CN115759526A - Crop harvest index inversion method based on convolutional neural network and crop model - Google Patents
Crop harvest index inversion method based on convolutional neural network and crop model Download PDFInfo
- Publication number
- CN115759526A CN115759526A CN202211345984.8A CN202211345984A CN115759526A CN 115759526 A CN115759526 A CN 115759526A CN 202211345984 A CN202211345984 A CN 202211345984A CN 115759526 A CN115759526 A CN 115759526A
- Authority
- CN
- China
- Prior art keywords
- crop
- data
- research
- neural network
- convolutional neural
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
- 238000000034 method Methods 0.000 title claims abstract description 56
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 39
- 238000003306 harvesting Methods 0.000 title claims abstract description 38
- 230000005855 radiation Effects 0.000 claims abstract description 23
- 239000002028 Biomass Substances 0.000 claims abstract description 11
- 230000009467 reduction Effects 0.000 claims abstract description 11
- 239000002352 surface water Substances 0.000 claims abstract description 8
- 238000001914 filtration Methods 0.000 claims abstract description 5
- 238000011160 research Methods 0.000 claims description 58
- 239000011159 matrix material Substances 0.000 claims description 25
- 101000583150 Homo sapiens Membrane-associated phosphatidylinositol transfer protein 3 Proteins 0.000 claims description 14
- 102100030351 Membrane-associated phosphatidylinositol transfer protein 3 Human genes 0.000 claims description 14
- DZSYJVXGONVNKA-UHFFFAOYSA-L NIR-1 dye Chemical compound [K+].[K+].C1=CC2=C(S([O-])(=O)=O)C=C(S([O-])(=O)=O)C=C2C(C2(C)C)=C1[N+](CC)=C2C=CC=CC=CC=C1C(C)(C)C2=CC(C(O)=O)=CC=C2N1CCCCS([O-])(=O)=O DZSYJVXGONVNKA-UHFFFAOYSA-L 0.000 claims description 14
- 238000002310 reflectometry Methods 0.000 claims description 8
- 238000012549 training Methods 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 7
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 7
- 238000007781 pre-processing Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 210000000056 organ Anatomy 0.000 claims description 4
- 238000003860 storage Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000015572 biosynthetic process Effects 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 230000004907 flux Effects 0.000 claims description 3
- 230000000630 rising effect Effects 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 239000002689 soil Substances 0.000 claims description 3
- 238000003786 synthesis reaction Methods 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 2
- 238000012544 monitoring process Methods 0.000 abstract description 3
- 238000009825 accumulation Methods 0.000 abstract description 2
- 238000004364 calculation method Methods 0.000 description 5
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 230000004927 fusion Effects 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
Images
Landscapes
- Image Processing (AREA)
Abstract
The embodiment of the invention discloses a crop harvest index inversion method based on a convolutional neural network and a crop model, which is characterized in that the method selects remote sensing indexes related to crop biomass accumulation and based on MODIS data-crop absorbed photosynthetically active radiation APAR and daytime surface temperature LST D Night surface temperature LST N And the land surface water index LSWI, and performing histogram dimensionality reduction to be used as an input layer of the convolutional neural network; assimilating a crop growth model WOFOST by using LAI data of MODIS through an ensemble Kalman filtering method, and realizing inversion in a large range through a small amount of ground calibration work so as to obtain more HI samples as an output layer; the model is trained and optimized through the convolutional neural network, so that the large-range and efficient remote sensing monitoring of the crop harvest index is realized.
Description
Technical Field
The embodiment of the invention relates to the technical field of crop yield estimation of agricultural satellite remote sensing, in particular to a crop harvest index inversion method based on a convolutional neural network and a crop model.
Background
Crop yield can be understood from the point of carbon assimilation as the fraction of total net accumulated biomass that is converted into storage organs. Can be expressed as: yield = NPP HI. In the formula, NPP generally refers to the net accumulated biomass of a crop, HI is the harvest index, and Yield per unit area.
