectance characteristics of herbaceous and woody vegetation in the visible and infrared bands are similar, and the NDVI values ​​are very close. This method cannot accurately separate woody vegetation and herbaceous vegetation, and does not consider the impact of dry vegetation on the total coverage of forest and grass vegetation. Although some mixed pixel decomposition models consider dry vegetation and green vegetation endmembers, in highly heterogeneous areas where multiple vegetation cover types are mixed and vegetation site conditions are complex, the same vegetation type may have multiple endmember spectra. It is difficult to directly find high-quality endmember spectra of different vegetation types on medium and low resolution remote sensing images, which cannot support the mixed pixel decomposition model to estimate the abundance of green vegetation, dry vegetation and soil components with high accuracy, and it is impossible to decompose the coverage of woody and herbaceous vegetation. Therefore, there is an urgent need for a method to automatically obtain multi-endmember spectra of green vegetation, dry vegetation and bare soil to improve the accuracy of endmember spectrum extraction, thereby improving the accuracy level of mixed pixel decomposition model solution. Summary of the invention

[0012] In order to solve the problem that the existing methods have low coverage of green vegetation and dry vegetation in inversion, and cannot separate the coverage of woody vegetation and herbaceous vegetation in green vegetation, the present invention proposes a method and system for fine inversion of forest and grass vegetation coverage based on multi-scale remote sensing, and uses the idea of ​​hierarchical classification inversion to invert the coverage of woody vegetation, herbaceous vegetation and dry vegetation layer by layer. It can be applied to the fine inversion of forest and grass vegetation coverage (woody vegetation coverage, herbaceous coverage, dry vegetation coverage) in arid / semi-arid areas, and can provide new technical means for high-precision monitoring of vegetation dynamic processes, and can also provide high-precision underlying surface data support for vegetation eco-hydrological benefit evaluation based on surface water and soil process models such as ecological models, hydrological models and soil erosion models.

[0013] In order to achieve the above object, the present invention adopts the following technical solutions:

[0014] On the one hand, the present invention proposes a method for refining forest and grass vegetation coverage based on multi-scale remote sensing, comprising:

[0015] Step 1: Obtain ground sample plot drone images and classify the acquired drone images into green vegetation, dry vegetation, and soil;

[0016] Step 2: The average surface reflectance of the Landsat pixel corresponding to the coordinates of the ground sample plot range is used as the mixed pixel observation signal value of the Landsat sample plot; green vegetation, dry vegetation and soil are used as the end member types in the pixel; based on the classification results of the drone image of each sample plot, the coverage value of each end member type is obtained, and the coverage value is used as the abundance prior knowledge of the end member type;

[0017] Step 3: Based on the prior knowledge of the abundance of endmember types, the endmember spectra are reversely calculated using the linear mixed pixel decomposition model to obtain the endmember spectral library;

[0018] Step 4: Based on the acquired endmember spectral library data, the abundance of green vegetation, dry vegetation and soil is forward inverted using a multi-endmember mixed pixel decomposition model to obtain the abundance of green vegetation, dry vegetation and soil at a regional scale, and then the dry vegetation coverage is obtained. The sum of the abundances of green vegetation and dry vegetation is calculated to obtain the forest and grass vegetation coverage;

[0019] Step 5: Use human-computer interactive interpretation to obtain the woody vegetation in each sample plot UAV image and calculate the woody vegetation coverage, thereby constructing a sample dataset of woody vegetation coverage;

[0020] Step 6: Extract spectral and texture features of Landsat and high-resolution satellite image data;

[0021] Step 7: Take the woody vegetation coverage sample data set as the target variable, the extracted spectral features, texture features and the forest and grass vegetation coverage obtained in step 4 as feature variables, and use the random forest machine learning model to construct the nonlinear relationship between woody vegetation coverage and spectral features, texture features and forest and grass vegetation coverage, and predict the woody vegetation coverage;

[0022] Step 8: Based on the physical shielding relationship between forest and grass vegetation coverage, woody vegetation coverage, and herbaceous vegetation coverage, the herbaceous vegetation coverage is calculated in reverse.

