CN114694020A

CN114694020A - Construction method of cotton aphid remote sensing forecast model

Info

Publication number: CN114694020A
Application number: CN202210208227.XA
Authority: CN
Inventors: 周学林; 蒋平安; 周建勤
Original assignee: Xinjiang Agricultural University
Current assignee: Xinjiang Agricultural University
Priority date: 2022-03-04
Filing date: 2022-03-04
Publication date: 2022-07-01
Anticipated expiration: 2042-03-04
Also published as: CN114694020B

Abstract

The invention belongs to the technical field of agricultural remote sensing, and particularly relates to a method for constructing a cotton aphid remote sensing forecast model. The method comprises the steps of extracting cotton canopy spectral data of a seedling stage, analyzing the spectrums of healthy cotton plants and cotton aphid pest cotton plants, and constructing cotton aphid recognition models (a multispectral image cotton aphid recognition model, a hyperspectral image cotton aphid recognition model and a ground hyperspectral cotton aphid recognition model) based on spectral characteristics; constructing a cotton growth model by researching the phenological growth state of cotton; the method comprises the steps of researching diffusion rules and influence factors of aphids, constructing a cotton aphid diffusion model, constructing a cotton aphid remote sensing forecasting model by adopting a 4DVar assimilation method, realizing the accurate identification and diffusion judgment of cotton aphids by using a remote sensing technology, being beneficial to mastering insect pest occurrence conditions and pesticide using conditions of cotton in a cotton field, and further having important significance for guiding accurate pesticide application to kill the cotton aphid pests, saving pesticide application cost, improving ecological environment quality and the like.

Description

Construction method of cotton aphid remote sensing forecast model

Technical Field

The invention belongs to the technical field of agricultural remote sensing, and particularly relates to a method for constructing a cotton aphid remote sensing forecasting model.

Background

Cotton is the main economic crop in China, Xinjiang cotton has planted area and output for 27 consecutive years and is the first place in the country, and the cotton industry plays an important role in realizing desert greening, farmer income increasing, national economic development and the like. Compared with island cotton, the upland cotton is more prone to pest and disease damage in the growing process, more than 10 kinds of harmful pests are caused all the year round, the yield of cotton is reduced by about 10-15% due to the pest and the disease damage, the cotton is affected by global climate environment change, ecological natural regulation and control of cotton fields are weakened due to continuous planting, and the pest and disease damage of the cotton tends to be increased day by day.

Aphids are one of the main pests harmful to cotton, are parasitized on young and tender parts of plants, are small in size, the length of adults is less than 2 mm, early-stage harmful symptoms are not easy to capture and detect, yellow white spots, curling or shrinking of leaves are caused after aphids appear, development and growth retardation of new leaves are affected seriously, the growth cycle is delayed, and the quality and yield of cotton are affected seriously. There are 3 aphids harmful to Xinjiang cotton, namely cotton black aphid at seedling stage, then cotton aphid and long-tube cotton aphid, and the occurrence and harm of the cotton aphid are the most serious; meanwhile, the cotton aphids have migratory flight property and strong mobility, and the insect situation outbreak is rapid and spread rapidly, which can cause the loss of the cotton peanut yield in an unappreciable way.

At present, prediction and forecast of cotton aphids mainly depend on field manual investigation experience judgment of technical personnel of agricultural plant protection departments in various regions, time and labor are consumed, and meanwhile, because the number and the area of investigation sample plots are limited, actual field occurrence and dangerous pest dynamic conditions are difficult to accurately reflect, and the requirement of large-area cotton field pest control is difficult to meet; the plant diseases and insect pests in the field are in the form of spot sheets in the initial stage, if the investigation result is not accurate enough and timely, the plant diseases and insect pests are damaged in a large range due to the spot sheets, and the plant protection work is not facilitated to be smoothly carried out. Meanwhile, the prevention and control mode of the diseases and insect pests at the present stage is that pesticides are passively sprayed in a large range for prevention and control, timely and rapid accurate prevention and control cannot be realized aiming at insect sources, the application range and dosage of the pesticides are enlarged, and negative effects such as mutagenesis, carcinogenesis, teratogenesis, pesticide residues and environmental pollution caused by chemical pesticides are increasingly prominent; in addition, the extensive overuse of chemical pesticides leads to a gradual increase in the resistance of Aphis gossypii.

When crops are infected with insect pests, the appearance forms of leaves such as color, shape and texture of the crops can be changed correspondingly, and scholars at home and abroad analyze the color, shape and texture characteristics of the leaves of the crops by using different information technical means and diagnose the diseases of the crops by characteristic processing. Research shows that the crop disease diagnosis by using information technology means such as computer vision and the like can reach a certain accuracy rate, the method has good feasibility, the current-stage pest nondestructive monitoring is mainly based on remote sensing monitoring, a common monitoring system comprises a visible light-near infrared spectrum sensor, a hyperspectral sensor, a fluorescence and thermal sensor, a synthetic aperture radar and the like, the remote sensing monitoring characteristics of plants are established according to different types of acquired data, and the method is applied to the monitoring process to visually express the plant damage condition.

In the prior art, a cotton aphid damage diagnosis model research (approved, Cotton science, 2020, 32 (2): 133-142) based on leaf textural features obtains hyperspectral images of healthy and cotton aphid-damaged cotton leaves, extracts the textural features of the images by utilizing a gray level symbiotic matrix, constructs a cotton aphid damage diagnosis model, and obtains a texture feature vector based model capable of effectively realizing identification of cotton aphid damage leaves.

