Wheat leaf area index whole growth period estimation method based on canopy pure hyperspectral reflectivity

Abstract

The invention provides a wheat leaf area index whole growth period estimation method based on canopy pure hyperspectral reflectivity, which comprises the following steps of: firstly, extracting wheat canopy spike, leaf and soil pure pixels based on an RGB (Red, Green and Blue) image combined with a deep learning algorithm; secondly, separating the canopy pure leaf hyperspectral reflectivity by using the extracted pure pixels and canopy hyperspectral reflectivity data in combination with a non-negative matrix factorization algorithm; and finally, calculating a vegetation index based on the canopy pure leaf hyperspectral reflectivity, and establishing a whole-growth-period LAI estimation model in combination with a simple linear regression algorithm. According to the method, the problem that the LAI inversion precision is reduced after heading is solved by combining near-end remote sensing and a machine learning algorithm, and the wheat LAI estimation method based on the canopy pure hyperspectral reflectivity has the potential of accurately predicting the LAI of the wheat in the whole growth period and can be expanded to accurate management of field crops and breeding of high-yield wheat varieties.

Description

Technical Field

[0001] The method of the present invention relates to the field of remote sensing monitoring of crop agro-meteorology, in particular to a monitoring method for improving the estimation accuracy of the leaf area index of wheat during the whole growth period based on canopy hyperspectral reflectance. Background Art

[0002] Wheat is one of the most important food crops in the world, and its high yield and good quality are of great significance for ensuring food security. The leaf area index (LAI), as a key ecological parameter describing the canopy structure of plants, is crucial for wheat growth monitoring and optimizing crop management.

[0003] At present, many scholars have used spectral technology to estimate the LAI of wheat. The common method is to construct an estimation model based on vegetation indices using canopy hyperspectral reflectance. However, the canopy hyperspectral reflectance is easily affected by the external environment. In the early growth stage of crops, it is mainly affected by the soil background. In the later growth stage, the ears of wheat and leaves are in the same canopy. The appearance of ears not only changes the canopy structure characteristics and light interception, but also has a significant impact on the canopy hyperspectral reflectance. In the early growth stage, some vegetation indices have been developed, such as SAVI and OSAVI, which have weakened the influence of the soil background on the inversion accuracy of crop growth parameters to a certain extent. However, in the later growth stage, the proportion of the soil background decreases, and the ear layer is located at the top of the canopy. The weakening of the soil background cannot reduce the influence of the ear layer on the canopy spectrum. At present, studies on rice have shown that eliminating the influence of the vegetation index on LAI estimation caused by the difference between rice ears and leaves in the later growth stage helps to improve the estimation accuracy of LAI after rice heading. However, there are few reports on the influence of wheat ears on LAI estimation.

[0004] Spectral unmixing technology can effectively separate the vegetation spectrum from the mixed spectrum of the wheat canopy and has been widely used in the decomposition of soil and vegetation spectra. However, the spectral decomposition of wheat ears and leaves is more complex. The difficulty lies in that first, it is necessary to obtain the high-precision proportion of ear and leaf components in the canopy. The currently commonly used canopy component segmentation method is based on the threshold method. By setting one or more thresholds, the pixels in the image are divided into two categories: vegetation and non-vegetation. However, this method is greatly affected by light and canopy structure. The canopy structure after heading will experience changes such as green leaves - green ears, yellow leaves - green ears, and yellow leaves - yellow ears. Therefore, higher requirements are placed on the stability of the segmentation algorithm. In view of this, there is currently a lack of a canopy ear and leaf segmentation method with high precision that can combine spectral unmixing technology to extract pure leaf spectral data of the canopy in the later growth stage, so as to be used for LAI estimation in multiple growth periods and have a high estimation accuracy after heading. Summary of the Invention

[0005] The present invention provides a method for estimating the wheat leaf area index (LAI) throughout the whole growth period based on the pure canopy hyperspectral reflectance, which includes the following steps: First, based on the RGB images and combined with a deep learning algorithm, the pure pixels of wheat ears, leaves and soil in the wheat canopy are extracted; Second, using the extracted pure pixels and the canopy hyperspectral reflectance data, the non-negative matrix factorization algorithm is combined to separate the pure leaf hyperspectral reflectance of the canopy; Finally, based on the pure leaf hyperspectral reflectance of the canopy, vegetation indices are calculated to establish an LAI estimation model for the whole growth period. This method improves the problem of the decline in the LAI inversion accuracy after heading by combining proximal remote sensing and machine learning algorithms. The proposed method for estimating wheat LAI based on pure canopy hyperspectral reflectance has the potential to accurately predict wheat LAI throughout the whole growth period and can be extended to the precise management of field crops and the breeding of high-yield wheat varieties.