[0023] Furthermore, in step 1, the acquired drone images are classified in the following manner:

[0024] Step 1.1: Select green vegetation, dry vegetation and soil samples based on field survey photos and drone images;

[0025] Step 1.2: Use the color space transformation function to convert the RGB color value of the drone image to obtain four color spaces: RGB, HSV, XYZ, and L * a * b * The color characteristics of 12 dimensions;

[0026] Step 1.3: Use the support vector machine supervised classification method to automatically classify the color features after RGB transformation, and combine it with the visual interpretation method to obtain the classification results of drone images of green vegetation, dry vegetation and soil.

[0027] Furthermore, the calculation formula of the linear mixed pixel decomposition model is:

[0028] x k =M T f k +ε

[0029] Among them, x k is a b×1 vector, representing the surface reflectance of the kth sample plot, where b is the number of bands; f k is a c×1 vector, representing the endmember coverage in the kth plot, c is the number of endmember categories; M is a c×b matrix, representing the endmember spectral matrix, and each row of array elements represents the spectrum of a specific endmember type; ε represents the model fitting error.

[0030] Furthermore, in step 3, for n sample plots, the linear mixed pixel decomposition model is modified as follows:

[0031] X=FM

[0032] Where X is an n×b matrix, X=[x1,x2,…,x k ,…,x n ]T ; F is an n×c matrix, F=[f1,f2,…fk k ,…f n ] T ;

[0033] The end-member spectrum matrix M of the study area is obtained by using the least squares method:

[0034] M=(F T F) -1 F T X.

[0035] Furthermore, the calculation formula of the multi-end-member mixed pixel decomposition model is:

[0036]

[0037] Where X is the actual observed value of the image pixel; N is the number of endmember categories; d i is the number of endmember spectra of the i-th endmember type; e ij is the jth endmember spectrum of the i-th type endmember; p ij is the proportion of the jth spectrum of the i-th endmember in the pixel; q is 0 or 1, representing the label of whether the endmember spectrum participates in the unmixing; n is the error component; 0≤p ij ≤1 and

[0038] Furthermore, in step 4, the multi-terminal mixed pixel decomposition model with the smallest root mean square error of the pixel spectral simulation value is selected as the optimal model for spectral unmixing, and the p calculated by the optimal model for spectral unmixing is ij That is, the abundance value of each end member type in the pixel. The abundance of green vegetation, dry vegetation and soil at the regional scale is calculated using the optimal model of spectral unmixing. The sum of the abundance of green vegetation and dry vegetation is the forest and grass vegetation coverage.

[0039] Furthermore, the spectral characteristics include surface reflectance of all bands, normalized vegetation index, visible light band difference vegetation index, normalized difference moisture index, ratio vegetation index, enhanced vegetation index, improved soil adjusted vegetation index, difference vegetation index, green light normalized vegetation index, green band index and near infrared vegetation index.

[0040] Furthermore, in step 6, a gray level co-occurrence matrix based on statistical analysis is used to calculate the texture features of the surface reflectance of all bands.

[0041] Furthermore, in step 8, the herbaceous vegetation coverage is obtained by reverse calculation according to the following formula:

[0042] f total =f woody +(1-fwoody )×f herb

[0043] Among them, f total is the forest and grass vegetation coverage, f woody and f herb are the coverage of woody vegetation and herbaceous vegetation, respectively.

[0044] Another aspect of the present invention provides a multi-scale remote sensing-based forest and grass vegetation coverage refinement inversion system, comprising:

[0045] The ground sample plot UAV image acquisition and classification module is used to acquire ground sample plot UAV images and classify the acquired UAV images into green vegetation, dry vegetation and soil;

[0046] The module for obtaining the coverage value of end-member types is used to take the average surface reflectance of the Landsat pixel corresponding to the coordinates of the ground sample plot range as the mixed pixel observation signal value of the Landsat sample plot; take green vegetation, dry vegetation and soil as the end-member types in the pixel; based on the classification results of the drone image of each sample plot, obtain the coverage value of each end-member type, and use the coverage value as the abundance prior knowledge of the end-member type;

[0047] The endmember spectrum library back-calculation module is used to reversely calculate the endmember spectrum based on the prior knowledge of the endmember type abundance and use the linear mixed pixel decomposition model to obtain the endmember spectrum library;