In the research on remote sensing monitoring of cotton aphids on a regional scale (Master thesis of Henan university of Henan, 2019), cotton aphids in the northern region of Kurler are monitored by a spectral angle method and a Logistic modeling method, the identification precision of the two methods is analyzed and evaluated, and the aphid monitoring method based on the Logistic model is better obtained by specific analysis in the aspects of test equipment price, data acquisition, data processing, classification precision and the like.

In the research on spectral identification and remote sensing monitoring of cotton plant diseases and insect pests (field, university of Shandong agriculture university Master thesis of 2016), cotton in Xinjiang planting area of China is taken as a research object, HJ satellite data is combined with ground canopy hyperspectral data to extract the area of the cotton, early identification and extraction of the occurrence of the plant diseases and insect pests are carried out, 6 selected vegetation indexes capable of identifying and distinguishing cotton verticillium wilt and cotton leaf mites and wavelet characteristics of 9 wave bands with the largest correlation coefficients with sample categories in different wave band ranges under 10 different scales are selected, and the accuracy of the established discrimination model is generally higher than that of the model established by the vegetation indexes.

An improved deep confidence network is applied in the prediction of cotton diseases and insect pests (Wangbaifeng, Cotton journal, 2018, 30 (4): 300-307) to provide 1 cotton disease and insect pest prediction model based on environmental information and deep belief network, which consists of 3 layers of limiting Boltzmann machine network and 1 supervising back propagation network, RBM is used to convert the environmental information data to a new characteristic space related to the occurrence of the diseases and insect pests, BP network is used to classify and predict the characteristic vector output by the last 1 layer, the training process of RBM is accelerated by using dynamic learning rate and contrast dispersion criterion, and the model is used to perform prediction test on bollworm, cotton aphid, red spider, verticillium wilt and blight of cotton in nearly 6 years. The model integrates massive and multi-source environmental information data, and overcomes the defects of the existing prediction method to a certain extent; however, the number of hidden layers and nodes of hidden layers of the model is almost not set according to theory, the model is generally set according to experience and test results, the time is long, and the stability and accuracy of the prediction performance of the model need to be further verified and improved.

There have been some relevant research achievements in the field of remote sensing monitoring of cotton in China. In the aspect of cotton pest monitoring, the invention with application number of 201910336544.8 discloses unmanned aerial vehicle equipment for monitoring cotton pests, which can shoot images from the upper part of cotton and the back of leaves, realize the monitoring of the cotton pests in all aspects and know the conditions of the cotton pests in advance; the invention with application number 202011162451.7 discloses a three-dimensional monitoring method and system for cotton pests based on a small unmanned aerial vehicle cluster, which effectively monitors pest information in the middle and at the bottom of cotton in a short distance by utilizing the unmanned aerial vehicle cluster based on a three-dimensional vision technology. In the aspect of cotton crop remote sensing monitoring, the invention with application number 201711500930.3 discloses a cotton remote sensing monitoring method based on phenological analysis, which realizes automatic extraction of cotton planting information by regional cotton spectral feature analysis and infected crop phenological analysis in combination with a remote sensing image layered classification method. The invention with the application number of 201911087521.4 discloses a method for extracting spectral characteristics of cotton canopy of aphid pests in the bud period of cotton and estimating the level of the aphid pests, which utilizes an aphid pest level estimation model of a partial least square method of a sensitive spectral ratio derivative spectral value to be applied to imaging hyperspectral images.

However, the prior art center still has the following problems:

(1) at present, the research on cotton plant diseases and insect pests is mostly carried out in an indoor laboratory or in a small range by using an unmanned aerial vehicle, the main targets are plant diseases and insect pests such as spider mites, verticillium wilt and the like in a boll stage, the research on field remote sensing monitoring tests on aphid damage in a seedling stage is not carried out, and the identification method and the occurrence rule of the field aphid damage are not analyzed.

(2) The application of the prior art is mostly carried out in a laboratory environment, the environment in the laboratory is single, noise interference is less, a good effect can be obtained, various interferences exist in the actual field environment, and the problems that the accuracy is low, research objects and technologies are disjointed and the like are caused by low applicability of a hyperspectral technology in the field cotton aphid damage diagnosis process are caused.

(3) The existing research develops spectral analysis research, but the analysis of crop spectral characteristics is to consider the influence of insect pests on physiological changes in crops less and lack the analysis research on all waveband information of crops infected with insect pests; many factors influence the change of spectral characteristics, most remote sensing monitoring results aim at specific plant diseases and insect pests, other problems in the crop production process are not involved, and the method cannot be specifically practiced in agricultural production management.

(4) The existing technical method for identifying cotton aphids is carried out by adopting a single data source, does not adopt multi-source data for analysis, does not carry out multi-source data fusion assimilation research, and cannot overcome the problem of low identification precision caused by information loss such as space-time spectrum and the like of the single data source.

Disclosure of Invention

Aiming at the problems that the influence of insect pests on the physiological change in crops is less considered in the prior art, and the analysis and research on all wave band information of crops infected with insect pests are lacked; the invention provides a method for diagnosing the cotton aphid pests in the field, which solves the problems of low accuracy, disjointed research objects and technologies and the like in the hyperspectral technology in the diagnosis process of the cotton aphid pests in the field due to low applicability.