[0006] The technical solution for achieving the object of the present invention is as follows:

[0007] A method for estimating the wheat leaf area index (LAI) throughout the whole growth period based on the pure canopy hyperspectral reflectance, which includes the following steps:

[0008] Step 1: Obtain the measured LAI in the field, the canopy RGB images and the hyperspectral reflectance during the tillering-filling period;

[0009] Step 2: Based on the canopy RGB images and combined with a deep learning algorithm, extract the pure pixels of wheat ears, leaves and soil in the wheat canopy;

[0010] Step 3: Based on the pure pixels obtained in Step 2 and the canopy hyperspectral reflectance data, combine the non-negative matrix factorization (NMF) algorithm to calculate the pure leaf hyperspectral reflectance of the canopy;

[0011] Step 4: Based on the pure leaf hyperspectral reflectance of the canopy obtained in Step 3, calculate the vegetation indices and combine the simple linear regression algorithm to construct an LAI estimation model for the whole growth period.

[0012] Furthermore, for the method for estimating the wheat leaf area index (LAI) throughout the whole growth period based on the pure canopy hyperspectral reflectance proposed by the present invention, the specific steps for obtaining the measured LAI in the field, the proximal RGB images and the hyperspectral reflectance data during the tillering-filling period in Step 1 include:

[0013] Step 1-1: Measure the canopy hyperspectral reflectance

[0014] The vertical reflectance spectrum of the wheat canopy was measured using a FieldSpec Pro FR2500 back-mounted field hyperspectral radiometer. During the spectral measurement, the height of the spectrometer probe from the canopy was approximately 1 m vertically. Before each measurement, a standard whiteboard calibration was performed. Three points were selected for measurement in each plot, and three spectral reflectances were recorded at each point. The average value was used as the canopy spectral reflectance of the plot.

[0015] Step 1-2: Measure the hyperspectral reflectance of the wheat ears in the canopy

[0016] The spectral reflectance of the wheat ears in the canopy was measured using a FieldSpec Pro FR2500 back-mounted field hyperspectral radiometer at the heading stage, flowering stage, early filling stage, and mid-filling stage. The spectral measurement time was selected from 10:00 to 14:00 on sunny, cloudless, or slightly cloudy days. The field of view angle of the instrument was 25°. A standard whiteboard calibration was performed before the measurement. Twenty representative wheat main stems were selected in each plot. The wheat ears were cut and held together in the hand. The spectrometer probe was placed directly above the wheat ears to ensure that only wheat ears were within the field of view angle of the instrument, and the spectral reflectance of the wheat ears was measured.

[0017] Step 1-3: Measure the hyperspectral reflectance of the canopy leaves

[0018] After the wheat headed, the spectral reflectance of the original wheat canopy was first measured using a FieldSpec Pro FR2500 spectrometer, and then all the wheat ears within the field of view of the instrument were cut off. The spectral reflectance of the wheat canopy after removing the ear layer was measured again. The measurement requirements were the same as those for the vertical canopy reflectance.

[0019] Step 1-4: Measure the hyperspectral reflectance of the soil

[0020] The height of the spectrometer probe from the ground was approximately 0.1 m. Before each measurement, a standard whiteboard calibration was performed. One piece of bare soil was selected for measurement in each plot. Three spectral reflectances were recorded at each point, and the average value was used as the soil spectral reflectance of the plot.