[0048] The multi-endmember mixed pixel decomposition module is used to perform forward inversion on the abundance of green vegetation, dry vegetation and soil based on the acquired endmember spectral library data using the multi-endmember mixed pixel decomposition model, obtain the abundance of green vegetation, dry vegetation and soil at the regional scale, and then obtain the dry vegetation coverage, and calculate the sum of the abundance of green vegetation and dry vegetation to obtain the forest and grass vegetation coverage;

[0049] The module for constructing a sample dataset of woody vegetation coverage is used to obtain the woody vegetation in each plot drone image using a human-computer interactive interpretation method and calculate the woody vegetation coverage, thereby constructing a sample dataset of woody vegetation coverage;

[0050] Feature variable set building module, used to extract spectral and texture features of Landsat and high-resolution satellite image data;

[0051] The woody vegetation coverage prediction module is used to take the woody vegetation coverage sample data set as the target variable, and the extracted spectral features, texture features and forest and grass vegetation coverage obtained by the multi-end-member mixed pixel decomposition module as the feature variable, and use the random forest machine learning model to construct the nonlinear relationship between woody vegetation coverage and spectral features, texture features and forest and grass vegetation coverage, and predict the woody vegetation coverage;

[0052] The herbaceous vegetation coverage prediction module is used to reversely calculate the herbaceous vegetation coverage based on the physical shielding relationship between the forest and grass vegetation coverage and the woody vegetation coverage and the herbaceous vegetation coverage.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] The existing mixed pixel decomposition mainly extracts endmember spectra directly from remote sensing images, but this method is difficult to effectively find endmember spectra that meet the 100% abundance endmember type in arid / semi-arid savanna areas, which will affect the analytical accuracy of the mixed pixel decomposition model. The present invention uses the field observation data of unmanned aerial vehicles to provide prior knowledge of endmember types and endmember abundance, and uses a multi-endmember mixed pixel spectral decomposition model to inversely calculate the theoretical value of the endmember spectrum, which can improve the accuracy of extracting endmember spectra.

[0055] The existing woody vegetation coverage inversion mainly uses medium and low resolution satellite remote sensing spectral features to invert woody vegetation coverage, but the inversion effect of this method in arid / semi-arid savanna areas is poor. This is because the spectral features of herbaceous vegetation are very similar to those of woody vegetation, and herbaceous vegetation will seriously affect the inversion accuracy of woody vegetation coverage. The present invention makes full use of the texture features of high-resolution satellite (remote sensing) image data to reflect the difference between woody vegetation and herbaceous vegetation, combines medium-resolution spectral features with high-resolution texture features, and uses a random forest model to construct the best response relationship between different remote sensing feature combinations and woody vegetation coverage measured data, which can improve the inversion accuracy of woody vegetation coverage and provide data support for herbaceous vegetation coverage estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a flow chart of a method for fine-grained inversion of forest and grass vegetation coverage based on multi-scale remote sensing according to an embodiment of the present invention;

[0057] Figure 2 A schematic diagram of the composition structure of the end member spectral library provided in an embodiment of the present invention;

[0058] Figure 3 A multi-terminal mixed pixel decomposition flow chart provided by an embodiment of the present invention;

[0059] Figure 4The present invention is a schematic diagram of the architecture of a system for refining forest and grass vegetation coverage based on multi-scale remote sensing according to an embodiment of the present invention. DETAILED DESCRIPTION

[0060] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments:

[0061] like Figure 1 As shown in the figure, a method for refining forest and grass vegetation coverage based on multi-scale remote sensing includes:

[0062] S101: Acquire drone images of ground sample plots, and classify the acquired drone images into green vegetation, dry vegetation, and soil;

[0063] S102: taking the average value of the surface reflectance of the Landsat pixel corresponding to the coordinates of the ground sample plot range as the mixed pixel observation signal value of the Landsat of the sample plot; taking green vegetation, dry vegetation and soil as the end member types in the pixel; obtaining the coverage value of each end member type based on the classification result of the drone image of each sample plot, and taking the coverage value as the abundance prior knowledge of the end member type;

[0064] S103: Based on the prior knowledge of the abundance of endmember types, the endmember spectra are reversely calculated using a linear mixed pixel decomposition model to obtain an endmember spectrum library;

