Vegetation canopy shadow coverage remote sensing calculation method based on mixed pixel decomposition

Abstract

The invention relates to the field of remote sensing calculation, and discloses a vegetation canopy shadow coverage remote sensing calculation method based on mixed pixel decomposition, comprising the following steps: S1, data acquisition and preprocessing; s2, red edge parameter extraction; and S3, constructing a mixed pixel decomposition model, introducing an NDCSI-BSI pixel trisection model, decomposing the pixel into three end members including an illumination canopy, a shadow canopy and bare soil, and calculating the abundance value of each end member based on a non-negative least square method. The method improves the precision of coverage estimation, expands the application range of a traditional model, and is suitable for the fields of regional ecological monitoring, precision agriculture, carbon cycle research and the like.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing computing, and particularly to a method for remotely sensing and calculating the shadow coverage of a vegetation canopy based on mixed pixel decomposition. Background Art

[0002] The light and shadow distribution of the vegetation canopy has an important impact on the photosynthesis, evapotranspiration, biomass estimation, etc. of vegetation. Traditional methods for estimating vegetation coverage usually ignore the influence of canopy shadows, resulting in large errors in the estimation results. With the development of remote sensing technology, especially the wide application of hyperspectral and multispectral remote sensing data, the mixed pixel decomposition technology has become an effective means to solve the problem of estimating the shadow coverage of the vegetation canopy. The mixed pixel decomposition technology decomposes the mixed pixels in the remote sensing image into multiple endmembers (such as illuminated canopy, shadow canopy, bare soil, etc.), and calculates the abundance (i.e., coverage) of each endmember, so as to achieve accurate estimation of the light and shadow coverage of the vegetation canopy.

[0003] However, the existing mixed pixel decomposition methods still have the following problems in the estimation of the shadow coverage of the vegetation canopy: single data source: the existing methods usually only rely on a single type of spectral data (such as hyperspectral or multispectral), lacking the fusion of multi-source data, resulting in limited estimation accuracy; insufficient shadow quantification: the existing methods have insufficient quantitative expression of the vegetation canopy shadow, making it difficult to accurately distinguish between the illuminated canopy and the shadow canopy; high model complexity: the existing methods mostly adopt complex non-linear models, with a large amount of calculation, making it difficult to be extended to large-scale remote sensing data processing. Therefore, we propose a method for remotely sensing and calculating the shadow coverage of a vegetation canopy based on mixed pixel decomposition to solve the above-mentioned problems. Summary of the Invention

[0004] The purpose of the present invention is to solve the disadvantages existing in the background art, and to propose a method for remotely sensing and calculating the shadow coverage of a vegetation canopy based on mixed pixel decomposition.

[0005] To achieve the above object, the technical solution adopted by the present invention is as follows: 1. A method for remotely sensing and calculating the shadow coverage of a vegetation canopy based on mixed pixel decomposition, characterized by comprising the following steps:

[0006] S1. Data acquisition and preprocessing, the steps include:

[0007] S11. Obtain the vegetation canopy image by using unmanned aerial vehicle hyperspectral data and Sentinel-2A multispectral data;

[0008] S12. Perform geometric correction, radiometric correction and spectral normalization processing on the image;

[0009] S2. Red-edge parameter extraction, the steps include:

[0010] S21. Extract the red-edge features based on spectral derivative technology, including the red-edge slope and red-edge reflectance;

[0011] S22. Combine the normalized difference vegetation index (NDVI) to construct the normalized difference canopy shadow index (NDCSI);

[0012] S3. Construct a mixed pixel decomposition model. Introduce the NDCSI-BSI three-endmember pixel model to decompose the pixel into three endmembers: sunlit canopy, shaded canopy, and bare soil. Calculate the abundance values of each endmember based on the non-negative least squares method, and estimate the sunlit and shaded coverages. The steps include:

[0013] S31. Remote sensing calculation of the shaded canopy coverage of vegetation based on the pixel dichotomy model;

[0014] S32. Remote sensing calculation of the shaded canopy coverage of vegetation based on the NDCSI-BSI three-endmember pixel model.