[0021] Step 1-5: Take RGB images of the wheat canopy

[0022] RGB images of the canopy were acquired using an iphone12 smartphone. The camera settings were as follows: focal length, 4 mm; aperture value, f / 1.6; ISO speed, ISO-32; auto exposure. The time for image acquisition was selected from 16:00 to 17:00 on the afternoon of the day of the field spectral test. During the shooting, the phone was located 1 m above the wheat canopy, and vertical-angle canopy images were taken with a resolution of 3024×3024. Canopy images of wheat in each plot were obtained under natural environmental conditions at the tillering stage, jointing stage, booting stage, heading stage, flowering stage, and early, middle, and late filling stages of wheat.

[0023] Step 1-6: Measure the LAI of the plot

[0024] After spectral measurement, 10 representative wheat plants were selected from each plot, and the green leaf area was measured using a leaf area instrument (LI-3000, LI-COR Inc., USA). The leaves were put into envelopes and then into an oven. After blanching at 105 °C for 30 minutes, they were dried to a constant weight at 80 °C, and the dry weight was weighed. The leaf area index was calculated by the dry weight method.

[0025] Furthermore, for a method for estimating the wheat leaf area index during the whole growth period based on pure hyperspectral reflectance proposed by the present invention, the specific steps of step 2 include:

[0026] Step 2-1: Construct a wheat canopy ear segmentation model

[0027] The deep learning model has the ability to automatically extract non-linear complex features from high-dimensional and massive data. In order to accurately and quickly extract the ear proportion from the canopy image, a Unet-Vgg16 ear semantic segmentation model was built in this study to achieve accurate segmentation of the ear organs in the RGB images of wheat fields. The model uses U-net as the backbone and Vgg16 as the main feature extraction network. Multiscale features of the input image are extracted, and all features are fused to achieve semantic segmentation, and accurate pixel classification of ears and background is performed. The model is applicable to RGB images at all stages after wheat heading, and the mean intersection over union (mIoU) and mean pixel accuracy (PA) of the model reach 90.05% and 94.14% respectively.

[0028] Step 2-2: Extract the proportion of canopy soil

[0029] The CAN_EYE software was used to perform supervised classification on the canopy images taken at each growth stage, so as to extract the proportion of soil pixels under the vertical observation angle of wheat, that is, pure soil pixels;

[0030] Step 2-3: Obtain the proportion of canopy leaves

[0031] Based on the fact that the sum of the proportions of the three components of leaves, ears and soil in the image is equal to 1, after obtaining the data of the ear pixel proportion (step 2-1) and the soil pixel proportion (step 2-2) in the canopy image, the proportion of canopy leaf pixels was calculated, that is, pure canopy leaf pixels of wheat.

[0032] Furthermore, for a method for estimating the wheat leaf area index during the whole growth period based on pure canopy hyperspectral reflectance proposed by the present invention, the specific steps of step 3 include:

[0033] Step 3-1: Generate simulated canopy hyperspectral reflectances with different proportions of canopy ears, leaves and soil

[0034] Obtain the measured hyperspectral reflectance of the canopy leaves, canopy wheat ears and soil obtained in step 1 and the corresponding canopy ear-leaf-soil ratio extracted in step 2, simulate the linear mixed spectra of plots with different ear-leaf-soil ratios, generate the simulated canopy spectral reflectance through equation (1), and denote it as V 模拟 ;

[0035] V simulated = f leaf ×R can + f panicle ×R panicle + f soil ×R soil #(1)

[0036] V 模拟 is the simulated canopy hyperspectral reflectance; R can is the measured hyperspectral reflectance of the canopy leaves; R panicle is the measured hyperspectral reflectance of the canopy wheat ears; R soil is the measured hyperspectral reflectance of the soil; f leaf 、f panicle and f soil are the ratios of the canopy leaf, wheat ear and soil components respectively, and the sum of the three is 1;

[0037] Step 3-2: Calculate the non-negative matrix NMF coefficient H

[0038] Obtain the key coefficient H through equation (2);

[0039] V = WH + N#(2)

[0040] V is the canopy hyperspectral reflectance matrix; H is a coefficient matrix; W is the number of sources, which is 2 before heading and 3 after heading; N is the noise.

[0041] Step 3-3: Obtain the hyperspectral reflectance of the pure canopy leaves

[0042] Input the measured canopy hyperspectral reflectance V 实测 , combine with the coefficient H obtained in step 3-2, iterate 100 times according to equation (2), and output the matrix W, which contains the pure hyperspectral reflectance of the ears, leaves and soil.