[0065] S104: Based on the acquired endmember spectral library data, the abundance of green vegetation, dry vegetation and soil is forward inverted using a multi-endmember mixed pixel decomposition model to obtain the abundance of green vegetation, dry vegetation and soil at a regional scale, and then obtain the dry vegetation coverage, and calculate the sum of the abundances of green vegetation and dry vegetation to obtain the forest and grass vegetation coverage;

[0066] S105: Use human-computer interactive interpretation to obtain woody vegetation in each plot drone image and calculate the woody vegetation coverage, thereby constructing a sample dataset of woody vegetation coverage;

[0067] S106: Extract spectral and texture features of Landsat and high-resolution satellite image data;

[0068] S107: Taking the sample data set of woody vegetation coverage as the target variable, the extracted spectral features, texture features and the forest and grass vegetation coverage obtained in S104 as feature variables, and using the random forest machine learning model to construct the nonlinear relationship between woody vegetation coverage and spectral features, texture features and forest and grass vegetation coverage, the woody vegetation coverage is predicted;

[0069] S108: According to the physical shielding relationship between the forest and grass vegetation coverage, the woody vegetation coverage, and the herbaceous vegetation coverage, the herbaceous vegetation coverage is obtained by reverse calculation.

[0070] The method specifically includes:

[0071] 1. Inversion of total forest and grass vegetation coverage

[0072] (1) Acquisition and classification of ground sample plot drone images

[0073] Conduct field sample surveys in the target area, using 30m×30m sample plots, and use differential RTK GPS to accurately locate the sample center and four corners. Use drones to take vertical photos of each sample plot, ensuring that the overlap of each photo is greater than 60%, and the shooting range can completely cover the 30m×30m ground sample plot, and use professional software to stitch and orthorectify drone pictures. Ensure that the spatial distribution of woody vegetation, herbaceous vegetation, dry vegetation and soil can be clearly distinguished in the drone photos.

[0074] The present invention designs an automatic classification and processing flow of UAV images: ① based on field survey photos and UAV images, green vegetation, dry vegetation and soil samples are selected; ② the RGB color value of the UAV image is converted using the color space change function to obtain four color spaces (RGB, HSV, XYZ and L * a * b * )’s 12-dimensional color features, and the calculation formulas are shown in (1) to (5); ③ The color features after RGB transformation are automatically classified using the support vector machine supervised classification method, and the high-precision UAV image classification results of green vegetation, dry vegetation and soil are obtained in combination with the visual interpretation method.

[0075] The calculation formula for converting RGB to HSV is as follows:

[0076]

[0077] Where:

[0078]

[0079] The calculation formula to convert RGB to XYZ is as follows:

[0080]

[0081] Convert RGB to L * a * b * The calculation formula is as follows:

[0082]

[0083] Where:

[0084]

[0085] Among them, X n , Y n and Z n is the value of a specific white object point.

[0086] (2) Back calculation of end member spectral library

[0087] The average surface reflectance of the 2×2 area adjacent to the Landsat pixel corresponding to the sample plot range coordinates is taken as the mixed pixel observation signal value of the Landsat remote sensing satellite of the sample plot; green vegetation, dry vegetation and soil are taken as the end-member types in the pixel; based on the classification results of the drone image of each sample plot (including green vegetation, dry vegetation and soil), the total number of pixels of a certain type can be divided by the total number of all pixels in the sample plot to calculate the coverage value of the type, and this coverage value is used as the abundance prior knowledge of the end-member type.

[0088] The linear mixed pixel decomposition model (LSMA) is used to reversely calculate the end member spectrum, and the calculation formula is as follows:

[0089] x k =M T f k +ε(6)

[0090] In the formula, x k is a b×1 vector, representing the surface reflectance of the kth sample plot, where b is the number of bands; f k is a c×1 vector, representing the endmember coverage in the kth sample plot, c is the number of endmember categories; M is a c×b matrix, each row of array elements represents the spectrum of a specific endmember type, and the M matrix is ​​called the endmember spectrum matrix; ε represents the model fitting error.

[0091] For n samples, the linear mixed pixel decomposition model can be modified as follows:

[0092] X=FM (7)

[0093] Where X is an n×b matrix, X=[x1,x2,…,x k ,…,x n ] T ; F is an n×c matrix, F=[f1,f2,…f k ,…f n ] T .