The technical scheme adopted by the invention is as follows:

a method for constructing a cotton aphid remote sensing forecast model comprises the following steps:

step 1: acquiring hyperspectral data of cotton canopy ground of a normal cotton plant and a cotton plant stressed by aphids for multiple times in a cotton seedling stage;

step 2: analyzing and processing the cotton canopy ground hyperspectral data obtained in the step (1), screening out a sensitive spectrum, and constructing a ground hyperspectral Aphis gossypii Glover identification model according to the sensitive spectrum;

and step 3: synchronously acquiring multi-source remote sensing data according to the time nodes in the step 1, analyzing and processing unmanned aerial vehicle multispectral image data, unmanned aerial vehicle hyperspectral image data, satellite multispectral image data and satellite hyperspectral image data in the multi-source remote sensing data, and constructing a cotton aphid pest multispectral identification method;

analyzing and processing the satellite multispectral image data, and obtaining a multispectral image cotton aphid recognition model by combining the cotton aphid pest multispectral recognition method;

analyzing and processing the satellite hyperspectral image data, and combining the ground hyperspectral cotton aphid recognition model obtained in the step 2 to obtain a hyperspectral image cotton aphid recognition model;

and 4, step 4: acquiring growth phenological data according to the time nodes in the step 1, and constructing a cotton growth model according to the growth phenological data and the multi-source remote sensing data acquired in the step 3;

and 5: determining an influence factor of cotton aphid diffusion, collecting and recording the population number of the cotton aphids, the natural enemy number of the cotton aphids and relevant data of the influence factor for many times according to the time node in the step 1, and constructing a cotton aphid diffusion model according to the collected data;

step 6: and (4) constructing a cotton aphid remote sensing forecasting model according to the multispectral image cotton aphid recognition model and the hyperspectral image cotton aphid recognition model obtained in the step (3), the cotton growth model obtained in the step (4) and the cotton aphid diffusion model obtained in the step (5).

After the technical scheme is adopted, the method extracts the spectral data of the cotton canopy at the seedling stage, analyzes the spectrums of healthy cotton plants and cotton aphid pest cotton plants, and constructs a cotton aphid recognition model (a multispectral image cotton aphid recognition model, a hyperspectral image cotton aphid recognition model and a ground hyperspectral cotton aphid recognition model) based on spectral characteristics; constructing a cotton peanut growth model by researching the phenological growth state of cotton; the method is characterized by researching the diffusion rule and the influence factors of aphids, constructing an aphid diffusion model, constructing a cotton aphid remote sensing forecasting model by adopting a 4DVar assimilation method, realizing the accurate identification and diffusion judgment of cotton aphids by using a remote sensing technology, being beneficial to mastering the pest occurrence condition and the pesticide application condition of cotton in a cotton field, and further having important significance for guiding accurate pesticide application to kill the cotton aphids, saving pesticide application cost, improving ecological environment quality and the like.

Preferably, in the step 1, the cotton canopy ground hyperspectral data is obtained through a surface feature spectrometer, the height of an observation probe of the surface feature spectrometer from a cotton canopy is H, and the calculation formula of the height H is as follows:

wherein theta is an observation probe visual field angle, and L is a cotton planting row spacing.

After the technical scheme is adopted, the height H of the spectrometer observation probe from the cotton crown layer is calculated according to the field angle of the probe and the cotton planting row spacing, and the method is more accurate.

Preferably, the specific steps of obtaining the ground hyperspectral aphis gossypii aphid recognition model in the step 2 comprise:

step 2.1: cleaning: cleaning the cotton canopy ground hyperspectral data obtained in the step 1, and removing abnormal values;

step 2.2: pretreatment: denoising and smoothing the cotton canopy ground hyperspectral data obtained in the step 2.1;

step 2.3: correlation derivative and differential calculation: carrying out derivative and differential calculation on the cotton canopy ground hyperspectral data obtained in the step 2.2;

wherein the derivative is taken

And

calculating a formula model;

wherein R is^1st(λ_i) In the spectral region lambda_iThe value of the first derivative of (A) above, R (λ)_i) Is the value of the reflectivity at the i-band, R (λ)_i+1) Is the value of the reflectivity at the (i +1) band; r^2nd(λ_i) In the spectral region lambda_iA second derivative value of (d);

differential adoption

And ρ' (λ)_i)＝[ρ′(λ_i+1)-ρ′(λ_i-1)]/2(λ_i+1-λ_i-1) Calculating a formula model;

where ρ' (λ)_i) In the spectral region lambda_iThe first order differential value above; ρ (λ)_i+1) Is the value of the reflectivity at the (i +1) band; ρ (λ)_i-1) Is the value of the reflectivity at the (i-1) wavelength band;ρ″(λ_i) In the spectral region lambda_iSecond order differential value of (d); ρ' (λ)_i+1) Is the first order differential value at the (i +1) band; ρ' (λ)_i-1) Is the first order differential value at (i-1) band; lambda [ alpha ]_iIs the wavelength value of the i wave band; lambda [ alpha ]_i+1Is the wavelength value of the (i +1) band.