[0043] Furthermore, for a method for estimating the wheat leaf area index during the whole growth period based on the pure canopy hyperspectral reflectance proposed by the present invention, the specific steps of step 4 include:

[0044] Twelve vegetation indices commonly used in LAI estimation were selected to compare and evaluate the accuracy of modeling based on the original canopy hyperspectral reflectance and the canopy pure hyperspectral reflectance constructed by this method. The data from 2022 - 2023 were used as the modeling dataset, and the data from 2021 - 2022 were used as the independent validation set. An LAI estimation model for the entire growth period was constructed based on the simple linear regression algorithm, using the coefficient of determination R 2 and the root mean square error RMSE as the evaluation indices for the estimation model.

[0045] Compared with the prior art, the present invention adopting the above technical solutions has the following technical effects:

[0046] 1. A method for estimating the wheat leaf area index for the entire growth period based on the canopy pure hyperspectral reflectance of the present invention has the characteristics of low cost and high efficiency in obtaining the proportion of ears and leaves in the temporal canopy.

[0047] 2. A method for estimating the wheat leaf area index for the entire growth period based on the canopy pure hyperspectral reflectance of the present invention analyzes the influence of different ear proportions on the canopy reflectance, and clarifies that the ear layer will affect the inversion accuracy of LAI after wheat heading.

[0048] 3. A method for estimating the wheat leaf area index for the entire growth period based on the canopy pure hyperspectral reflectance of the present invention. The vegetation indices calculated based on this method are applicable to the estimation of wheat LAI for the entire growth period, especially can significantly improve the estimation accuracy of LAI after heading.

[0049] Explanation of the attached figures

[0050] Figure 1 is the technical flow chart of a method for estimating the wheat leaf area index for the entire growth period based on the canopy pure hyperspectral reflectance of the present invention.

[0051] Figure 2 is a schematic diagram of the correlation between the vegetation index calculated based on the original canopy hyperspectral reflectance and LAI under different ear proportions in the canopy.

[0052] Figure 3 is a comparison chart of the linear fitting results of the vegetation index calculated based on the original canopy hyperspectral reflectance and the canopy pure leaf hyperspectral reflectance extracted by the method of the present invention before (a - c) and after (d - f) heading with LAI.

[0053] Figure 4 is a comparison chart of the validation accuracy of the LAI estimation model for the entire growth period constructed based on the vegetation indices calculated from the original canopy hyperspectral reflectance and the canopy pure leaf hyperspectral reflectance extracted by the method of the present invention. Detailed implementation manners

[0054] The embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as a limitation to the present invention.

[0055] The implementation of the present invention is based on multi-variety plot trials, with each plot being 3m × 20m in size. It includes a total of 2 field trials, involving 2 nitrogen application levels and 16 wheat varieties. The specific experimental design is shown in Table 1 and the sampling time is shown in Table 2.

[0056] Table 1 Wheat field trial design

[0057]

[0058] Table 2 Data acquisition dates

[0059]

[0060]

[0061] As Figure 1 shown, a method for estimating the wheat leaf area index during the whole growth period based on the pure canopy hyperspectral reflectance specifically includes the following steps:

[0062] Step 1: Obtain the measured LAI, canopy RGB images, and hyperspectral reflectance in the field during the tillering-filling period

[0063] Step 1-1: Measure the canopy hyperspectral reflectance

[0064] Use a FieldSpec Pro FR2500 type backpack-mounted field hyperspectral radiometer to measure the vertical reflectance spectrum of the wheat canopy. When measuring the spectrum, the distance between the spectrometer probe and the canopy vertical height is about 1m. Before each measurement, a standard whiteboard correction is performed. Three points are selected for measurement in each plot, and three spectral reflectances are recorded at each point, and the average value is used as the canopy spectral reflectance of the plot.