The current remote sensing inversion method of NPP is mature, but problems exist when NPP is used for estimating yield, 1) HI is set as a constant, namely final total NPP is multiplied by a fixed constant, but crop growth needs to be influenced by multiple periods such as day and night, seasons and the like; 2) When a crop growth model such as WOFOST is directly used for simulation, a large number of parameters are needed to be calibrated, and the large-range inversion efficiency is low. Campoy et al performed exponential function curve fitting on HI using remote sensing indicators related to crop biomass in 2020 to obtain better correlation. The convolutional neural network has excellent nonlinear fitting capability, and the traditional harvest index acquisition method needs a large amount of ground measured data and consumes a large amount of manpower and material resources.
Disclosure of Invention
Therefore, the embodiment of the invention provides a crop harvest index inversion method based on a convolutional neural network and a crop model, so as to solve the technical problems, and thus, accurate and efficient remote sensing monitoring for judging the crop harvest index is realized.
In order to achieve the above object, an embodiment of the present invention provides the following:
in a first aspect of embodiments of the present invention, there is provided a method of crop harvest index inversion based on a convolutional neural network and a crop model, comprising the steps of:
s1, selecting a research area and research crops, and performing parameter calibration on a crop growth model WOFOST by using the data of the agricultural gas station, wherein the parameters comprise crop parameters, soil parameters, field management parameters and the like;
s2, downloading MODIS data in the crop growth period, and preprocessing to obtain LST D 、LST N Taking the LSWI index and the APAR index as features to be input;
and S3, taking the NDVI average value of two phenological nodes of the research crop at the beginning and the end of the three-year growth period of the research crop in the research area as a judgment threshold value of the corresponding node according to the data record of the agricultural gas station.
S4, screening all images in the growing period of each block crop according to the threshold judgment method in the S3, performing histogram dimensionality reduction, and generating a matrix of the characteristic images to be input to serve as an input layer of a convolutional neural network;
s5, performing data assimilation on the WOFOST model by using MODIS LAI products in a research area and a research crop growth period, outputting simulated aboveground biomass and yield data of crops corresponding to the block level in the step S2 of the research area, and calculating a harvest index to serve as an output layer of a convolutional neural network;
and S6, setting structural parameters of the convolutional neural network, and training and verifying the model.
Furthermore, the MODIS data downloading is to use the crop distribution vector data as a mask to extract and download images of the whole growth cycle of the research area and the research crops, and the images include data of 5 wave bands, namely the daytime and nighttime surface temperature LST of MYD11A2 respectively D 、LST N NDVI data of MOD13A1, surface reflectance data of NIR1 wave band and surface reflectance data of SWIR1 wave band are preprocessed to obtain a surface water index LSWI and photosynthetically active radiation APAR.
Further, the preprocessing process comprises the steps of converting downloaded data into UTM projection based on a WGS-84 ellipsoid, setting the uniform resolution of the data to be 500m, and extracting, splicing and cutting the remote sensing image according to the region where crops are located; and calculating a surface water index LSWI by using NIR1 and SWIR1, calculating a ratio vegetation index SR by using NDVI data, and calculating photosynthetically active radiation APAR absorbed by the crops by using NDVI and SR in combination with meteorological data.
Further, the calculation formula of the land water index LSWI is as follows:
LSWI＝(ρ NIR1 –ρ SWIR1 )/(ρ NIR1 +ρ SWIR1 )
where ρ is NIR1 Reflectivity of near infrared band (841-875 nm), rho SWIR1 The reflectivity of the short wave infrared band is 1628-1652 nm.
Further, the calculation formula of the photosynthetically active radiation APAR is as follows:
APAR(x,t)＝SOL(x,t)×FPAR(x,t)×0.5
wherein SOL is solar radiant flux with unit of MJ/m 2 0.5 is the proportion of photosynthetically active radiation in the total solar radiation and FPAR is the proportion of photosynthetically active radiation absorbed by the crop.
Further, the formula for fitting the NDVI time sequence of the research region research crop in the whole year NDVI of the research region research crop in the research year by using the Beck-Double Logistic method is as follows:
f(t)＝a base +(a max -a base )×[(1/(1+e -m1×(t-m2) ))+1/(1+e n1×(t-n2) )]
where t is the day t of the year, i.e., DOY, f (t) refers to the NDVI value by Beck method fitting at time t; a is base Refers to the plot base NDVI value before crop emergence; a is max Is the maximum value of NDVI in a year; m2 is an inflection point of the rising stage, and n2 is an inflection point of the falling stage; m1 and n1 are slope values at points m2, n 2.
