阅读说明：本技术 基于逆高斯过程反演叶面积指数的农作物长势监测方法 (Crop growth monitoring method based on inversion of leaf area index by inverse Gaussian process ) 是由 黄健熙 尹峰 于 2019-08-26 设计创作，主要内容包括：本发明属农业遥感领域,涉及基于逆高斯过程反演叶面积指数的农作物长势监测方法,其为：获一致性哨兵2号地表反射率,提取作物空间分布图；获不同背景下PROSAIL模拟作物光谱并转为波段光谱；对黑土壤背景下模拟地表反射率主成分分析得纯冠层反射特征；对实测土壤光谱反射率主成分分析得纯土壤反射特征；线性分解含土壤信息的模拟地表反射率得模拟纯冠层反射率；高斯过程模拟学习模拟纯冠层反射率与LAI映射关系得训练好的高斯过程模型；线性分解哨兵2号影像得纯冠层反射率,选与其最接近的模拟纯冠层反射率输入训练好的高斯过程模型得到LAI,以往年平均LAI轨迹为基准判断当前作物长势。本发明的方法规避多参同效问题,精度较高。(The invention belongs to the field of agricultural remote sensing, and relates to a crop growth monitoring method based on inverse Gaussian process inversion leaf area index, which comprises the following steps: obtaining the surface reflectivity of the sentinel No. 2 with consistency, and extracting a crop space distribution map; acquiring PROSAIL simulated crop spectra under different backgrounds and converting the spectra into waveband spectra; analyzing the main components of the simulated earth surface reflectivity under the black soil background to obtain the pure canopy reflection characteristics; analyzing the main components of the actually measured soil spectral reflectivity to obtain the pure soil reflection characteristics; linearly decomposing the simulated earth surface reflectivity containing soil information to obtain the simulated pure canopy reflectivity; simulating a mapping relation between pure canopy reflectivity and LAI by Gaussian process simulation learning to obtain a trained Gaussian process model; linearly decomposing the sentinel No. 2 image to obtain pure canopy reflectivity, inputting the simulated pure canopy reflectivity closest to the pure canopy reflectivity into a trained Gaussian process model to obtain LAI, and judging the current crop growth by taking the annual average LAI track as a reference. The method provided by the invention avoids the problem of multiple parameter and effect, and has high precision.)

1. A crop growth monitoring method based on inverse Gaussian process inversion leaf area index is characterized by comprising the following specific steps:

s1, obtaining the consistent ground surface reflectivity of the sentinel No. 2 by using a sensor indifference atmospheric correction method SIAC;

s2, extracting each key growth period of the crop to be detected by using the time sequence data of the sentinel No. 2 surface reflectivity corresponding to the sample point, and performing supervision and classification to obtain a crop space distribution map in a research area;

S3, carrying out extensive sampling aiming at a PROSAIL model parameter set corresponding to crops to obtain a PROSAIL simulated crop spectrum under the background of black soil and a PROSAIL simulated crop spectrum under the background containing soil information;

s4, converting the crop spectrum simulated in S3 into a wave band spectrum corresponding to the sentinel No. 2 by using the spectrum response function of the sentinel No. 2, namely obtaining the surface reflectivity simulated by the PROSAIL under the background of black soil and the surface reflectivity simulated by the PROSAIL under the background containing soil information;

S5, performing principal component analysis on the PROSAIL simulated earth surface reflectivity under the black soil background obtained in the S4, and selecting the first 4 principal components as the spectral reflection characteristics of the pure canopy; meanwhile, performing principal component analysis on the reflectivity of the sentinel No. 2 corresponding to the actually measured soil spectral curve, and selecting the first 2 principal components as the spectral reflection characteristics of the pure soil;

s6, carrying out linear decomposition on the ground surface reflectivity simulated by the PROSAIL under the background containing the soil information obtained in S4 based on the spectral reflection characteristics of the pure canopy and the spectral reflection characteristics of the pure soil obtained in S5 to obtain the pure canopy reflectivity simulated by the PROSAIL;

S7, utilizing the mapping relation between the PROSAIL simulated earth surface reflectivity and the LAI under the black soil background obtained in the Gaussian process simulation learning S4 to realize the prediction from the pure canopy spectral reflectivity to the LAI, and obtaining a well-trained Gaussian process model;

s8, performing linear decomposition on the original sentinel No. 2 image based on the result obtained by the principal component analysis of S5 to obtain the corresponding pure canopy reflectivity; comparing the pure canopy reflectivity simulated by the PROSAIL obtained in the step S6 with the pure canopy reflectivity, and inputting the pure canopy reflectivity simulated by the PROSAIL closest to the pure canopy reflectivity into a Gaussian process model trained in the step S7 to obtain LAI; and taking the average LAI track of the plot in the last 3 years as a reference, and comparing the LAI track in the current year with the LAI track in the current year, thereby judging the growth vigor of the current crops.