[0065] Step 1-2: Measure the hyperspectral reflectance of the ear in the canopy

[0066] The spectral reflectance of wheat ears in the canopy was measured using a FieldSpec Pro FR2500 back-mounted field hyperspectral radiometer at the heading stage, flowering stage, early filling stage, and mid-filling stage. The spectral measurements were carried out at 10:00 - 14:00 on sunny, cloudless or slightly cloudy days. The field of view angle of the instrument was 25°. Standard whiteboard calibration was performed before the measurement. Twenty representative wheat main stems were selected from each plot. The wheat ears were cut and held together in the hand. The spectrometer probe was placed directly above the wheat ears to ensure that the field of view of the instrument was entirely covered by the wheat ears, and the spectral reflectance of the wheat ears was measured.

[0067] Steps 1 - 3: Measuring the hyperspectral reflectance of canopy leaves

[0068] After the wheat headed, the spectral reflectance of the original wheat canopy was first measured using a FieldSpec Pro FR2500 spectrometer. Then, all the wheat ears within the field of view of the instrument were cut off, and the spectral reflectance of the wheat canopy after removing the ear layer was measured again. The measurement requirements were the same as those for the vertical reflectance of the canopy.

[0069] Steps 1 - 4: Measuring the hyperspectral reflectance of soil

[0070] The height of the spectrometer probe from the ground was approximately 0.1 m. Standard whiteboard calibration was performed before each measurement. One piece of bare soil was selected from each plot for measurement. Three spectral reflectances were recorded at each point, and the average value was used as the soil spectral reflectance of the plot.

[0071] Steps 1 - 5: Taking RGB images of the wheat canopy

[0072] The RGB images of the canopy were acquired using an iphone12 smartphone. The camera settings were as follows: focal length, 4 mm; aperture value, f / 1.6; ISO speed, ISO - 32; auto exposure. The image acquisition was carried out at 16:00 - 17:00 on the afternoon of the day of the field spectral test. When taking the pictures, the phone was located 1 m above the wheat canopy, and vertical - angle canopy images were taken with a resolution of 3024×3024. Canopy images of wheat in each plot were obtained under natural environmental conditions at the tillering stage, jointing stage, booting stage, heading stage, flowering stage, and early, middle, and late filling stages of wheat respectively.

[0073] Steps 1 - 6: Measuring the LAI of the plot

[0074] After the spectral measurement, ten representative wheat plants were selected from each plot. The green leaf area was measured using a leaf area instrument (LI - 3000, LI - COR Inc., USA). The leaves were put into an envelope and then into an oven. After blanching at 105°C for 30 minutes, they were dried to a constant weight at 80°C, and the dry weight was measured. The leaf area index was calculated using the dry weight method.

[0075] Step 2: Extract pure pixel images of wheat canopy ears, leaves, and soil based on canopy RGB images combined with deep learning algorithms. Step 2-1: Construct a wheat canopy ear segmentation model

[0076] Deep learning models have the ability to automatically extract non-linear complex features from high-dimensional and massive data. To accurately and quickly extract the ear proportion from canopy images, this study built a Unet-Vgg16 ear semantic segmentation model to achieve accurate segmentation of ear organs in wheat field RGB images. The model uses U-net as the backbone and Vgg16 as the main feature extraction network. It extracts multi-scale features of the input image, fuses all features to achieve semantic segmentation, and precisely classifies pixels of ears and the background. The model is applicable to RGB images at all stages after wheat heading, and the average intersection over union (mIoU) and average pixel accuracy (PA) of the model reach 90.05% and 94.14% respectively.

[0077] Step 2-2: Extract the proportion of canopy soil

[0078] The CAN_EYE software is used to perform supervised classification on the canopy images taken at each growth stage, so as to extract the proportion of soil pixels under the vertical observation angle of wheat, that is, pure soil pixels;

[0079] Step 2-3: Obtain the proportion of canopy leaves

[0080] Based on the fact that the sum of the proportions of the three components of leaves, ears, and soil in the image is equal to 1, after obtaining the data of the ear pixel proportion (Step 2-1) and the soil pixel proportion (Step 2-2) in the canopy image, the proportion of canopy leaf pixels is calculated, that is, pure leaf pixels of the wheat canopy.