[0094] The endmember spectrum matrix M of the study area can be obtained by using the least squares method:

[0095] M=(FT F) -1 F T X(8)

[0096] (3) Multi-end element mixed pixel decomposition

[0097] The endmember spectral library consists of class A, class B, and class C. The endmembers of class A have different spectra such as A1, A2, A3, etc. (same object, different spectra phenomenon), and the endmembers of class B have B1, etc., and so on. When forming an unmixing model, one or more or no spectra are selected from each endmember class as the endmember spectra of the model for unmixing. The endmember spectral library composition structure is as follows: Figure 2 .

[0098] Based on the endmember spectral library data obtained in the above steps, the abundance of green vegetation, dry vegetation and soil components is forward inverted using the multi-endmember mixed pixel decomposition model (MESMA). The calculation process is as follows: Figure 3 As shown. The MESMA calculation formula is as follows:

[0099]

[0100] Where X is the actual observed value of the image pixel; N is the number of endmember categories; d i is the number of endmember spectra of the i-th endmember type; e ij is the jth endmember spectrum of the i-th type endmember; p ij is the proportion of the jth spectrum of the i-th endmember in the pixel; q is 0 or 1, representing a label of whether this endmember spectrum participates in the unmixing; n is the error component. Like the linear mixed model, the multi-endmember mixed pixel decomposition model also has two constraints, 0≤p ij ≤1 and

[0101] The root mean square error (RMSE) of the spectral simulation value of the selected pixel during the calculation process s )The smallest MESMA model is used as the optimal model for spectral unmixing. The p ij It is the abundance value of each end member type in the pixel.

[0102]

[0103] Where m represents the number of spectral bands, ε b Represents the root mean square error of the simulated pixel values ​​in band b.

[0104] The model (i.e., the optimal spectral unmixing model) can be used to calculate the abundance of green vegetation, dry vegetation, and soil at the regional scale. The abundance of green vegetation represents the coverage of green woody and green herbaceous vegetation, and the abundance of dry vegetation represents the coverage of yellow herbaceous vegetation (dry woody vegetation does not produce coverage because it has no canopy), i.e., the coverage of dry vegetation. Therefore, the sum of the abundance of green vegetation and dry vegetation is the coverage of forest and grass vegetation.

[0105] 2. Fine inversion of woody vegetation and herbaceous vegetation coverage

[0106] (1) Construction of woody vegetation coverage sample dataset

[0107] Since the automatic classification results of green woody vegetation and green herbaceous vegetation have great uncertainty, in order to improve the accuracy of sample data, the present invention adopts a human-computer interactive interpretation method to obtain the woody vegetation canopy in each sample plot UAV image and calculates its woody vegetation coverage, thereby constructing a high-quality sample data set of woody vegetation coverage.

[0108] (2) Construction of feature variable set

[0109] Extract spectral and texture features of Landsat (visible light, near infrared, shortwave infrared) and high-resolution satellite image data (visible light, near infrared). Spectral features include surface reflectance in all bands, Normalized Difference Vegetation Index (NDVI), Visible-band Difference Vegetation Index (VDVI), Normalized Difference Moisture Index (NDMI), Ratio Vegetation Index (RVI), Enhanced Vegetation Index (EVI), Modified Soil-Adjusted Vegetation Index (MSAVI), Difference Vegetation Index (DVI), Green Normalized Difference Vegetation Index (GNDVI), Green-Red Vegetation Index (GRVI) and Near-Infrared Reflectance of The texture characteristics of the surface reflectance in all bands were calculated using the Gray-Level-Co-occurrence Matrix (GLCM) based on statistical analysis.

[0110] (3) Construction of woody vegetation coverage prediction model

[0111] The sample dataset of woody vegetation coverage was used as the target variable, and the spectral characteristics, texture characteristics of Landsat and high-resolution satellite image data corresponding to the sample area, and the inversion results of forest and grass vegetation coverage in 1 were used as feature variables. The random forest machine learning model was used to construct the complex nonlinear relationship between woody vegetation coverage and spectral characteristics, texture characteristics and forest and grass vegetation coverage, and a remote sensing inversion model of woody vegetation coverage at the regional scale was constructed. Then, recursive feature elimination (RFE) was used to select the optimal variables for all predictor variables, so as to determine the most robust and accurate machine learning prediction model for woody vegetation coverage.