[0015] Further, in step S21, the red-edge feature extraction formula is as follows:

[0016]

[0017] In the formula, R(λ i ) represents the spectral reflectance at the wavelength λ of the i-th band, and Δλ represents the adjacent wavelength interval. i at which.

[0018] Further, in step S22, the calculation formula of the normalized difference canopy shadow index (NDCSI) is as follows:

[0019]

[0020]

[0021] In the formula, RedEdge represents the band reflectance at the red edge; NDVI is the normalized difference vegetation index, and RedEdge min and RedEdge max are respectively the minimum and maximum values of the band reflectance at the red edge of all pixel points in the research image; NIR is the reflectance of the near-infrared band, and Red is the reflectance of the red light band.

[0022] Further, in step S31, the calculation formula of the shaded canopy coverage of vegetation is as follows:

[0023] Fc = (S - Ssha) / (Slig - Ssha)

[0024] Wherein, S is the surface information observed by the sensor element, Ssha is the remote sensing information obtained from a pure pixel completely covered by the vegetation shadow canopy of the pixel, Slig is the remote sensing information obtained from a pure pixel completely covered by the vegetation illuminated canopy, and Ssha and Slig are two parameters of the pixel dichotomy model, which are obtained through information measurement calculation or through vegetation index inversion.

[0025] Furthermore, by solving the following system of equations, the coverage of the vegetation illuminated canopy and the shadow canopy of each mixed pixel can be obtained:

[0026] C = f LC C LC + f SC C SC + f BS C BS

[0027] B = f LC B LC + f SC B SC + f BS B BS

[0028] f LC + f SC + f BS = 1

[0029] Wherein, C represents the NDCSI index of the pixel, B represents the BSI index of the pixel; f LC 、f SC 、f BS respectively represent the proportions of the vegetation illuminated canopy, the vegetation shadow canopy, and the soil within the pixel, C LC 、C SC 、C BS respectively represent the endmember characteristic values of the vegetation illuminated canopy LC, the vegetation shadow canopy SC, and the bare soil BS in the NDCSI index, B LC 、B SC 、B BS respectively represent the endmember characteristic values of the vegetation illuminated canopy LC, the vegetation shadow canopy SC, and the bare soil BS in the BSI index.

[0030] Furthermore, the remote sensing calculation method for the vegetation canopy shadow coverage based on mixed pixel decomposition of the present invention further includes:

[0031] S4. Multi-source data integration, simulating the red-edge characteristics of hyperspectral data to multispectral data to expand the applicable range of the model; in step S4, the near-infrared band available in all remote sensing satellite sensors is selected to replace the red-edge band parameter information, and a normalized vegetation canopy shadow index applicable to multispectral remote sensing images is constructed:

[0032]

[0033] In the formula, NIR represents the reflectance in the near-infrared band.

[0034] Furthermore, the remote sensing calculation method for the shadow coverage of the vegetation canopy based on mixed pixel decomposition in the present invention further includes:

[0035] S5. Accuracy verification, the steps include:

[0036] S51. Use the ground-measured coverage data to verify the model results;

[0037] S52. Evaluate the correlation between the estimated coverage value and the actual value through regression analysis.

[0038] Among them, in the step S52, four types of indicators based on the confusion matrix are selected to represent the accuracy verification evaluation, including the Kappa coefficient Kappa, the overall classification accuracy OA, the mapping accuracy PA, and the user accuracy UA. The calculation formulas are as follows:

[0039]

[0040] In the formula, p i,j represents the number of samples in the i-th row and j-th column of the confusion matrix, represents the number of samples that are actually in category i but are classified as j, N represents the total number of samples, that is, the number of all verification points, and n represents the number of classification categories.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] By combining hyperspectral and multispectral remote sensing data and using the mixed pixel decomposition technology, the present invention quantitatively estimates the light and shadow coverage of the vegetation canopy, solves the problems of insufficient estimation accuracy of the shadow coverage of the vegetation canopy, single data source, and high model complexity in the prior art. This method has high accuracy and broad application prospects, and can provide technical support for vegetation ecological research, carbon cycle monitoring, vegetation stress analysis, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is the overall technical route flowchart of a remote sensing calculation method for the shadow coverage of a vegetation canopy based on mixed pixel decomposition according to the present invention;