[0081] Step 3: Calculate the hyperspectral reflectance of pure canopy leaves based on the pure pixels obtained in Step 2 and the canopy spectral reflectance data combined with the non-negative matrix factorization (NMF) algorithm

[0082] Step 3-1: Generate simulated canopy hyperspectral reflectance with different proportions of canopy ears, leaves, and soil

[0083] Input the measured hyperspectral reflectance of canopy leaves, canopy ears, and soil obtained in Step 1 and the corresponding proportions of canopy ear-leaf-soil extracted in Step 2, simulate the linear mixed spectra of plots with different ear-leaf-soil proportions, and generate the simulated canopy hyperspectral reflectance through Equation (1), denoted as V 模拟 ;

[0084] Vsimulated = f leaf ×R can + f panicle ×R panicle + f soil ×R soil #(1)

[0085] V 模拟 is the simulated hyperspectral reflectance of the canopy; R can is the measured hyperspectral reflectance of the canopy leaves; R panicle is the measured hyperspectral reflectance of the canopy wheat ears; R soil is the measured hyperspectral reflectance of the soil; f leaf 、f panicle and f soil are the proportions of the canopy leaves, wheat ears and soil components respectively, and the sum of the three is 1;

[0086] Step 3-2: Calculate the non-negative matrix NMF coefficient H

[0087] Obtain the key coefficient H through Equation (2);

[0088] V = WH + N #(2)

[0089] V is the canopy hyperspectral reflectance matrix; H is a coefficient matrix; W is the number of sources, which is 2 before heading and 3 after heading; N is the noise.

[0090] Step 3-3: Obtain the hyperspectral reflectance of the pure canopy leaves

[0091] Input the measured canopy hyperspectral reflectance V 实测 , combined with the coefficient H obtained in Step 3-2, and iterate 100 times according to Equation (2) to output the matrix W, which contains the pure hyperspectral reflectance of the ears, leaves and soil.

[0092] Step 4: Calculate the vegetation index based on the hyperspectral reflectance of the pure canopy leaves obtained in Step 3 and construct the LAI estimation model for the whole growth period by combining the simple linear regression algorithm

[0093] Step 4-1: Calculate the vegetation index

[0094] Extract the corresponding bands of the original canopy hyperspectral reflectance and the hyperspectral reflectance of the pure canopy leaves obtained in Step 3-3, and substitute them into the vegetation index calculation formula in Table 3 to calculate the vegetation index based on the original and pure canopy leaf hyperspectral reflectance at each growth stage;

[0095] Table 3 Vegetation indices used in the present invention

[0096]

[0097]

[0098] Note: R 800 ,R 750 ,R 670 ,R 550 and R 450Reflectance at wavelengths of 800 nm, 750 nm, 670 nm, 550 nm, and 450 nm, representing the near-infrared, red-edge, red, green, and blue bands respectively.

[0099] Step 4-2: Correlation analysis

[0100] Extract six sets of original canopy spectral reflectances with wheat ear proportions of 0%, 3%, 5%, 7%, 10%, and 15%, and analyze the correlation coefficient R between existing vegetation indices and LAI 2 Change with the increase of wheat ear proportion.

[0101] Step 4-3: Model construction

[0102] Select the top three vegetation indices with better performance (RDVI, MSAVI, and MTVI2 in this invention), and establish a simple linear regression model for estimating LAI during the whole growth period through Equation (3):

[0103] y = β0 + β1x + ∈ #(3)

[0104] Where y is LAI, x is the vegetation index, β0 is the intercept, β1 is the slope, and ∈ is the error term, representing the part that the model fails to explain. It includes random errors and other factors not included in the model.

[0105] Step 4-4: Model fitting

[0106] Through the ordinary least squares (OLS) method, the parameters β0 and β1 can be estimated. The goal of the least squares method is to minimize the sum of squares of the differences between the measured sample values y and the model predicted values y′, that is:

[0107] minimize∑(y i -(β0 + β1x i )) 2 (4)

[0108] Step 4-5: Accuracy verification

[0109] Use the coefficient of determination R 2 and the root mean square error RMSE as evaluation indicators for the estimation model. R 2 and RMSE calculation formulas are as follows:

[0110]

[0111] Where SSE and SST are the sum of squared errors and the total sum of squares respectively; y i and y′ i are the measured sample values and the predicted sample values respectively, is the mean of the measured samples, is the mean of the predicted samples; n is the total number of sample data.