Step 2.4: and (3) index calculation: calculating NNIR index and red edge index R according to the cotton canopy ground hyperspectral data obtained in the step 2.2_(550,760)And disease plant spectral index beta;

wherein, the NNIR index adopts a formula model

Calculation was performed where NNIR is the vegetation index, ρ₇₆₀Is the reflectance value at the wavelength band of 760 nm; rho₆₅₀A reflectance value at a wavelength band of 650 nm; rho₅₅₀A reflectance value at a wavelength band of 550 nm;

red edge index R_(550,760)Using a formula model

Performing a calculation in which R_(550,760)Is a specific vegetation index; rho₇₆₀Is the reflectance value at the wavelength band of 760 nm; rho₅₅₀A reflectance value at a wavelength band of 550 nm;

disease plant spectral index beta adopts a formula model

Calculation was performed where NNIR is the vegetation index, R_(550,760)Is a specific vegetation index;

step 2.5: insect pest judgment: red edge index R calculated according to step 2.4_(550,760)Judging whether the cotton has insect damage and the degree of the insect damage according to the disease plant spectral index beta;

step 2.6: characteristic spectrum analysis: comprehensively analyzing the cotton canopy ground hyperspectral data obtained in the step 2.2 and the derivative data and differential data calculated in the step 2.3, and screening out a characteristic spectrum section of cotton aphid pests;

step 2.7: constructing a model: and (4) constructing a ground hyperspectral cotton aphid identification model by adopting a spectrum angle model according to the characteristic spectrum section of the cotton aphid pests screened out in the step 2.6.

Preferably, the specific steps of obtaining the multispectral image aphis gossypii identification model and the hyperspectral image aphis gossypii identification model in the step 3 include:

step 3.1: acquisition of remote sensing data of the unmanned aerial vehicle: acquiring multispectral data sets of a cotton canopy of the unmanned aerial vehicle of a normal cotton plant and a cotton plant stressed by aphids in a cotton seedling stage by using the unmanned aerial vehicle according to the time node in the step 1;

step 3.2: unmanned aerial vehicle remote sensing data processing: preprocessing the multispectral data set of the cotton canopy of the unmanned aerial vehicle obtained in the step 3.1 to obtain multispectral image data of the unmanned aerial vehicle and hyperspectral image data of the unmanned aerial vehicle;

step 3.3: acquiring satellite remote sensing data: acquiring satellite high-resolution remote sensing data and satellite hyperspectral remote sensing data according to the time node in the step 1;

step 3.4: satellite remote sensing data processing: preprocessing the satellite high-resolution remote sensing data obtained in the step 3.3, using an ENVI Registration tool to complete geographic Registration by adopting a second-order polynomial method to obtain satellite multispectral image data and satellite panchromatic image data, and preprocessing the satellite hyperspectral remote sensing data obtained in the step 3.3 to obtain satellite hyperspectral image data;

step 3.5: the construction of the cotton aphid pest multispectral identification method comprises the following steps: analyzing the identification characteristics of multispectral cotton aphid pests according to the multispectral image data of the unmanned aerial vehicle obtained in the step 3.2 and the multispectral image data of the satellite obtained in the step 3.4 to construct a multispectral cotton aphid pest identification method, wherein the construction method of the multispectral cotton aphid pest identification method comprises the following steps:

step 3.5.1: and (3) selecting a multispectral Aphis gossypii characteristic spectrum segment and an index: calculating NNIR of the satellite multispectral image data and the unmanned aerial vehicle multispectral image data according to the formula listed in the step 2.4_mulIndices (F ═ 1.26; df ═ 4, 12;^ap-0.0368), red edge index R_mul(550，760)(F＝1.57；df＝4,12；^aP-0.0460) and spectrum index beta of diseased plant_mul(F＝9.73； df＝4,12；^bP ═ 0.0008), where F is the ratio of the two mean squares, df is the factor degree of freedom, P determines the factor significance,^aP＜0.05；^bP＜0.001；

step 3.5.2 construction

(F-101.73; df-4.12; P-0.0005) second order polynomial regression models, where y is the aphid pest interpretation index; phi, phi,

Gamma is the yield coefficient; and b is an adjusting coefficient.

Step 3.5.3: setting the cotton aphid pest distinguishing threshold value as T₁I.e. having y < T₁Cotton aphids were not obvious; y is more than or equal to T₁When the aphid occurs, the cotton aphid occurs;

step 3.6: constructing a multispectral image cotton aphid recognition model: after the satellite multispectral image data obtained in the step 3.4 and the satellite panchromatic image data are subjected to homologous fusion, fused super-resolution multispectral image data are obtained, the cotton aphid pest multispectral identification model obtained in the step 3.5 is applied to the fused super-resolution multispectral image, a multispectral image cotton aphid identification model is comprehensively constructed, and the construction method of the multispectral image cotton aphid identification model comprises the following steps:

step 3.6.1: performing homologous fusion on the satellite multispectral image data and the satellite panchromatic image data to obtain fused multiresolution multispectral image data;

step 3.6.2: calculating NNIR of satellite multispectral image data based on the formula of step 2.4 and step 3.5.1_mulIndex, red edge index R_mul(550, 760) and spectrum index beta of diseased plants_mulSynchronously confirming the values of F, df and P, and keeping the values within a confidence interval that P is less than 0.001;

step 3.6.3: determining a multispectral image aphis gossypii glover identification model based on the model obtained in the step 3.5.2 and the discrimination rule in the step 3.5.3;

step 3.7: constructing a hyperspectral image cotton aphid recognition model: and 3.5, performing principal component analysis on the satellite hyperspectral image data obtained in the step 3.4, using the satellite hyperspectral image data and the image data subjected to data dimensionality reduction as low-pass parameters, using the panchromatic band of the fused ultra-resolution multispectral image data obtained in the step 3.6 as high-pass parameters to be fused, obtaining fused high-resolution hyperspectral image data, performing data assimilation operation on the cotton canopy ground hyperspectral data and the fused high-resolution hyperspectral image data, and constructing a hyperspectral image cotton aphid recognition model by using a deep learning algorithm according to the step 3.5.