Further, the step S3 of obtaining the growth period corresponding to the crop in the current year by using the threshold value adopts histogram reduction and normalization, and the specific method includes:
determining the time sequence number of remote sensing images of research year, research district block level and research crop growth period, namely N period total images,
and fusing the N images of the crops in the growth period on a time sequence to form an N x N matrix, and finally generating an N x N4 matrix as an input layer of the convolutional neural network.
Further, the specific method for simulating the crop harvest index by the wobest and constructing the output layer of the convolutional neural network in the step S4 is as follows:
downloading day-by-day meteorological data of a research area, wherein the day-by-day meteorological data comprise the highest air temperature and the lowest air temperature of each day, the daily total rainfall, the average water vapor pressure, the sunlight radiation quantity and the daily average wind speed, and interpolating the meteorological data by using ANUSPLIN to form 500m grid data with the resolution equal to that of the downloaded remote sensing image;
utilizing MODIS LAI data, and carrying out data assimilation on a crop growth model by an EnKF ensemble Kalman filtering method, wherein the formula is as follows:
Aa＝Af+K*(Dt-HA)
K＝Ac*HT*(HAc*HT+Dc)-1,K∈Rn*N
wherein A is a To analyze the matrix, A f Is a prediction matrix, K is an ensemble Kalman gain coefficient, D t Is a matrix of observed variables, D c Is a covariance matrix of the observed variables, H is a nonlinear operator, A is from a prediction equation, and HA is an error covariance matrix of the observed variables;
and (4) carrying out time-period-by-time-period synthesis on the output result of the WOFOST according to the time sequence scale and the spatial position corresponding to the histogram in the step (3), and calculating a harvest index HI.
Further, the harvest index HI is calculated by the formula: HI = Yield/AGB
Wherein HI is the harvest index, YIeld is the amount of dry matter that net accumulates in the storage organ simulated by WOFOST, and AGB is the total biomass of the upper part of the crop field simulated by WOFOST; constructing a sample for each land parcel, generating a harvest index sample library as an output layer of the convolutional neural network, selecting 2/3 as a training set, and selecting 1/3 as a verification set.
Further, the specific steps of constructing the convolutional neural network in step S6 are:
performing batch normalization and Relu function activation on each convolution layer, and adding a Dropout layer into the full-connection layer;
and (4) adopting the decision coefficient and the root mean square error as evaluation indexes, and calculating the harvest index through a convolutional neural network in the same time interval.
According to the embodiment of the invention, compared with the prior art, the inversion method has the following advantages: the patent selects remote sensing index related to crop biomass accumulation based on MODIS data-photosynthetic effective radiation APAR absorbed by crops and daytime surface temperature LST D Surface temperature LST at night N And the land surface water index LSWI, and performing histogram dimensionality reduction to be used as an input layer of the convolutional neural network; assimilating a crop growth model WOFOST by using LAI data of MODIS through an ensemble Kalman filtering method, and realizing inversion in a large range through a small amount of ground calibration work so as to obtain more HI samples as an output layer; the model is trained and optimized through the convolutional neural network, so that large-range and efficient remote sensing monitoring of the crop harvest index is realized.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below. It should be apparent that the drawings in the following description are merely exemplary, and that other embodiments can be derived from the drawings provided by those of ordinary skill in the art without inventive effort.
The structures, ratios, sizes, and the like shown in the present specification are only used for matching with the contents disclosed in the specification, so as to be understood and read by those skilled in the art, and are not used to limit the conditions that the present invention can be implemented, so that the present invention has no technical significance, and any structural modifications, changes in the ratio relationship, or adjustments of the sizes, without affecting the effects and the achievable by the present invention, should still fall within the range that the technical contents disclosed in the present invention can cover.
FIG. 1 is a flow chart of a method for crop harvest index inversion based on a convolutional neural network and a crop model according to an embodiment of the present invention;
fig. 2 is a structure route diagram of an implementation of the crop harvest index inversion method based on the convolutional neural network and the crop model according to the embodiment of the present invention.
Detailed Description
The present invention is described in terms of particular embodiments, other advantages and features of the invention will become apparent to those skilled in the art from the following disclosure, and it is to be understood that the described embodiments are merely exemplary of the invention and that it is not intended to limit the invention to the particular embodiments disclosed. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In the present specification, the terms "upper", "lower", "left", "right", "middle", and the like are used for clarity of description, and are not intended to limit the scope of the present invention, and changes or modifications in the relative relationship may be made without substantial changes in the technical content.