[0112] The following is a comparison of the performance of the proposed method and the vegetation index constructed by the original canopy hyperspectral reflectance in estimating LAI during the whole growth period. The results show that as the proportion of wheat ears in the canopy increases, the correlation between the vegetation index calculated based on the original canopy hyperspectral reflectance and LAI has been decreasing ( Figure 2 ); The three vegetation indices with the best correlation with LAI, RDVI, MSAVI and MTVI2, were selected, and the correlation between the original canopy hyperspectral reflectance and the canopy pure hyperspectral reflectance calculation method and LAI before and after heading was compared ( Figure 3 ), and finally compared the validation scatter plots of the LAI estimation models constructed by each method ( Figure 4 ).from Figure 3 and Figure 4 It can be seen that the method of the present invention can improve the estimation accuracy of LAI after heading, while taking into account the elimination of the influence of soil background before heading, and improving the adaptability of LAI estimation in the whole growth period.

[0113] The above are only some embodiments of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements can be made without departing from the principles of the present invention. These improvements should be within the scope of protection of the present invention.

Claims

1. A method for estimating wheat leaf area index throughout its growth period based on pure hyperspectral reflectance of the canopy, characterized in that The following steps are involved: Step 1: Obtain the field-measured leaf area index LAI, canopy RGB image and hyperspectral reflectance during the tillering-grain filling period; Step 2: Extract pure pixels of wheat canopy ears, leaves and soil based on canopy RGB images combined with deep learning algorithms; Step 3: Based on the pure pixel and canopy spectral reflectance data obtained in step 2, the canopy pure leaf hyperspectral reflectance is calculated using the non-negative matrix factorization (NMF) algorithm; Step 4: Based on the canopy pure leaf hyperspectral reflectance obtained in step 3, the vegetation index is calculated and combined with a simple linear regression algorithm to construct a full growth period LAI estimation model.

2. The method according to claim 1, characterized in that The specific steps for obtaining field-measured LAI, canopy RGB images, and hyperspectral reflectance data during the tillering-grain filling period in step 1 include: Step 1-1: Measure canopy hyperspectral reflectance The vertical reflectance spectrum of wheat canopy was measured. A standard white plate calibration was performed before each measurement. Three points were selected in each 3m×20m plot for measurement. Three spectral reflectances were recorded at each point, and the average value was taken as the canopy spectral reflectance of the plot. Step 1-2: Measure the hyperspectral reflectance of wheat ears in the canopy The spectral reflectance of wheat ears in the canopy was measured at the heading stage, flowering stage, early grain filling stage, and mid grain filling stage. Standard white plate calibration was performed before measurement. Steps 1-3: Measure canopy leaf hyperspectral reflectance The hyperspectral reflectance of the canopy leaves before heading is the canopy hyperspectral reflectance, which is the same as step 1-1; the hyperspectral reflectance of the canopy after heading and after heading after removing the ear layer is the canopy leaf hyperspectral reflectance; Steps 1-4: Measure soil spectral reflectance The soil spectral reflectance was measured three times on the bare soil in each plot, and the average value was taken as the soil spectral reflectance of the plot; Step 1-5: Take RGB images of wheat canopy The wheat canopy images were obtained in the natural environment of each plot at the tillering stage, jointing stage, booting stage, heading stage, flowering stage, and before, middle and late stages of grain filling. Step 1-6: Measure the cell LAI After spectral measurement, 10 representative wheat plants were selected from each plot, and the leaf area index was calculated using the dry weight method.

3. The method according to claim 1, characterized in that The specific steps of extracting pure pixels of wheat canopy ears and leaves in step 2 include: Step 2-1: Construct a canopy wheat ear segmentation model to extract pure pixels of wheat canopy ears Construct the Unet-Vgg16 model, with the canopy RGB image as input and the canopy wheat ear pixel ratio as output, that is, the pure pixels of wheat ears in the wheat canopy; Step 2-2: Extract pure pixels of canopy soil The canopy images taken at each growth period were supervised and classified using the CAN_EYE software to extract the soil pixel ratio at the vertical observation angle of wheat, i.e., the soil pure pixel. Step 2-3: Extract pure wheat leaf pixels Based on the fact that the sum of the proportions of leaves, wheat ears and soil in the image is equal to 1, after obtaining the data of the proportion of wheat ear pixels and the proportion of soil pixels in the canopy image, the proportion of canopy leaf pixels, that is, the pure pixels of wheat canopy leaves, is obtained by calculation.