Preferably, the specific steps of constructing the cotton growth model in the step 4 include:

step 4.1: model parameter establishment and collection: the model parameters comprise predicted images, the predicted images are obtained by processing the multi-source remote sensing data obtained in the step 3, physical parameters, chemical parameters, structural parameters and biological parameters are obtained according to the same time node of the multi-source remote sensing data obtained in the step 3, and the physical parameters, the chemical parameters, the structural parameters and the biological parameters comprise crop data, weather data, soil data and field management data which are collected on the spot;

step 4.2: and (3) observation data collection and processing: processing the multi-source remote sensing data collected in the step 3 to obtain a fusion image;

step 4.3: and (3) calibrating model parameters: uniformly distributed U formed by parameter maximum values according to the parameters acquired in the step 4.1_(min，max) Or calibrating the parameters by normal distribution, wherein the specific calibration formula is as follows: logL ═ logL_LAI+logL_yield；

logL_LAf＝-0.5(x-x_obs)^T∑^-1(x-x_obs)-0.5Klog(2π)-0.5log(Δ∑)；

Wherein L represents a likelihood function; x and x_obsModel simulation value and observation value corresponding to LAI respectivelyA time series vector of; Σ represents a covariance matrix of LAI observations; k represents a vector dimension; Δ Σ represents a column-column value of Σ; y and Y_obsAn analog value and an observed value respectively representing the yield; σ represents the standard deviation of the yield observation;

step 4.4: establishing an assimilation algorithm: carrying out 4Dvar assimilation algorithm on the model parameters obtained in the step 4.1 and the observation data obtained in the step 4.2 to obtain assimilation data, wherein the algorithm formula is

Wherein x_kRepresenting model parameters, B is a model error, y is an observation number, Q is an observation error, and H is an observation operator;

step 4.5: and (3) constructing a cotton growth model, namely constructing the cotton growth model by combining the assimilation data obtained in the step 4.4 according to the WOFOST model of Waherty root in the Netherlands.

Preferably, the specific step of obtaining a predictive image in step 4.1 comprises:

step 4.1.1: carrying out accurate geometric registration on the multi-source remote sensing data obtained in the step 3;

step 4.1.2: performing translation invariance wavelet transformation on the multisource remote sensing data to obtain corresponding wavelet transformation low-frequency and high-frequency component coefficients with different resolutions and different directions;

step 4.1.3: fusing the low-frequency coefficients obtained in the step 4.1.2 by using weighted average, and fusing the high-frequency coefficients and the low-frequency coefficients by using a large operator;

step 4.1.4: performing inverse wavelet transform to obtain a predicted image;

preferably, the specific steps of obtaining the fused image in step 4.2 include:

step 4.2.1: performing contrast tower decomposition on each source image obtained in the step 3 respectively to establish a contrast pyramid of each image;

step 4.2.2: respectively performing fusion processing on each decomposition layer of the image pyramid, adopting a weighted average operator for a low-frequency part of the decomposed image, and using a large operator for a high-frequency part to process to obtain a contrast pyramid of the fused image;

step 4.2.3: and performing inverse pyramid transformation on the fused contrast pyramid to obtain a reconstructed image which is a fused image.

Preferably, the specific operation steps of the 4Dvar assimilation algorithm include:

step 4.3.1: taking a mode forecast field as an initial estimation field;

step 4.3.2: carrying out initialization processing on the updated field;

step 4.3.3: and (4) forecasting the mode forwards for a plurality of steps, taking the new forecasting field as an initial estimation field of the next updating, and then returning to the step 4.3.1 to form a loop.

Preferably, the specific steps of constructing the cotton aphid diffusion model in the step 5 comprise:

step 5.1: data acquisition and processing: determining an influence factor of cotton aphid diffusion, acquiring and recording the number of cotton aphids, the number of natural enemies of the cotton aphids and relevant data of the influence factor for multiple times according to the time node in the step 1, and analyzing the relevant data of the influence factor, the number of the cotton aphids and the data of the natural enemies of the cotton aphids to obtain a rule of occurrence time of the cotton aphids, a rule of quantity of the cotton aphids in different periods and a rule of influence of the influence factor on the quantity of the cotton aphids;

step 5.2: image space information statistics: according to the multi-source remote sensing data collected in the step 3, the spatial information characteristics on the image are counted, the evolution of the cotton aphid pest situation is analyzed, and the spectral reflectivity of the image is obtained according to the multi-source remote sensing data;

step 5.3: and (3) correlation analysis: establishing the relationship between the number of cotton aphids and the spectral reflectivity of the image, the rule of the occurrence time of the cotton aphids and the spectral reflectivity of the image, and the rule of the number of the cotton aphids in different periods and the cotton plant disease spectral index beta and the number of the cotton aphids and the influence factor obtained in the step 2.4 according to the data obtained in the step 5.1 and the step 5.2;

step 5.4: constructing a cotton aphid diffusion model: and (4) constructing a cotton aphid diffusion model by adopting a mathematical statistics and quantitative analysis method, a spatial interpolation method, network infectious disease dynamics and a deep neural network method based on the correlation analysis result of the step 5.3.