[0044] Figure 2 is the aerial hyperspectral image taken by the S185 imaging spectrometer (upper right: local larch; lower right: local Chinese pine);

[0045] Figure 3 is a schematic diagram of the spectral curves of larch pixels with different shadow degrees (80 points);

[0046] Figure 4Schematic diagram of spectral curves of Pinus tabuliformis pixels with different shading degrees (80 points);

[0047] Figure 5 Schematic diagram of the characteristic space of the NDCSI - BSI pixel three - component model;

[0048] Figure 6 Schematic diagram of the linear relationship between the slope at the red edge and the reflectance at the red edge of Larix gmelinii (80 points);

[0049] Figure 7 Schematic diagram of the linear relationship between the slope at the red edge and the reflectance at the red edge of Pinus tabuliformis (80 points);

[0050] Figure 8 Schematic diagram of the linear relationship between the slope at the red edge and the reflectance at the red edge without distinguishing tree species (160 points);

[0051] Figure 9 Schematic diagram of the linear relationship between the slope at the red edge and NDVI of Larix gmelinii (80 points);

[0052] Figure 10 Schematic diagram of the linear relationship between the slope at the red edge and NDVI of Pinus tabuliformis (80 points);

[0053] Figure 11 Schematic diagram of the vegetation canopy study area and its NDVI calculation;

[0054] Figure 12 Schematic diagram of the estimation result of vegetation light canopy coverage;

[0055] Figure 13 Schematic diagram of endmember extraction based on the PPI index and the vertices of the two - dimensional scatter plot;

[0056] Figure 14 Hyperspectral image abundance map based on three - component model mixed pixel decomposition (the upper left is the original image, the upper right is the light canopy, the lower left is the light bare soil, and the lower right is the shaded canopy);

[0057] Figure 15 Schematic diagram of the linear relationship between the cumulative visible light reflectance and the light canopy coverage (40 points);

[0058] Figure 16 Linear relationship between the cumulative visible light reflectance and the light canopy coverage (26 points). Detailed implementation manners

[0059] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments in the following description are only examples, and those skilled in the art can think of other obvious variations.

[0060] Example: In this example, Wangyedian Forest Farm in Chifeng City, Inner Mongolia is taken as the study area, and the main vegetation types are Chinese pine and larch.

[0061] As Figure 1 This is the flowchart of the remote sensing calculation method for the vegetation canopy shadow coverage based on mixed pixel decomposition in this example. The remote sensing calculation method for the vegetation canopy shadow coverage of the present invention includes the following steps:

[0062] S1. Data acquisition and preprocessing, the steps include:

[0063] S11. Use an unmanned aerial vehicle hyperspectral imager to obtain image data (hyperspectral data) with a resolution of 1 meter, and use Sentinel-2A images to simulate multispectral data with a resolution of 10 meters. The vegetation canopy images of the target area are included in the unmanned aerial vehicle hyperspectral data and Sentinel-2A multispectral data. As Figure 2 Shown is the aerial hyperspectral image taken by the S185 imaging spectrometer. In the figure, the upper right is a partial view of larch; the lower right is a partial view of Chinese pine. The spectral acquisition range of this spectrometer is 450 - 950 nm, the spectral resolution is 4 nm. When operating at a flight altitude of 50 m, the focal length of the S185 imaging spectrometer reaches 17 mm, and hyperspectral images with a spatial resolution of up to 0.21 m can be obtained under good environmental conditions.