As shown in fig. 1, which illustrates a method for inverting a crop harvest index based on a convolutional neural network and a crop model according to an embodiment of the present invention, and is shown in fig. 2, the method includes the following steps:
s1, selecting a research area and research crops, and performing parameter calibration on a crop growth model WOFOST by using the station data of the agricultural gas, wherein the parameters comprise crop parameters, soil parameters, field management parameters and the like;
s2, downloading MODIS (spectral sensing instrument) data in the crop growth period, and preprocessing the data;
wherein, the MODIS data download is to extract and download images of the whole growth cycle of research areas and research crops by using crop distribution vector data as a mask, and the data comprises 5 wave bands of data, namely the daytime and nighttime surface temperature LST of MYD11A2 D 、LST N MOD13A1 NDVI data, NIThe R1 wave band earth surface reflectivity data and the SWIR1 wave band earth surface reflectivity data are preprocessed to obtain the earth surface temperature LST in the daytime and at night D 、LST N LSWI and APAR surface water index LSWI and photosynthetically active radiation APAR are used as the characteristics to be input.
The preprocessing process comprises the steps of converting downloaded data into UTM projection based on a WGS-84 ellipsoid, setting the unified resolution of the data to be 500m, and extracting, splicing and cutting the wave bands of the remote sensing images according to the plot where crops are located; and calculating a surface water index LSWI by using NIR1 and SWIR1, calculating a ratio vegetation index SR by using NDVI data, and calculating photosynthetically active radiation APAR absorbed by the crops by using NDVI and SR in combination with meteorological data.
The calculation formula of the land water index LSWI is as follows:
LSWI＝(ρ NIR1 –ρ SWIR1 )/(ρ NIR1 +ρ SWIR1 )
in the formula, ρ NIR1 Reflectivity in near infrared band (841-875 nm), rho SWIR1 The reflectivity is in a short wave infrared band (1628-1652 nm);
the calculation formula of the photosynthetically active radiation APAR is as follows:
APAR(x,t)＝SOL(x,t)×FPAR(x,t)×0.5
wherein SOL is solar radiant flux with unit of MJ/m 2 0.5 is the proportion of photosynthetically active radiation in the total solar radiation and FPAR is the proportion of photosynthetically active radiation absorbed by the crop.
Specifically, the FPAR is obtained from the normalized vegetation index NDVI and the ratio vegetation index SR, and the formula is:
FPAR(x,t)＝0.5×(FPAR NDVI (x,t)+FPAR SR (x,t))
FPAR NDVI (x,t)＝((NDVI(x,t)–NDVI min )×(FPAR max -FPAR min ))/(NDVI max
-NDVI min )+FPAR min
FPAR SR (x,t)＝((SR(x,t)-SR min )×(FPAR max -FPAR min ))/(SR max -SR min )+FPAR min
SR(x,t)＝(1+NDVI(x,t))/(1–NDVI(x,t))
wherein, FPAR max Taking 0.95 of FPAR min Taking 0.001; NDVI max And NDVI min Respectively 95% and 5% lower percentage, and obtaining SR by the same method max And SR min . Finally generate LST D ,LST N Four indexes, LSWI and APAR;
and S3, taking the NDVI average value of two phenological nodes of the research crop in the research area, which grow for three years and start (SoS) and end (EoS), as the judgment threshold value of the corresponding node according to the data record of the agricultural gas station. Carrying out time sequence curve fitting on NDVI of the research crops in the research region in the whole year by using a Beck-Double Logistic method, and obtaining the corresponding growth period of the crops in the current year by using the threshold;
the formula for fitting the NDVI time sequence of the research crop in the research area by using the Beck-Double Logistic method is as follows:
f(t)＝a base +(a max -a base )×[(1/(1+e -m1×(t-m2) ))+1/(1+e n1×(t-n2) )]
where t is the day t of the year, i.e., DOY, f (t) refers to the NDVI value by Beck method fitting at time t; a is a base Refers to the plot base NDVI value before crop emergence; a is max Is the maximum value of NDVI in a year; m2 is an inflection point of the rising stage, and n2 is an inflection point of the falling stage; m1 and n1 are slope values at points m2, n 2; in the present example, performing nonlinear least square fitting on the NDVI time series data of the plot to obtain each parameter value; and averaging NDVI values corresponding to the beginning of crop growth (SoS) and the end of crop growth (EoS) of the research crop in the research area in three years, and substituting the average value into the fitted curve equation to obtain the growth period time of the plot crop in the research year.