Preferably, the specific construction formula for constructing the cotton aphid diffusion model in the step 5.4 comprises:

let u (x, y, t) be the number of Aphis gossypii at time (x, y) t, and K (u) be the population dynamics, then there is the formula: k (u) a (x, y, t) u + b (x, y, t), wherein b (x, y, t) is the cotton aphid diffusivity;

wherein p (x, y, t) is the diffusion coefficient;

a (x, y, t) ═ e (x, y, t) -d (x, y, t) + c (x, y, t), where e (x, y, t) is the cotton aphid birth rate; c (x, y, t) is the migration rate of Aphis gossypii; d (x, y, t) is the cotton aphid death rate;

wherein q (x, y, t) is the mobility coefficient;

the birth rate and death rate of cotton aphid are independent of density, and the experimental environment is uniform, and p (x, y, t) is p (t), q (x, y, t) is q (t), e (x, y, t) is e (t), d (x, y, t) is d (t), then:

wherein the birth rate e (t) and the death rate d (t) are obtained from field investigations.

Preferably, the specific steps of constructing the cotton aphid remote sensing forecasting model in the step 6 comprise:

step 6.1: heterogeneous data assimilation: performing heterogeneous data assimilation on the multispectral image aphis gossypii glover identification model and the hyperspectral image aphis gossypii glover identification model obtained in the step 3, the cotton growth model obtained in the step 4 and the cotton aphid diffusion model obtained in the step 5 based on a 4DVar assimilation method to obtain an assimilation coupling system;

step 6.2: and (3) selecting a cotton growth and aphid pest stress model, namely constructing the cotton aphid pest stress model by adopting a partial least square model based on the assimilation coupling system in the step 6.1 and the cotton aphid data identified on the image obtained from the step 3 through the multispectral image cotton aphid identification model and the hyperspectral image cotton aphid identification model, wherein a formula model is as follows:

X＝TP^T+E

Y＝UQ^T+F

wherein X is a prediction matrix of nxm; y is a response matrix of nxp; t is an nxl X projection matrix; u is an n multiplied by l Y projection matrix; p is an m × l orthogonal load matrix; q is an orthogonal load matrix of p x l; the matrixes E and F are error terms and are normal distribution random variables which are independent and distributed in the same way;

step 6.3: constructing a cotton aphid remote sensing forecasting model: based on the assimilation coupling system obtained in the step 6.1, the system analyzes the relationship among the cotton aphid stress state, the cotton aphid diffusion state, the cotton spectral information and the air-space multi-source remote sensing image information in the cotton chemical growth process, a cotton aphid remote sensing forecasting model is constructed by adopting a least square twin support vector machine method, and the calculation formula is as follows:

where h (x) is the wavelet mother function and σ is the kernel function.

In summary, due to the adoption of the technical scheme, the invention has the beneficial effects that:

1. the method comprises the steps of extracting cotton canopy spectral data at a seedling stage, analyzing spectrums of healthy cotton plants and cotton aphid insect pest cotton plants, and constructing a cotton aphid recognition model based on spectral characteristics; constructing a cotton growth model by researching the phenological growth state of cotton; researching the diffusion rule and the influence factor of the aphids, constructing an aphid diffusion model, and constructing a cotton aphid remote sensing forecasting model by adopting a 4DVar assimilation method.

2. The method realizes the accurate discrimination of the remote sensing technology on the cotton aphids and the diffusion, is beneficial to mastering the pest occurrence condition and the pesticide use condition of cotton in a cotton field, and further has important significance for guiding the accurate pesticide use to eliminate and kill the cotton aphids, saving the pesticide use cost, improving the quality of the ecological environment and the like.

3. The method provides decision support for timely and accurate monitoring of cotton aphid information and prevention and control of insect pest situation, and further has important application value in reducing pesticide and stabilizing yield of cotton fields and protecting farmland ecological environment.

Drawings

FIG. 1 is a flow chart of the present invention;

FIG. 2 is a schematic view of a ground survey unit arrangement of the present invention;

FIG. 3 is a flow chart of the construction of a ground hyperspectral aphis gossypii aphid recognition model.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings attached to the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all the embodiments. The components of the embodiments of the present application, generally described and illustrated in the figures herein, can be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present application, as presented in the figures, is not intended to limit the scope of the claimed application, but is merely representative of selected embodiments of the application. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present application without making any creative effort, shall fall within the protection scope of the present application.

Example 1

A method for constructing a cotton aphid remote sensing forecast model comprises the following steps (as shown in figure 1):

this example was carried out in shaya county, aksu, of the Uygur autonomous region of Xinjiang. The area is located around the Tarim river, soil salinization and desertification are serious, cotton is a local main planting crop, the scale is large, and the planting structure is simple. The cotton planting variety is recommended in Aksu area, sowing is started in the middle of 4 months, and cotton aphids naturally occur in the field. The data acquisition time is from the late 5 months to the early 7 months of the cotton seedling stage.