[0064] S12. Perform geometric correction and atmospheric radiation correction on the image data, remove noise interference, ensure the authenticity of spectral information, and perform spectral normalization processing. As Figure 3 And Figure 4 Shown are the spectral curve schematic diagrams of larch pixels and Chinese pine pixels with different shadow degrees in the Sentinel-2A multispectral data. It can be seen from the figure that although the spectral curve shapes of the same vegetation under different shadow degrees are similar, their amplitudes are significantly different. The brighter the vegetation canopy of larch and Chinese pine, the greater the amplitude of its spectral curve, and vice versa, the darker, the smaller the amplitude of its spectral curve. It can be seen from the red edge parameter information statistical table that for both larch and Chinese pine, the red edge slope and red edge reflectance of the bright vegetation canopy are significantly higher than the corresponding red edge data of the shadow canopy.

[0065] S2. Red edge parameter extraction:

[0066] S21. Extract the red edge slope and red edge reflectance of each pixel based on the hyperspectral data.

[0067] The red edge feature extraction formula is as follows:

[0068]

[0069] In the formula, R(λi ) represents the wavelength λ of the i-th band i of the spectral reflectance at that point, and Δλ represents the adjacent wavelength interval.

[0070] S22. Combine the normalized difference vegetation index NDVI to construct the normalized difference canopy shadow index NDCSI. In step S22, the calculation formula of the normalized difference canopy shadow index NDCSI is as follows:

[0071]

[0072] In the formula, RedEdge represents the band reflectance at the red edge; NDVI is the normalized difference vegetation index, and the NDVI multiplier factor can ensure the morphological characteristics of vegetation and partially eliminate the influence of radiation changes related to atmospheric conditions such as satellite viewing angle, solar elevation angle, terrain, and cloud shadow. RedEdge min and RedEdge max are respectively the minimum value and the maximum value of the band reflectance values at the red edge of all pixel points in the research image; in practical applications, according to the statistical information of the red edge band histogram of the image, a threshold within a certain range (for example, the first 1% and the first 99%, which can be adjusted by oneself) can be selected as the minimum value and the maximum value.

[0073] As Figure 6 , Figure 7 and Figure 8 shown, in this embodiment, it can be seen from the schematic diagram of the linear relationship of the red edge parameters of single tree species at 80 points that there is an excellent linear regression relationship between the red edge slope and the red edge reflectance of Larix gmelinii and Pinus tabuliformis single tree species, and the R2 values reach 0.8504 and 0.927 respectively. In the schematic diagram obtained by combining and statistically analyzing the data of the two tree species, the results of the slope at the red edge and the reflectance at the red edge of the overall 160 points also have an excellent linear regression relationship, and the R2 reaches 0.8917. Whether it is the canopy of Larix gmelinii or Pinus tabuliformis, the brighter the vegetation canopy, the greater the reflectance at the red edge and the slope at the red edge, and vice versa, the smaller the red edge parameter values of the more shaded canopy. Moreover, by combining the sample points of the two tree species to establish a regression relationship model, the results still have a good linear correlation. That is, the red edge information of the vegetation canopy can be used to quantitatively express the brightness / shadow degree of the vegetation canopy and is not affected by the information of different tree species.

[0074] As Figure 9 , Figure 10As shown, the NDVI values of the spectral data of the larch and Chinese pine vegetation canopies with 80 sample points at different brightness / shadow levels mostly range from 0.6 - 0.8 and 0.7 - 0.9 respectively, and are basically not affected by the slope at the red edge. Since it is known from the study of spectral red edge information that the slope at the red edge can quantitatively express the brightness / shadow level of the vegetation canopy, it can be concluded that the NDVI value of the vegetation canopy spectrum is basically not affected by the shadow level, and using NDVI to construct a vegetation index can ensure stable vegetation morphological characteristics.

[0075] Using the pixel dichotomy model composed of the illuminated canopy and shadow canopy of vegetation, based on the measured flight data of S185 UAV hyperspectral imaging in the experimental area of Wangyedian Forest Farm, the normalized vegetation canopy shadow index NDCSI is used to estimate the coverage of the illuminated vegetation canopy. A 2000 * 2000 pixel dense Chinese pine canopy forest land is intercepted as the research area, as Figure 11 shown.