[0112] (4) Construction of herbaceous vegetation coverage prediction model

[0113] According to the physical shielding relationship between the total vegetation coverage in the vertical projection direction (i.e., forest and grass vegetation coverage) and the woody vegetation coverage and herbaceous vegetation coverage, the herbaceous vegetation coverage can be inversely calculated using formula (11).

[0114] f total =f woody +(1-f woody )×f herb (11)

[0115] In the formula, f total is the forest and grass vegetation coverage, f woody and f herb are the coverage of woody vegetation and herbaceous vegetation, respectively.

[0116] Based on the above embodiments, Figure 4 As shown, the present invention also proposes a forest and grass vegetation coverage refinement inversion system based on multi-scale remote sensing, comprising:

[0117] The ground sample plot UAV image acquisition and classification module is used to acquire ground sample plot UAV images and classify the acquired UAV images into green vegetation, dry vegetation and soil;

[0118] The module for obtaining the coverage value of end-member types is used to take the average surface reflectance of the Landsat pixel corresponding to the coordinates of the ground sample plot range as the mixed pixel observation signal value of the Landsat sample plot; take green vegetation, dry vegetation and soil as the end-member types in the pixel; based on the classification results of the drone image of each sample plot, obtain the coverage value of each end-member type, and use the coverage value as the abundance prior knowledge of the end-member type;

[0119] The endmember spectrum library back-calculation module is used to reversely calculate the endmember spectrum based on the prior knowledge of the endmember type abundance and use the linear mixed pixel decomposition model to obtain the endmember spectrum library;

[0120] The multi-endmember mixed pixel decomposition module is used to perform forward inversion on the abundance of green vegetation, dry vegetation and soil based on the acquired endmember spectral library data using the multi-endmember mixed pixel decomposition model, obtain the abundance of green vegetation, dry vegetation and soil at the regional scale, and then obtain the dry vegetation coverage, and calculate the sum of the abundance of green vegetation and dry vegetation to obtain the forest and grass vegetation coverage;

[0121] The module for constructing a sample dataset of woody vegetation coverage is used to obtain the woody vegetation in each plot drone image using a human-computer interactive interpretation method and calculate the woody vegetation coverage, thereby constructing a sample dataset of woody vegetation coverage;

[0122] Feature variable set building module, used to extract spectral and texture features of Landsat and high-resolution satellite image data;

[0123] The woody vegetation coverage prediction module is used to take the woody vegetation coverage sample data set as the target variable, and the extracted spectral features, texture features and forest and grass vegetation coverage obtained by the multi-end-member mixed pixel decomposition module as the feature variable, and use the random forest machine learning model to construct the nonlinear relationship between woody vegetation coverage and spectral features, texture features and forest and grass vegetation coverage, and predict the woody vegetation coverage;

[0124] The herbaceous vegetation coverage prediction module is used to reversely calculate the herbaceous vegetation coverage based on the physical shielding relationship between the forest and grass vegetation coverage and the woody vegetation coverage and the herbaceous vegetation coverage.

[0125] In summary, due to the strong spatial heterogeneity of the surface landscape under complex terrain conditions, it is difficult to directly obtain endmember spectra from satellite remote sensing images in practical applications, and the method of indirectly extracting endmember spectra only uses the coarse-scale observation information of medium and low-resolution satellites, which makes it difficult to effectively extract endmember types and their spectral information that can meet the actual surface coverage. In response to this problem, the present invention innovatively uses the field observation data of unmanned aerial vehicles to first provide endmember types and endmember abundance information, and then inputs these prior knowledge into the linear mixed pixel decomposition model to inversely calculate the endmember spectral information of the corresponding endmember type. This idea improves the traditional method of directly selecting endmember spectra from remote sensing images, and uses the high-resolution observation data of unmanned aerial vehicles as prior knowledge to improve the accuracy of endmember spectrum extraction, providing technical support for improving the inversion accuracy of total coverage of forest and grass vegetation.

[0126] Since the spectral characteristics of woody and herbaceous vegetation are similar, the traditional vegetation coverage inversion methods (empirical models, mechanism models or mixed pixel decomposition) based on medium-resolution spectral characteristics cannot finely separate woody and herbaceous vegetation coverage, which will cause large errors in the inversion results of woody vegetation coverage. In response to this problem, the present invention proposes to make full use of the texture characteristics of high-resolution satellite observation data, combine the medium-resolution spectral characteristics with the high-resolution texture characteristics, and use the random forest model to construct the best response relationship between different remote sensing feature combinations and the measured data of woody vegetation coverage, so as to achieve high-precision inversion of woody vegetation coverage in a large area and provide data support for the estimation of herbaceous vegetation coverage.