2. The crop growth monitoring method for inverting the leaf area index based on the inverse Gaussian process as claimed in claim 1, wherein the specific method of S3 is as follows: acquiring the minimum value and the maximum value of main biophysical parameters of a PROSAIL model corresponding to crops according to a large number of actually measured sample points, randomly and uniformly sampling by using a Latin hypercube sampling method in the range, inputting a large number of sampled parameters into the PROSAIL model, and simulating the spectrum of the crops under different growth conditions to obtain the spectrum of the crops simulated by PROSAIL under the black soil background and the spectrum of the crops simulated by PROSAIL under the background containing soil information.

3. The method for monitoring the growth of crops based on the inversion of leaf area index by the inverse gaussian process as claimed in claim 1 or 2, wherein the prediction of the LAI from the pure canopy spectral feature is realized in step S7 by using the following formula:

f_*And the training sample point y with Gaussian noise has a multivariate Gaussian distribution:

the mean value of the multivariate Gaussian distribution is 0 and the covariance matrix iswherein: f. of_*y is the LAI of the prediction sample point and the training sample point respectively; x_*X is respectively the reflectivity data of the prediction sample point and the training sample point;Is the covariance of the training samples and,is the variance of Gaussian noise, K (X), artificially added to the training sample_*,X_*) Is the covariance of the prediction sample, K (X, X)_*) Is the covariance of the training sample and the prediction sample, representing obedience, N represents the Gaussian distribution;

thus, the conditional distribution probability of the prediction sample is:

calculated according to the equations (3), (4), wherein the meaning of E is desired,Is defined in the sense that it is defined as,^－1Represents the inverse of the matrix, | represents the condition:

let K be K (X, X), K_*＝K(X,X_*) Then predict the predicted mean of the sampleand prediction covariance cov (f)_*) See formula (5), formula (6):

the Gaussian process training is an optimization process to obtain the hyper-parameters that determine the covariance matrix in equation (1), so that the mean of the predicted samples can be calculated using equations (5), (6)And uncertainty prediction covariance cov (f)_*) (ii) a The meaning of T is transposed;

In the Gaussian process, the radial basis function kernel is as follows:

Where σ is the hyper-parameter of the kernel function.

4. The method for monitoring the growth of a crop based on inversion of leaf area index by inverse gaussian process as claimed in claim 1 or 2, wherein said crop is a staple grain crop.

5. the use of the method for monitoring the growth of crops based on inversion of leaf area index by inverse Gaussian process as claimed in any one of claims 1 to 4, which is to guide the agricultural measures of field irrigation, fertilization, plant protection and the like according to the growth monitoring result.

Technical Field

The invention belongs to the field of agricultural remote sensing, and particularly relates to a crop growth monitoring method for inverting a leaf area index based on an inverse Gaussian process.

Background

In current practice, growth monitoring is generally based on NDVI, and is less based on leaf area index LAI, since LAI inversion is much more difficult than NDVI that can be obtained by band calculations alone. However, NDVI only reflects the spectral information of the crops in the red band and the near red band, while LAI reflects the sum of the area of single-sided green leaves on the unit surface area, and the NDVI is closely related to important biophysical processes of the crops such as canopy interception, evapotranspiration, photosynthesis and the like, and can reflect the growth vigor of the crops more comprehensively.

Currently, there are two main methods for acquiring LAI by using remote sensing technology, one is through statistical empirical relationship between canopy optical radiation information and LAI, and the other is through inversion of LAI from canopy optical radiation information by using a radiation transmission model; the first method is lack of universality, an empirical relationship established in one place is difficult to popularize in another place, the second method is very complex in calculation and has a plurality of input parameters, different and identical effects are easy to occur during inversion, and the inversion accuracy of the LAI is reduced.

Disclosure of Invention

in order to solve the problem of multiple parameter and common effects of the traditional method for acquiring the LAI from the inversion of the canopy light radiation information through the radiation transmission model, the invention simulates the radiation transmission model based on the inverse Gaussian process, can quickly invert the LAI, does not need ground actual measurement data in the inversion, and can also avoid different parameter and common effects. By comparing LAI tracks of the growth of crops in the past year, the growth vigor of the current crops can be monitored in time.

the invention provides a crop growth monitoring method based on inverse Gaussian process inversion leaf area index, which comprises the following specific steps:

s1, obtaining the consistent ground surface reflectivity of the sentinel No. 2 by using a sensor indifference atmospheric correction method SIAC;

S2, extracting each key growth period of the crop to be detected by using the time sequence data of the sentinel No. 2 surface reflectivity corresponding to the sample point, and performing supervision and classification to obtain a crop space distribution map in a research area;

S3, carrying out extensive sampling aiming at a PROSAIL model parameter set corresponding to crops to obtain a PROSAIL simulated crop spectrum under the background of black soil and a PROSAIL simulated crop spectrum under the background containing soil information;

S4, converting the crop spectrum simulated in S3 into a wave band spectrum corresponding to the sentinel No. 2 by using the spectrum response function of the sentinel No. 2, namely obtaining the surface reflectivity simulated by the PROSAIL under the background of black soil and the surface reflectivity simulated by the PROSAIL under the background containing soil information;