[0076] In the research area of the dense Chinese pine vegetation canopy, a pixel can be regarded as consisting of and only consisting of the covered part of the illuminated vegetation canopy and the covered part of the vegetation shadow canopy. The spectral information received by the remote sensing satellite sensor elements is obtained by mathematically linearly weighting and mixing the pure spectra of these two endmembers. The ratio of the area of each pure endmember in the pixel to the total area is the abundance weight of each endmember. That is, the normalized vegetation canopy shadow index NDCSI can be used to estimate the weights of the illuminated vegetation canopy and the shadow canopy, that is, the corresponding coverage information. Based on the research conclusion of the vegetation red edge parameters, for the convenience of calculating the canopy red edge parameter information of the entire scene image, the 710 nm band is selected from the S185 UAV hyperspectral image to approximately represent the red edge band data at that point. The coverage information of the vegetation shadow canopy and the illuminated vegetation canopy in the research area of the dense Chinese pine canopy forest land can be calculated. The closer the calculated result is to 1, the greater the coverage of the illuminated canopy within the pixel; conversely, the closer the calculated result is to 0, the greater the coverage of the shadow canopy within the pixel. The results are as Figure 12 shown.

[0077] S3. Construction of the mixed pixel decomposition model. Introduce the NDCSI - BSI pixel trichotomy model, decompose the pixel into three endmembers: the illuminated canopy, the shadow canopy, and bare soil, and calculate the abundance values of each endmember based on the non - negative least - squares method to estimate the illumination and shadow coverage. The steps include:

[0078] S31. Remote sensing calculation of the vegetation shadow canopy coverage based on the pixel dichotomy model (the most ideal state is all canopy vegetation). In step S31, the calculation formula for the vegetation shadow canopy coverage is as follows:

[0079] Fc = (S - Ssha) / (Slig - Ssha)

[0080] In the formula, S represents the surface information observed by the sensor element, Ssha represents the remote sensing information obtained from a pure pixel completely covered by the vegetation shadow canopy of the pixel, Slig represents the remote sensing information obtained from a pure pixel completely covered by the vegetation illuminated canopy, and Ssha and Slig are two parameters of the pixel dichotomy model, which are obtained by information measurement calculation or by inversion of vegetation indices. Using the spectral information of the vegetation illuminated canopy and the shadow canopy of remote sensing to estimate the coverage, this model shows the relationship between the remote sensing spectral information and the coverage of the vegetation illuminated canopy. Its parameters Ssha and Slig have actual physical meanings, that is, the remote sensing information reflected by the pure pixels of the vegetation illuminated canopy and the vegetation shadow canopy. In this way, the influence of noise interference backgrounds such as soil, atmosphere, and vegetation types on the estimation of remote sensing information is minimized, leaving only the information of the coverage of the vegetation illuminated canopy.

[0081] S32. Remote sensing calculation of the coverage of the vegetation shadow canopy based on the NDCSI-BSI pixel trichotomy model (there is bare soil component in the actual image).

[0082] On the basis of the theory of the pixel dichotomy model, Guerschman et al. proposed the pixel trichotomy model, that is, it is assumed that the mixed pixel is composed of three parts: photosynthetic vegetation (PV), non-photosynthetic vegetation (NPV), and soil (BaredSoil, BS). The pixel information received by the sensor is a linear combination of the information of the three basic components of PV, NPV, and BS with the proportion of the pixel area they occupy as the weight coefficient. Then, two appropriate vegetation indices with non-linear relationships are selected to construct the relevant pixel trichotomy model.

[0083] Rikimaru proposed the bare soil index BSI (bare soil index, BSI) in 1996, and the calculation formula is as follows:

[0084]

[0085] In the formula, ρ mir1 represents the reflectance value of the mid-infrared 1 band, ρ red represents the reflectance value of the red light band, ρ nir represents the reflectance value of the near-infrared band, ρ blue represents the reflectance value of the blue light band.