Obtaining the corresponding growth period of the crop in the year by utilizing the threshold value, and performing histogram reduction and normalization treatment, wherein the specific method comprises the following steps:
a. determining the research year, the research region block level and the time sequence number of remote sensing images of the research crop growth period, namely N-period images, performing histogram statistics on the N-period images on a time sequence, determining the dimension reduction range of the histogram by visualizing the change of each wave band on the time sequence, dividing N intervals after determining the dimension reduction range of the histogram, and performing discretization statistics on the number of pixels one by one to generate a pixel histogram;
b. carrying out normalization processing on the generated histogram, wherein the formula is as follows:
[H 1 ,H 2 ,…H N ]＝[h 1 ,h 2 ,…,h N ]/∑N h＝1hi
wherein [ h ] 1 ,h 2 ,…,h N ]Is a pixel histogram generated after dimensionality reduction, [ H ] 1 ,H 2 ,H N ]Is a normalized pixel histogram;
c. time sequence fusion, which comprises the following steps:
LST contained in N images of crops in growth period D ,LST N And performing histogram extraction on the LSWI and APAR 4 wave bands to enable the image of each wave band to generate a vector with the length of N, performing fusion on the time sequence to form an N x N matrix, and finally generating the N x N4 matrix as an input layer of the convolutional neural network.
S4, screening out all N-stage LSTs in the growth period of each block crop according to the threshold judgment method in the step S3 D 、LST N Performing histogram dimensionality reduction on the images of the LSWI and APAR four wave bands to generate an NxNx4 matrix as an input layer of the convolutional neural network;
the specific method for simulating the crop harvest index by using WOFOST and constructing the output layer of the convolutional neural network comprises the following steps:
a. downloading the daily meteorological data of the research area, including the highest temperature and the lowest temperature of each day, the total rainfall, the average water vapor pressure and the average water vapor pressure of each day,
Carrying out interpolation on meteorological data by using ANUSPLIN to form 500m grid data with the resolution equal to that of the downloaded remote sensing image;
b. utilizing MODIS LAI data, and carrying out data assimilation on a crop growth model by an EnKF ensemble Kalman filtering method, wherein the formula is as follows:
Aa＝Af+K*(Dt-HA)
K＝Ac*HT*(HAc*HT+Dc)-1,K∈Rn*N
wherein A is a To analyze the matrix, A f Is the prediction matrix, K is the ensemble Kalman gain coefficient, D t Is a matrix of observed variables, D c The method is a covariance matrix of observed variables, wherein the method refers to an MODIS LAI value, H is a nonlinear operator, A is from a prediction equation, the method refers to simulation of a WOFOST model on LAI in the invention, and HA is an error covariance matrix of the observed variables.
c. And (4) carrying out time-period-by-time-period synthesis on the output result of the WOFOST according to the time sequence scale and the spatial position corresponding to the histogram in the step (3), and calculating a harvest index HI.
In the process, a data assimilation method is used for carrying out regional use expansion on the WOFOST model, the assimilated WOFOST is used for generating a HI sample data set to serve as an output layer, and large-scale actual measurement sampling is avoided.
Wherein, the calculation formula of the harvest index HI is as follows:
HI＝Yield/AGB
wherein HI is the harvest index, YIeld is the amount of dry matter that net accumulates in the storage organ simulated by WOFOST, and AGB is the total biomass of the upper part of the crop field simulated by WOFOST; constructing a sample for each plot, generating a harvest index sample library as an output layer of the convolutional neural network, selecting 2/3 of the samples as a training set, and selecting 1/3 of the samples as a verification set.
S5, interpolating meteorological data into a 500m grid, corresponding to the MODIS image data resolution, performing data assimilation on a WOFOST model by using MODIS LAI products in a research area and a research crop growth period to output simulated aboveground biomass and yield data of the crops corresponding to the block level in the step S2 of the research area, and calculating a harvest index to serve as an output layer of a convolutional neural network;
and S6, setting structural parameters of the convolutional neural network, and training and verifying the model.