Step 1: acquisition of cotton canopy ground hyperspectral dataThe portable surface feature spectrometer is SVC HR768, the spectral band is 350nm-2500nm, 768 wave band channels and the spectral resolution is 3.5nm @1000nm, a 25-degree field angle optical fiber probe is adopted to collect the high spectral data of the ground of the cotton canopy, and the observation time interval is 10:30-16:30 at a specified place so as to ensure a sufficient solar altitude; the meteorological requirements are as follows: the ground visibility is not less than 10 km; the light cloud volume in a solid angle range of 90 degrees around the sun is less than 2 percent, and no cirrus cloud, dense cloud and the like exist; wind power is less than 3 grades; the spectral data of each point of the sampling data is not less than 10 times, and the spectral data passes through a formula according to a planting mode of 12 cotton plants at a row spacing of 12cm, a row spacing of 66cm and 12 cotton plants, and the principle of covering one row spacing

And calculating the height H of the optical fiber probe from the cotton canopy to be 150 cm.

Taking the mode of planting three lines of the cotton semi-membrane as a reference, beginning from the third semi-membrane, surveying according to a survey unit of a 5-point sampling method, wherein each survey unit is arranged at intervals of three lines of the semi-membrane in the transverse direction and 10 meters in the longitudinal direction (as shown in figure 2), and once every 5 days, before and after satellite transit and 30 minutes of unmanned aerial vehicle data acquisition, the number of aphids and natural enemies are surveyed at the middle row, and the number of the aphids and the natural enemies are recorded

According to the technical Specification for measuring and reporting cotton aphids (GB/T15799. sup. 2011) in late 5 months to late 7 months, Beidou positioning equipment is adopted to accurately select sample plots with the height of 25m, the middle and the lower 25m in a field, cotton canopy ground hyperspectral data of cells with the length of 1m and the length of 1m in the sample plots are collected every 5 days (as shown in figure 2), and cotton growth climate data, meteorological information, cotton aphid diffusion data, aphid population number, natural enemy number and the like are synchronously collected;

step 2: analyzing and processing the cotton canopy ground hyperspectral data obtained in the step 1, screening out a sensitive spectrum section, and constructing a ground hyperspectral aphis gossypii aphid recognition model according to the sensitive spectrum section; the method comprises the following specific steps (as shown in figure 3):

step 2.2: pretreatment: adopting a Savitzky-Golay method for the cotton canopy ground hyperspectral data obtained in the step 2.1, setting a polynomial as 5 th order, and performing denoising smoothing treatment to eliminate high-frequency noise influence;

step 2.3: derivative and differential calculation: using a formula

And formula

Performing derivative calculation on the cotton canopy ground hyperspectral data obtained in the step 2.2 by adopting a formula

And formula

Performing differential calculation on the cotton canopy ground hyperspectral data obtained in the step 2.2 to obtain data such as first derivative, second derivative, fractional order differential (0.5 order and 1.5 order) and the like of the cotton canopy ground hyperspectral data; (wherein, R^1st(λ_i) In the spectral region lambda_iThe value of the first derivative of (A) above, R (λ)_i) Is the value of the reflectivity at the i-band, R (λ)_i+1) Is the reflectance value at the (i +1) wavelength band; r is^2nd(λ_i) In the spectral region lambda_iA second derivative value of (d); where ρ' (λ)_i) In the spectral region lambda_iThe first order differential value above; ρ (λ)_i+1) Is the value of the reflectivity at the (i +1) band; ρ (λ)_i-1) Is the value of the reflectivity at the (i-1) wavelength band; ρ' (λ ″)_i) In the spectral region lambda_iSecond order differential value of (d); ρ' (λ)_i+1) Is the first order differential value at the (i +1) band; ρ' (λ)_i-1) Is the first order differential value at (i-1) band; lambda [ alpha ]_iIs the wavelength value of the i wave band; lambda [ alpha ]_i+1Is the wavelength value of the (i +1) band. )

Step 2.4: and (3) index calculation: according to the cotton canopy ground hyperspectral data obtained in the step 2.2, a formula is adopted

Calculating the NNIR index by using the formula

Calculated red edge index R_(550,760)By the formula

Calculating a spectral index beta of a disease plant;

step 2.5: insect pest judgment: red edge index R calculated according to step 2.4_(550,760)And judging whether the cotton has insect damage or not and the degree of the insect damage according to the disease plant spectral index beta, and if R is equal to R, judging whether the cotton has the insect damage or not and judging the degree of the insect damage_(550,760)>0.2 can determine that the cotton is infected with spider mite pests; if R is_(550,760)<When the time is 0.13, the cotton can be judged to be infected with the spider mite insect pest preliminarily; if R is_(550,760)In the (0.13, 0.2) stage, the comprehensive judgment of beta is needed: when beta is more than 5.8, the cotton plant is in a normal state; when beta is less than 5.8, the plant is in the transformation process from the initial stage of disease infection to the moderate disease;

step 2.6: characteristic spectrum analysis: comprehensively analyzing the cotton canopy ground hyperspectral data preprocessed in the step 2.2 and the derivative data and differential data calculated in the step 2.3, and screening out a characteristic spectrum section of cotton aphid pests;

step 2.7: constructing a model: and (4) constructing a cotton aphid remote sensing identification model by adopting a spectrum angle model according to the characteristic spectrum of the cotton aphid pests screened in the step 2.6.