[0086] Analogous to the pixel trichotomy model proposed by Guerschman, the present invention proposes the NDCSI-BSI pixel trichotomy model (for the schematic diagram of the NDCSI-BSI pixel trichotomy model feature space, see Figure 5), assuming that the mixed pixel consists of three parts: the Lightness Canopy (LC) of vegetation, the Shadow Canopy (SC) of vegetation, and the Bared Soil (BS). The NDCSI and BSI indices conform to a linear relationship, and the NDCSI-BSI feature space of pixels with various proportional combinations will appear as a triangle (as shown in the following figure). The NDCSI of the Lightness Canopy of vegetation is high, the BSI is low and negative, located at the lower right corner of the triangle; the NDCSI of the Shadow Canopy of vegetation is low, the BSI is low and negative, located at the lower left corner of the triangle; the NDCSI of the soil is low, the BSI is high and positive, located at the upper left corner of the triangle, as Figure 5 shown.

[0087] By solving the following system of equations, the coverage of the Lightness Canopy and Shadow Canopy of vegetation for each mixed pixel can be obtained:

[0088] C = f LC C LC + f SC C SC + f BS C BS

[0089] B = f LC B LC + f SC B SC + f BS B BS

[0090] f LC + f SC + f BS = 1

[0091] In the formula, C represents the Normalized Difference Canopy Shadow Index NDCSI of the pixel, and B represents the Bare Soil Index BSI of the pixel; f LC , f SC , f BS represent the proportions of the Lightness Canopy of vegetation, the Shadow Canopy of vegetation, and the soil in the pixel respectively. C LC , C SC , C BS represent the endmember eigenvalues of the Lightness Canopy of vegetation LC, the Shadow Canopy of vegetation SC, and the Bare Soil BS in the NDCSI index respectively. B LC , B SC , B BS represent the endmember eigenvalues of the Lightness Canopy of vegetation LC, the Shadow Canopy of vegetation SC, and the Bare Soil BS in the BSI index respectively.

[0092] Based on the entire Sentinel-2A image, the number of pixels in the selected research area of Wangyedian Forest Farm in Chifeng is 1200*1200, and 91.9% of the pixels have an NDVI value greater than 0.4. The statistical data is as Figure 13 shown. The obtained vegetation illuminated canopy and shadow canopy abundance maps, that is, the coverage estimation results are as Figure 14 shown. It can be preliminarily concluded from the resulting abundance map that based on the Sentinel-2A multispectral data, using the pixel trisection model, by judging and selecting the spectral data of the vegetation illuminated canopy, shadow canopy, and bare soil endmembers, the abundance maps of the corresponding endmembers can be calculated, and the general distribution is compared with the true color map of the original image, and the texture details are clearly distinguishable.

[0093] S4. Integration of multi-source data, simulating the red-edge characteristics of hyperspectral data into multispectral data to expand the applicable range of the model.

[0094] The characteristics of hyperspectral remote sensing are the combination of spectrum and image, and there are numerous bands. Its spectral transmission channels are dozens or even hundreds, and the arrangement between each spectral channel is often continuous. Hyperspectral remote sensing data is easy to construct vegetation indices related to the parameters of the red-edge band (or bands near the range) using spectral differential transformation, but it is not conducive to being extended to multispectral remote sensing applications with fewer bands. In order to make the vegetation canopy shadow index easy to promote and have strong applicability, for the vast majority of multispectral data sources (except for special ones such as Sentinel satellites), the near-infrared band that all remote sensing satellite sensors have is selected to replace the red-edge band parameter information, and the normalized difference idea is used to avoid the inconvenience caused by too large or too small data values in use, and the following normalized vegetation canopy shadow index (Normalized Difference Canopy Shadow Index, NDCSI) applicable to multispectral remote sensing images is constructed:

[0095]

[0096] In the formula, NIR represents the reflectance of the near-infrared band, NIR min the minimum value of the reflectance of the near-infrared band, NIR max the maximum value of the reflectance of the near-infrared band. This is mainly for facilitating the extension to different multispectral satellite data for use.