In the model training process, in order to meet the requirement of growth cycle in multiple scenes of research areas and research crops, a convolutional neural network is constructedThe method comprises the following specific steps: a. the input layer is a matrix of N × 4, 7 convolution layers are arranged totally, and the number of convolution kernels is as follows in sequence: 64. 64, 128, 256, all with 3 x 3 convolution kernels, the sliding steps are respectively 2, 1, 2, 1 and 2, carrying out 1 all-zero filling on each convolution layer, carrying out batch normalization and Relu function activation on each convolution layer, and adding a Dropout layer into the all-connection layer; b.L 2 Parameter regularization alpha
Using L 2 Regularization prevents overfitting, so that the model can be changed to have generalization force, more scenes can be used, and the regularization formula is as follows:
J’(θ；x,y)＝J(θ；x,y)+αΩ(θ)
Ω(θ)＝0.5*||w|| 2 2
wherein Ω (θ) is L 2 Parameter regularization process, w is the regularized network layer parameter, α ∈ [0, ∞ ], is the tradeoff of L 2 The parameter regularization and the hyper-parameter of the relative contribution of the standard objective function J' (θ; x, y).
c. By introducing a model complexity index into the loss function, the noise of training data is weakened, and the difference between a model result and an actual value is reduced:
suppose the ith input feature x i True flags of (1):
y i ＝(y1,y2,…,y m ) T
m is the total dimension of the marked vector, and the regression prediction value of the sample network
The prediction error in the t-dimension from its true mark can be expressed as:
assuming n samples are present, the corresponding l 2 The loss function is:
d. index for evaluating precision of harvest index
Using a determining coefficient R 2 And the root mean square error RMSE is used as an evaluation index, and the formula is as follows:
wherein i represents the ith plot, Y i (ii) a harvest index, E, indicating that there is WOFOST in the ith plot, which is artificially and cumulatively combined in the corresponding satellite image time sequence interval i And (4) calculating a harvest index for the ith plot through a convolutional neural network in the same time interval.
Although the invention has been described in detail above with reference to a general description and specific examples, it will be apparent to one skilled in the art that modifications or improvements may be made thereto based on the invention. Accordingly, such modifications and improvements are intended to be within the scope of the invention as claimed.
Claims (10)
1. A crop harvest index inversion method based on a convolutional neural network and a crop model is characterized by comprising the following steps:
s1, selecting a research area and research crops, and performing parameter calibration on a crop growth model WOFOST by using the data of the agricultural gas station, wherein the parameters comprise crop parameters, soil parameters and field management parameters;
s2, downloading MODIS data in the crop growth period, and preprocessing to obtain LST D 、LST N Taking the LSWI index and the APAR index as features to be input;
s3, according to the data record of the agricultural gas station points, taking the NDVI average value of two phenological nodes of the research crop at the beginning and the end of the growth of the research crop in the research area for three years as the judgment threshold value of the corresponding node;
s4, screening all images in the growing period of each block crop according to the threshold judgment method in the S3, performing histogram dimensionality reduction, and generating a matrix of the characteristic images to be input to serve as an input layer of a convolutional neural network;
s5, utilizing MODIS LAI products in the research area and the growth period of the research crops to carry out data assimilation on the WOFOST model, outputting the simulated aboveground part biomass and yield data of the crops corresponding to the block level in the step S2 of the research area, and calculating a harvest index to serve as an output layer of a convolutional neural network;
and S6, setting structural parameters of the convolutional neural network, and training and verifying the model.
2. The method of claim 1, wherein in step 2, the MODIS data is downloaded by using the crop distribution vector data as a mask to extract and download images of the whole growth cycle of the research region and the research crops, and the images include 5 bands of data, namely, the daytime and nighttime surface temperature LST of MYD11A2 D 、LST N NDVI data of MOD13A1, surface reflectance data of NIR1 wave band and surface reflectance data of SWIR1 wave band are preprocessed to obtain a surface water index LSWI and photosynthetically active radiation APAR.