S5, performing principal component analysis on the PROSAIL simulated earth surface reflectivity under the black soil background obtained in the S4, and selecting the first 4 principal components as the spectral reflection characteristics of the pure canopy; meanwhile, performing principal component analysis on the reflectivity of the sentinel No. 2 corresponding to the actually measured soil spectral curve, and selecting the first 2 principal components as the spectral reflection characteristics of the pure soil;

s6, carrying out linear decomposition on the ground surface reflectivity simulated by the PROSAIL under the background containing the soil information obtained in S4 based on the spectral reflection characteristics of the pure canopy and the spectral reflection characteristics of the pure soil obtained in S5 to obtain the pure canopy reflectivity simulated by the PROSAIL;

S7, utilizing the mapping relation between the PROSAIL simulated earth surface reflectivity and the LAI under the black soil background obtained in the Gaussian process simulation learning S4 to realize the prediction from the pure canopy spectral reflectivity to the LAI, and obtaining a well-trained Gaussian process model;

s8, performing linear decomposition on the original sentinel No. 2 image based on the result obtained by the principal component analysis of S5 to obtain the corresponding pure canopy reflectivity; comparing the pure canopy reflectivity simulated by the PROSAIL obtained in the step S6 with the pure canopy reflectivity, and inputting the pure canopy reflectivity simulated by the PROSAIL closest to the pure canopy reflectivity into a Gaussian process model trained in the step S7 to obtain LAI; and taking the average LAI track of the plot in the last 3 years as a reference, and comparing the LAI track in the current year with the LAI track in the current year, thereby judging the growth vigor of the current crops.

In step S1, the SIAC is a method of the invention, the code of which is disclosed in the GitHub platform, and the website is https: com/MarcYin/SIAC.

the specific method of S3 is as follows: acquiring the minimum value and the maximum value of main biophysical parameters of a PROSAIL model corresponding to crops according to a large number of actually measured sample points, randomly and uniformly sampling by using a Latin hypercube sampling method in the range, inputting a large number of sampled parameters into the PROSAIL model, and simulating the spectrum of the crops under different growth conditions to obtain the spectrum of the crops simulated by PROSAIL under the black soil background and the spectrum of the crops simulated by PROSAIL under the background containing soil information.

The black soil background in S3 means that the simulated spectrum does not include soil information, and the spectrum in the black soil background is the spectrum of the pure canopy.

the prediction of the pure canopy spectral features to LAI is realized as described in step S7, using the following formula:

f_*and the training sample point y with Gaussian noise has a multivariate Gaussian distribution:

The mean value of the multivariate Gaussian distribution is 0 and the covariance matrix is

Wherein: f. of_*Y is the LAI of the prediction sample point and the training sample point respectively; x_*X is respectively the reflectivity data of the prediction sample point and the training sample point;is the covariance of the training samples and,Is the variance of Gaussian noise, K (X), artificially added to the training sample_*，X_*) Is the covariance of the prediction sample, K (X, X)_*) Is the covariance of the training sample and the prediction sample, representing obedience, N represents the Gaussian distribution;

thus, the conditional distribution probability of the prediction sample is:

Calculated according to the equations (3), (4), wherein the meaning of E is desired,Is defined in the sense that it is defined as,^-1represents the inverse of the matrix, | represents the condition:

Let K be K (X, X), K_*＝K(X，X_*) Then predict the predicted mean of the sampleAnd prediction covariance cov (f)_*) See formula (5), formula (6):

The Gaussian process training is an optimization process to obtain the hyper-parameters that determine the covariance matrix in equation (1), so that the mean of the predicted samples can be calculated using equations (5), (6)And uncertainty prediction covariance cov (f)_*) (ii) a The meaning of T is transposed;

In the Gaussian process, the radial basis function kernel is as follows:

Where σ is the hyper-parameter of the kernel function.

The crop is a staple grain crop selected from any one of wheat, rice, corn and the like.

The invention also provides an application of the crop growth monitoring method for inverting the leaf area index based on the inverse Gaussian process, which is used for guiding agricultural measures such as field irrigation, fertilization, plant protection and the like according to the growth monitoring result.

compared with the prior art, the invention has the beneficial effects that:

1. the mapping from the input parameters to the output of the Prosail model is realized through the inverse Gaussian process, the problem of multi-parameter synchronization is avoided, and the LAI inversion precision is improved.

2. The method overcomes the locality of the traditional machine learning, and the mapping of the inverse Gaussian process can be widely applied to various crops and various growing environments.

3. the biophysical and chemical parameters obtained by using all available wave bands and based on a physical radiation transmission model can more effectively and accurately reflect the growth condition of surface crops, and are greatly superior to the traditional plant index such as NDVI.

Drawings

FIG. 1 is a flow chart of a crop growth monitoring method based on inverse Gaussian process inversion leaf area index.

FIG. 2 is a plot of the LAI output obtained by the inversion in example 1.

FIG. 3 is a graph showing the results of judging the growth vigor of wheat in example 1.

Detailed Description

The following describes in further detail specific embodiments of the present invention with reference to examples. The following examples are intended to illustrate the invention but are not intended to limit the scope of the invention.