[0097] S5. Accuracy verification, the steps include:

[0098] S51. Using the ground-measured coverage data to verify the model results,

[0099] S52. Evaluating the correlation between the estimated coverage value and the actual value through regression analysis.

[0100] Four types of indicators based on the Confusion Matrix are selected to represent the accuracy verification evaluation (Hay, 1988; Zheng Mingguo et al., 2006), including the Kappa Coefficient (Kappa), the Overall Accuracy (OA), the Producer's Accuracy (PA), and the User's Accuracy (UA).

[0101] The calculation formulas for the Kappa coefficient, the overall classification accuracy OA, the mapping accuracy PA, and the user accuracy UA are as follows:

[0102]

[0103] In the formula, p i,j represents the number of samples in the i-th row and j-th column of the confusion matrix, represents the number of samples that are actually of class i but are classified as j, N represents the total number of samples, that is, the number of all verification points, and n represents the number of classification categories.

[0104] As Figure 15 shown, from the regression analysis diagram of 40 measured verification sample points of hyperspectral data, it can be seen that the calculated result of the light canopy coverage of Chinese pine vegetation has an excellent linear relationship with the cumulative spectral reflectance in the visible light range (representing the measured light brightness information) at this point, and R2 reaches 0.8862. This indicates that the normalized vegetation canopy shadow index (NDCSI) can be used to estimate the vegetation shadow canopy and light canopy coverage information with relatively high accuracy. However, it should be noted that for the 5 measured points where the NDCSI value is close to 0.8, compared with the other 35 points with smaller values, the degree of deviation from the linear relationship is relatively large. It can be considered that when the light intensity reaches a certain level or is close to saturation, it will affect the calculation result of the coverage at this point.

[0105] As Figure 16 shown, from the regression analysis diagram of 26 verification sample points of Sentinel-2A satellite image data, it can be seen that the result of the light canopy coverage of vegetation calculated by the NDCSI-BSI pixel three-component model has an excellent linear relationship with the cumulative spectral reflectance in the visible light range (representing the light brightness information) at this point, and R2 reaches 0.8475. This indicates that by using the normalized vegetation canopy shadow index (NDCSI) and combining it with auxiliary vegetation indices, the vegetation shadow canopy and light canopy coverage information can be estimated with relatively high accuracy.

[0106] Table 1 shows the red-edge parameter information of larch and Chinese pine pixels with different shadow degrees (partial)

[0107]

[0108]

[0109]

[0110] Table 2 shows the statistics of the coverage and cumulative reflectance of the verification points of the vegetation canopy with different brightnesses (40 points)

[0111]

[0112] Table 3 shows the statistics of the coverage and cumulative visible light reflectance of the vegetation canopy with different brightnesses (26 points)

[0113]

[0114]

[0115] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification is only the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection required by the present invention is defined by the appended claims and their equivalents.

Claims

1. A remote sensing calculation method for vegetation canopy shadow coverage based on mixed pixel decomposition, characterized in that: The following steps are involved: S1. Data acquisition and preprocessing, the steps include: S11. Obtain vegetation canopy images using UAV hyperspectral data and Sentinel-2A multispectral data; S12. Perform geometric correction, radiation correction and spectral normalization on the image; S2, red edge parameter extraction, the steps include: S21. Extract red edge features based on spectral differentiation technology, including red edge slope and red edge reflectivity; S22. Combine the Normalized Difference Vegetation Index NDVI to construct the Normalized Difference Vegetation Canopy Shadow Index NDCSI; S3. Construction of mixed pixel decomposition model. The NDCSI-BSI pixel trisection model is introduced to decompose the pixel into three end members: illuminated canopy, shadowed canopy and bare soil. The abundance value of each end member is calculated based on the non-negative least squares method to estimate the light and shadow coverage. The steps include: S31. Remote sensing calculation of vegetation shadow canopy coverage based on pixel binary model; S32. Remote sensing calculation of vegetation shadow canopy coverage based on NDCSI-BSI pixel trisection model.