[0127] The above is only a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for refining forest and grass vegetation coverage based on multi-scale remote sensing, characterized in that: include: Step 1: Obtain ground sample plot drone images and classify the acquired drone images into green vegetation, dry vegetation, and soil; Step 2: The average surface reflectance of the Landsat pixel corresponding to the coordinates of the ground sample plot range is used as the mixed pixel observation signal value of the Landsat sample plot; green vegetation, dry vegetation and soil are used as the end member types in the pixel; based on the classification results of the drone image of each sample plot, the coverage value of each end member type is obtained, and the coverage value is used as the abundance prior knowledge of the end member type; Step 3: Based on the prior knowledge of the abundance of endmember types, the endmember spectra are reversely calculated using the linear mixed pixel decomposition model to obtain the endmember spectral library; Step 4: Based on the acquired endmember spectral library data, the abundance of green vegetation, dry vegetation and soil is forward inverted using a multi-endmember mixed pixel decomposition model to obtain the abundance of green vegetation, dry vegetation and soil at a regional scale, and then the dry vegetation coverage is obtained. The sum of the abundances of green vegetation and dry vegetation is calculated to obtain the forest and grass vegetation coverage; Step 5: Use human-computer interactive interpretation to obtain the woody vegetation in each sample plot UAV image and calculate the woody vegetation coverage, thereby constructing a sample dataset of woody vegetation coverage; Step 6: Extract spectral and texture features of Landsat and high-resolution satellite image data; Step 7: Take the woody vegetation coverage sample data set as the target variable, the extracted spectral features, texture features and the forest and grass vegetation coverage obtained in step 4 as feature variables, and use the random forest machine learning model to construct the nonlinear relationship between woody vegetation coverage and spectral features, texture features and forest and grass vegetation coverage, and predict the woody vegetation coverage; Step 8: Based on the physical shielding relationship between forest and grass vegetation coverage, woody vegetation coverage, and herbaceous vegetation coverage, the herbaceous vegetation coverage is calculated in reverse.

2. The method for refining forest and grass vegetation coverage based on multi-scale remote sensing according to claim 1 is characterized in that: In step 1, the acquired drone images are classified in the following manner: Step 1.1: Select green vegetation, dry vegetation and soil samples based on field survey photos and drone images; Step 1.2: Use the color space transformation function to convert the RGB color value of the drone image to obtain four color spaces: RGB, HSV, XYZ, and L * a * b * The color characteristics of 12 dimensions; Step 1.3: Use the support vector machine supervised classification method to automatically classify the color features after RGB transformation, and combine it with the visual interpretation method to obtain the classification results of drone images of green vegetation, dry vegetation and soil.

3. The method for refining forest and grass vegetation coverage based on multi-scale remote sensing according to claim 1 is characterized in that: The calculation formula of the linear mixed pixel decomposition model is: x k =M T f k +ε Among them, x k is a b×1 vector, representing the surface reflectance of the kth sample plot, where b is the number of bands; f k is a c×1 vector, representing the endmember coverage in the kth plot, and c is the number of endmember categories; M is a c×b matrix, representing the endmember spectral matrix, and each row of array elements represents the spectrum of a specific endmember type; ε represents the model fitting error.