4. The method according to claim 1, characterized in that The specific steps of step 3 include: Step 3-1: Generate simulated hyperspectral reflectance with different canopy ear, leaf, and soil proportions Input the measured hyperspectral reflectance of canopy leaves, canopy wheat ears and soil obtained in step 1 and the corresponding canopy ear-leaf soil ratio extracted in step 2 to simulate the linear mixed spectrum of plots with different ear-leaf soil ratios. Generate the simulated canopy hyperspectral reflectance through formula (1), denoted as V 模拟 ; V 模拟 =f leaf ×R can +f panicle ×R panicle +f soil ×R soil #(1) V 模拟 is the simulated canopy hyperspectral reflectance; R can is the measured hyperspectral reflectance of canopy leaves; R panicle is the measured high spectral reflectance of the wheat ear canopy; R soil is the measured soil hyperspectral reflectance; f leaf 、f panicle and f soil They are the proportions of canopy leaves, wheat ears and soil components, and the sum of the three is 1; Step 3-2: Calculate the non-negative matrix NMF coefficients H The key coefficient H is obtained by formula (2); V=WH+N#(2) V is the canopy hyperspectral reflectance matrix; H is a coefficient matrix; W is the number of sources, 2 before heading and 3 after heading; N is noise; Step 3-3: Obtaining the hyperspectral reflectance of pure leaves in the canopy Input the measured canopy hyperspectral reflectance V 实测 , combined with the coefficient H obtained in step 3-2, combined with formula (2) iterated 100 times, and output matrix W, which contains the pure hyperspectral reflectance of ears, leaves and soil.

5. The method according to claim 1, characterized in that The specific steps of step 4 include: Step 4-1: Calculate vegetation index The corresponding bands of the original canopy hyperspectral reflectance and the canopy pure leaf hyperspectral reflectance are extracted, and the vegetation index based on the original and canopy pure leaf hyperspectral reflectance is calculated in each growth period; Step 4-2: Correlation Analysis Six original canopy spectral reflectances with canopy wheat ear proportions of 0%, 3%, 5%, 7%, 10% and 15% were extracted, and the correlation between the existing vegetation index and LAI was analyzed. 2 Changes with increasing proportion of wheat ears; Step 4-3: Model construction The first three vegetation indices with better performance were selected, and a simple linear regression model for estimating LAI during the whole growth period was established using formula (3): y=β0+β1x+∈#(3) Among them, y is LAI, x is vegetation index, β0 is intercept, β1 is slope; ∈ is the error term, which represents the part that the model cannot explain, which includes random errors and other factors not included in the model; Step 4-4: Model fitting The parameters β0 and β1 are estimated by the least squares method OLS. The goal of the least squares method is to minimize the difference between the measured sample value y and the model prediction value. The square difference between them, that is: minimize∑(y i -(β0+β1x i )) 2 (4) Step 4-5: Accuracy Verification Using the coefficient of determination R 2 and root mean square error RMSE as the evaluation indicators of the estimation model; R 2 And the RMSE calculation formula is as follows: Among them, SSE and SST are the cumulative sum of squares and the total sum of squares respectively; y i and y′ i are the measured sample value and the predicted sample value respectively, is the mean of the measured samples, is the mean of the predicted sample; n is the total number of sample data.

6. The method according to claim 5, characterized in that The vegetation index is calculated using the following table: Among them, R 800 , R 750 , R 670 , R 550 and R 450 It refers to the reflectivity of the bands at 800nm, 750nm, 670nm, 550nm and 450nm, representing the near infrared, red edge, red, green and blue bands respectively.

7. The method according to claim 5, characterized in that The three vegetation indices selected in step 4-3 are: RDVI, MSAVI and MTVI2.

Images