3. The method of claim 2, wherein the preprocessing process is to convert the downloaded data into UTM projection based on WGS-84 ellipsoid, set the uniform resolution of the data to 500m, perform band extraction, stitching and cropping on the remote sensing image according to the plot where the crop is located; and calculating a surface water index LSWI by using NIR1 and SWIR1, calculating a ratio vegetation index SR by using NDVI data, and calculating photosynthetically active radiation APAR absorbed by the crops by using NDVI and SR in combination with meteorological data.
4. The method of claim 3, wherein the land water index LSWI is calculated by the formula:
LSWI＝(ρ NIR1 –ρ SWIR1 )/(ρ NIR1 +ρ SWIR1 )
where ρ is NIR1 Reflectivity of near infrared band (841-875 nm), rho SWIR1 The reflectivity of the short wave infrared band is 1628-1652 nm.
5. The method according to claim 3, wherein the photosynthetically active radiation APAR is calculated by the formula:
APAR(x,t)＝SOL(x,t)×FPAR(x,t)×0.5
wherein SOL is solar radiation flux and has a unit of MJ/m 2 0.5 is the proportion of photosynthetically active radiation in the total solar radiation and FPAR is the proportion of photosynthetically active radiation absorbed by the crop.
6. The method of claim 1, wherein in step S3 the step of fitting the NDVI time series of the research crop over the entire year of the research year to the NDVI of the research crop using the Beck-Double Logistic method is:
f(t)＝a base +(a max -a base )×[(1/(1+e -m1×(t-m2) ))+1/(1+e n1×(t-n2) )]
where t is the day t of the year, i.e., DOY, f (t) refers to the NDVI value by Beck method fitting at time t; a is base Refers to the plot base NDVI value before crop emergence; a is max Is the maximum value of NDVI in a year; m2 is an inflection point of the rising stage, and n2 is an inflection point of the falling stage; m1 and n1 are slope values at points m2, n 2.
7. The method according to claim 6, wherein the step S3 of obtaining the growth period corresponding to the crop in the current year by using the threshold value adopts histogram reduction and normalization, and the specific method includes:
determining the time sequence number of remote sensing images of research year, research district block level and research crop growth period, namely N period total images,
and fusing the N images of the crops in the growth period on a time sequence to form an N x N matrix, and finally generating an N x N4 matrix as an input layer of the convolutional neural network.
8. The method of claim 1, wherein the step S4 of simulating the crop harvest index with wovast and constructing the convolutional neural network output layer is as follows:
downloading day-by-day meteorological data of a research area, wherein the day-by-day meteorological data comprise the highest air temperature and the lowest air temperature of each day, the total rainfall, the average water vapor pressure, the solar radiation quantity and the average wind speed of each day, and interpolating the meteorological data by using ANUSPLIN to form 500m grid data with the resolution equal to that of the downloaded remote sensing image;
utilizing MODIS LAI data, and carrying out data assimilation on a crop growth model by an EnKF ensemble Kalman filtering method, wherein the formula is as follows:
Aa＝Af+K*(Dt-HA)
K＝Ac*HT*(HAc*HT+Dc)-1,K∈Rn*N
in the formula, A a To analyze the matrix, A f Is a prediction matrix, K is an ensemble Kalman gain coefficient, D t Is a matrix of observed variables, D c Is a covariance matrix of the observed variables, H is a nonlinear operator, A is from a prediction equation, and HA is an error covariance matrix of the observed variables;
and (4) carrying out time-period-by-time-period synthesis on the output result of the WOFOST according to the time sequence scale and the spatial position corresponding to the histogram in the step (3), and calculating a harvest index HI.
9. The method according to claim 8, wherein the harvest index HI is calculated by the formula:
HI＝Yield/AGB
wherein HI is the harvest index, YIeld is the amount of dry matter that net accumulates in the storage organ simulated by WOFOST, and AGB is the total biomass of the upper part of the crop field simulated by WOFOST; constructing a sample for each plot, generating a harvest index sample library as an output layer of the convolutional neural network, selecting 2/3 of the samples as a training set, and selecting 1/3 of the samples as a verification set.