4. The method for refining forest and grass vegetation coverage based on multi-scale remote sensing according to claim 3 is characterized in that: In step 3, for n sample plots, the linear mixed pixel decomposition model is modified as follows: X=FM Where X is an n×b matrix, X=[x1,x2,…,x k ,…,x n ] T ; F is an n×c matrix, F=[f1,f2,…f k ,…f n ] T ; The end-member spectrum matrix M of the study area is obtained by using the least squares method: M=(F T F) -1 F T X。 5. The method for refining forest and grass vegetation coverage based on multi-scale remote sensing according to claim 1 is characterized in that: The calculation formula of the multi-end member mixed pixel decomposition model is: Where X is the actual observed value of the image pixel; N is the number of end member categories; d i is the number of endmember spectra of the i-th endmember type; e ij is the jth endmember spectrum of the i-th type endmember; p ij is the proportion of the jth spectrum of the i-th endmember in the pixel; q is 0 or 1, representing the label of whether the endmember spectrum participates in the unmixing; n is the error component; 0≤p ij ≤1 and 6. The method for refining forest and grass vegetation coverage based on multi-scale remote sensing according to claim 5 is characterized in that: In step 4, the multi-terminal mixed pixel decomposition model with the minimum root mean square error of the pixel spectral simulation value is selected as the optimal model for spectral unmixing during the calculation process. The p calculated by the optimal model for spectral unmixing is ij That is, the abundance value of each end member type in the pixel. The spectral unmixing optimal model is used to calculate the abundance of green vegetation, dry vegetation and soil at the regional scale. The sum of the abundance of green vegetation and dry vegetation is the forest and grass vegetation coverage.

7. The method for refining forest and grass vegetation coverage based on multi-scale remote sensing according to claim 1 is characterized in that: The spectral characteristics include surface reflectance of all bands, normalized difference vegetation index, visible light band difference vegetation index, normalized difference moisture index, ratio vegetation index, enhanced vegetation index, improved soil adjusted vegetation index, difference vegetation index, green light normalized difference vegetation index, green band index and near infrared vegetation index.

8. The method for refining forest and grass vegetation coverage based on multi-scale remote sensing according to claim 1 is characterized in that: In step 6, the texture features of the surface reflectance of all bands are calculated using a gray level co-occurrence matrix based on statistical analysis.

9. The method for refining forest and grass vegetation coverage based on multi-scale remote sensing according to claim 1 is characterized in that: In step 8, the herbaceous vegetation coverage is calculated by back-calculation according to the following formula: f total =f woody +(1-f woody )×f herb Among them, f total is the forest and grass vegetation coverage, f woody and f herb are the coverage of woody vegetation and herbaceous vegetation, respectively.

10. A multi-scale remote sensing-based forest and grass vegetation coverage refinement inversion system, characterized in that: include: The ground sample plot UAV image acquisition and classification module is used to acquire ground sample plot UAV images and classify the acquired UAV images into green vegetation, dry vegetation and soil; The module for obtaining the coverage value of end-member types is used to take the average surface reflectance of the Landsat pixel corresponding to the coordinates of the ground sample plot range as the mixed pixel observation signal value of the Landsat sample plot; take green vegetation, dry vegetation and soil as the end-member types in the pixel; based on the classification results of the drone image of each sample plot, obtain the coverage value of each end-member type, and use the coverage value as the abundance prior knowledge of the end-member type; The endmember spectrum library back-calculation module is used to reversely calculate the endmember spectrum based on the prior knowledge of the endmember type abundance and use the linear mixed pixel decomposition model to obtain the endmember spectrum library; The multi-endmember mixed pixel decomposition module is used to perform forward inversion on the abundance of green vegetation, dry vegetation and soil based on the acquired endmember spectral library data using the multi-endmember mixed pixel decomposition model, obtain the abundance of green vegetation, dry vegetation and soil at the regional scale, and then obtain the dry vegetation coverage, and calculate the sum of the abundance of green vegetation and dry vegetation to obtain the forest and grass vegetation coverage; The module for constructing a sample dataset of woody vegetation coverage is used to obtain the woody vegetation in each plot drone image using a human-computer interactive interpretation method and calculate the woody vegetation coverage, thereby constructing a sample dataset of woody vegetation coverage; Feature variable set building module, used to extract spectral and texture features of Landsat and high-resolution satellite image data; The woody vegetation coverage prediction module is used to take the woody vegetation coverage sample data set as the target variable, and the extracted spectral features, texture features and forest and grass vegetation coverage obtained by the multi-end-member mixed pixel decomposition module as the feature variable, and use the random forest machine learning model to construct the nonlinear relationship between woody vegetation coverage and spectral features, texture features and forest and grass vegetation coverage, and predict the woody vegetation coverage; The herbaceous vegetation coverage prediction module is used to reversely calculate the herbaceous vegetation coverage based on the physical shielding relationship between the forest and grass vegetation coverage and the woody vegetation coverage and the herbaceous vegetation coverage.

Images