2. The remote sensing calculation method for vegetation canopy shadow coverage based on mixed pixel decomposition according to claim 1 is characterized in that: In the step S21, the red edge feature extraction formula is as follows: In the formula, R(λ i ) represents the wavelength λ of the i-th band i Δλ represents the spectral reflectance at adjacent wavelengths.

3. The remote sensing calculation method for vegetation canopy shadow coverage based on mixed pixel decomposition according to claim 1 is characterized in that: In the step S22, the calculation formula of the normalized vegetation canopy shadow index NDCSI is as follows: In the formula, RedEdge represents the reflectance of the red edge band; NDVI is the normalized vegetation index, and RedEdge min and RedEdge max They are the minimum and maximum values ​​of the red edge reflectance of all pixels in the studied image; NIR is the near-infrared band reflectance, and Red is the red light band reflectance.

4. The remote sensing calculation method for vegetation canopy shadow coverage based on mixed pixel decomposition according to claim 1 is characterized in that: In step S31, the calculation formula of vegetation shadow canopy coverage is as follows: Fc=(S-Ssha) / (Slig-Ssha) Where S is the surface information observed by the sensor element, Ssha is the remote sensing information obtained by the pure pixel with full coverage of vegetation shadow canopy, Slig is the remote sensing information obtained by the pure pixel with full coverage of vegetation illumination canopy, Ssha and Slig are two parameters of the pixel binary model, which are obtained through information measurement and calculation or through vegetation index inversion.

5. The remote sensing calculation method for vegetation canopy shadow coverage based on mixed pixel decomposition according to claim 1 is characterized in that: In step S32, the vegetation illumination canopy and shadow canopy coverage of each mixed pixel are obtained by solving the following equation group: C=f LC C LC +f SC C SC +f BS C BS B=f LC B LC +f SC B SC +f BS B BS f LC +f SC +f BS =1 Where C represents the normalized vegetation canopy shadow index NDCSI of the pixel, and B represents the bare soil index BSI of the pixel; f LC 、f SC 、f BS Respectively represent the proportion of vegetation illumination canopy, vegetation shadow canopy and soil in the pixel, C LC , C SC , C BS They represent the end-member eigenvalues ​​of the vegetation light canopy LC, vegetation shadow canopy SC, and bare soil BS in the NDCSI index, respectively. LC , B SC , B BS They respectively represent the endmember eigenvalues ​​of vegetation light canopy LC, vegetation shadow canopy SC and bare soil BS in the BSI index.

6. The remote sensing calculation method for vegetation canopy shadow coverage based on mixed pixel decomposition according to claim 1 is characterized in that Also includes: S4, multi-source data integration, simulating the red edge characteristics of hyperspectral data to multispectral data, and expanding the scope of application of the model.

7. The remote sensing calculation method for vegetation canopy shadow coverage based on mixed pixel decomposition according to claim 6 is characterized in that: In step S4, the near-infrared band common to all remote sensing satellite sensors is selected to replace the red edge band parameter information, and a normalized vegetation canopy shadow index suitable for multispectral remote sensing images is constructed: In the formula, NIR represents the reflectivity in the near infrared band, NIR min The minimum reflectivity in the near infrared band, NIR max The maximum reflectivity in the near-infrared band.

8. The remote sensing calculation method for vegetation canopy shadow coverage based on mixed pixel decomposition according to claim 1 is characterized in that Also includes: S5, accuracy verification, the steps include: S51. Use ground-measured coverage data to verify the model results; S52. Assess the correlation between coverage estimates and actual values ​​by regression analysis.

9. The remote sensing calculation method for vegetation canopy shadow coverage based on mixed pixel decomposition according to claim 8, characterized in that: In the step S52, the accuracy verification evaluation is represented by four types of indicators based on the confusion matrix, including Kappa coefficient Kappa, overall classification accuracy OA, mapping accuracy PA and user accuracy UA, and the calculation formulas are as follows: In the formula, p i,j It represents the number of samples in the i-th row and j-th column in the confusion matrix, which means the number of samples whose actual category is i but classified as j. N represents the total number of samples, that is, the number of all verification points, and n represents the number of classification categories.

Images