10. The method according to claim 1, wherein the specific steps of constructing the convolutional neural network in step S6 are as follows:
performing batch normalization and Relu function activation on each convolution layer, and adding a Dropout layer into the full-connection layer;
and (4) adopting the decision coefficient and the root mean square error as evaluation indexes, and calculating the harvest index through a convolutional neural network in the same time interval.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211345984.8A CN115759526A (en) | 2022-10-31 | 2022-10-31 | Crop harvest index inversion method based on convolutional neural network and crop model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211345984.8A CN115759526A (en) | 2022-10-31 | 2022-10-31 | Crop harvest index inversion method based on convolutional neural network and crop model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115759526A true CN115759526A (en) | 2023-03-07 |
Family
ID=85354470
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211345984.8A Withdrawn CN115759526A (en) | 2022-10-31 | 2022-10-31 | Crop harvest index inversion method based on convolutional neural network and crop model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115759526A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117034778A (en) * | 2023-08-28 | 2023-11-10 | 黑龙江省网络空间研究中心(黑龙江省信息安全测评中心、黑龙江省国防科学技术研究院) | Method for inverting aboveground biomass based on hypershaper-transducer structure |
| CN117876870A (en) * | 2024-01-12 | 2024-04-12 | 中国农业科学院农业资源与农业区划研究所 | Crop estimated yield method and system based on multi-source remote sensing data |
-
2022
- 2022-10-31 CN CN202211345984.8A patent/CN115759526A/en not_active Withdrawn
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117034778A (en) * | 2023-08-28 | 2023-11-10 | 黑龙江省网络空间研究中心(黑龙江省信息安全测评中心、黑龙江省国防科学技术研究院) | Method for inverting aboveground biomass based on hypershaper-transducer structure |
| CN117876870A (en) * | 2024-01-12 | 2024-04-12 | 中国农业科学院农业资源与农业区划研究所 | Crop estimated yield method and system based on multi-source remote sensing data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110222475B (en) | Method for inverting moisture content of winter wheat plants based on multispectral remote sensing of unmanned aerial vehicle | |
| Boken et al. | Improving an operational wheat yield model using phenological phase-based Normalized Difference Vegetation Index | |
| CN106372592B (en) | A kind of winter wheat planting area calculation method based on winter wheat area index | |
| Chen et al. | Spatio-temporal prediction of leaf area index of rubber plantation using HJ-1A/1B CCD images and recurrent neural network | |
| CN109784300A (en) | A kind of crops science survey production method and system | |
| Lizaso et al. | Development of a leaf‐level canopy assimilation model for CERES‐Maize | |
| CN117218531A (en) | Sea-land ecological staggered zone mangrove plant overground carbon reserve estimation method | |
| CN118072178B (en) | Corn yield estimation method and system based on classified percentage data assimilation | |
| CN116415704A (en) | Regional precision irrigation method and system based on multi-data fusion and assimilation | |
| CN111401790B (en) | Method and device for calculating surface water consumption, computer equipment and storage medium | |
| Igder et al. | Multivariate assimilation of satellite-based leaf area index and ground-based river streamflow for hydrological modelling of irrigated watersheds using SWAT+ | |
| CN115759526A (en) | Crop harvest index inversion method based on convolutional neural network and crop model | |
| Pieri et al. | Estimation of actual evapotranspiration in fragmented Mediterranean areas by the spatio-temporal fusion of NDVI data | |
| CN115439402A (en) | Sugar cane yield-per-unit remote sensing estimation method and device based on SAFY-Sugar model | |
| Erencin | C-factor mapping using remote sensing and GIS: A case study of Lom Sak/Lom Kao, Thailand | |
| CN113534083B (en) | SAR-based corn stubble mode identification method, device and medium | |
| CN117690024B (en) | Integrated remote sensing identification method for paddy fields with multiple planting modes | |
| CN105678280B (en) | Mulching film mulching farmland remote sensing monitoring method based on textural features | |
| CN117496369A (en) | Mining area carbon sink data set construction method | |
| Pan et al. | Using QuickBird imagery and a production efficiency model to improve crop yield estimation in the semi-arid hilly Loess Plateau, China | |
| Zhang et al. | Using enhanced vegetation index and land surface temperature to reconstruct the solar-induced chlorophyll fluorescence of forests and grasslands across latitude and phenology | |
| Corbari et al. | Citizen science supporting agricultural monitoring with hundreds of low-cost sensors in comparison to remote sensing data | |
| CN115271991B (en) | Estimation method and device for net primary productivity of crops | |
| CN114663778B (en) | Method and device for improving crop yield | |
| CN115689100A (en) | Regional crop water use efficiency assessment method and system based on data assimilation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20230307 |