And 3, step 3: and step 3: synchronously acquiring multi-source remote sensing data according to the time nodes in the step 1, analyzing and processing unmanned aerial vehicle multispectral image data, unmanned aerial vehicle hyperspectral image data, satellite multispectral image data and satellite hyperspectral image data in the multi-source remote sensing data, and constructing a cotton aphid pest multispectral identification method;

analyzing and processing the satellite multispectral image data, and obtaining a multispectral image cotton aphid identification model by combining the cotton aphid pest multispectral identification method;

unmanned aerial vehicle remote sensing platform includes: the system comprises a quad-rotor unmanned aerial vehicle, a flight control system, a RedEdge-MX airborne multispectral imager, a ground control system, a data processing system and a miniature portable computer. Wherein the four-rotor unmanned aerial vehicle is a Xinjiang longitude and latitude M300, and the endurance time is 30 minutes; the RedEdge-MX airborne multi-spectral imager is a product of MicaSense company, adopts a hovering scanning imaging mode, and has a spectral range of 400-900nm (blue band central wavelength of 475nm and wave width of 40nm, green band central wavelength of 560nm and wave width of 20nm, red band central wavelength of 668nm and wave width of 10nm, red band central wavelength of 717nm and wave width of 10nm, near infrared band central wavelength of 840nm and wave width of 40nm), a horizontal field angle of 47.2 degrees, and a spatial resolution of 0.08m @120 m height. The unmanned aerial vehicle remote sensing operation is clear and cloudless in the day, the wind speed is less than 3 grades, the navigation speed is 3m/s, the navigation height is 50m, the course overlapping degree is 75%, and the sidewise overlapping degree is 75%. After data are collected according to the specification, PIX4D software is used for data preprocessing. The unmanned aerial vehicle remote sensing operation is clear and cloudless on the day, the wind speed is less than 3 grades, the navigation speed is 3m/s, the navigation height is 50m, the course overlapping degree is 75%, and the side direction overlapping degree is 75%. After data are collected according to the specification, PIX4D software is used for data preprocessing, and the method specifically comprises the following steps:

step 3.1: acquisition of remote sensing data of the unmanned aerial vehicle: acquiring multispectral data sets of a cotton canopy of an unmanned aerial vehicle of a normal cotton plant and a cotton plant stressed by aphids at a cotton seedling stage by using the unmanned aerial vehicle;

step 3.2: data processing of unmanned aerial vehicle remote sensing data: and (3) analyzing and processing the multispectral data set of the cotton canopy of the unmanned aerial vehicle obtained in the step (3.1) to obtain a multi/hyperspectral image, a panchromatic image, an RGB image, a near-infrared single-band image and a red-edge spectral band image of the unmanned aerial vehicle, performing Savitzky-Golay smoothing processing and spectral differential processing on the images to obtain standard preprocessing data, and cutting the standard preprocessing data by using an ENVI ROI tool to obtain the multispectral image data of the unmanned aerial vehicle of the target land parcel and the high-spectrum image data of the unmanned aerial vehicle.

Step 3.3: acquiring satellite remote sensing data: programming and collecting high spectral data of a 'Zhuhai No. one' satellite constellation and high spectral data of a 'Gao fen No. five' satellite constellation according to the time node in the step 1; synchronously acquiring a high-resolution satellite image of a high-resolution six-point satellite in a target area;

step 3.5: the construction of the cotton aphid pest multispectral identification method comprises the following steps: analyzing identification characteristics of multispectral cotton aphid insect pests according to the multispectral image data of the unmanned aerial vehicle obtained in the step 3.2 and the multispectral image data of the satellite obtained in the step 3.4, and constructing a multispectral cotton aphid insect pest identification method, wherein the construction method of the multispectral cotton aphid insect pest identification method comprises the following steps:

step 3.5.1: and (3) selecting a multispectral Aphis gossypii characteristic spectrum segment and an index: calculating NNIR of satellite multispectral image data and unmanned aerial vehicle multispectral image data according to the formula listed in step 2.4_mulThe indices (F ═ 1.26; df ═ 4, 12;^ap-0.0368), red edge index R_mul(550，760)(F＝1.57；df＝4,12；^aP-0.0460) and spectrum index beta of diseased plant_mul(F＝9.73； df＝4,12；^bP ═ 0.0008), where F is the ratio of the two mean squares, df is the factor degree of freedom, P determines the factor significance,^aP＜0.05；^bP＜0.001；

step 3.5.2 construction

Gamma is a yield coefficient; b is an adjustment coefficient;

step 3.5.3: aphid with cottonThe pernicious interpretation threshold is T₁I.e. having y < T₁Cotton aphids were not obvious; y is more than or equal to T₁When the aphid occurs, the cotton aphid occurs;

step 3.7: constructing a hyperspectral image cotton aphid recognition model: and 3, performing principal component analysis on the satellite hyperspectral image data obtained in the step 3.4, using the satellite hyperspectral image data and the image data subjected to data dimensionality reduction as low-pass parameters, using the panchromatic band of the fused ultra-resolution multispectral image data obtained in the step 3.6 as high-pass parameters for fusion to obtain fused high-resolution hyperspectral image data, performing data assimilation operation on the cotton canopy ground hyperspectral data and the fused high-resolution hyperspectral image data, and then adopting a deep learning algorithm to construct a hyperspectral image cotton aphid recognition model.

And 4, step 4: analyzing and processing the cotton growth climate data obtained in the step 1 to construct a cotton peanut length model;

step 4.1: step 4.1: model parameter establishment and collection: the model parameters comprise predicted images, the predicted images are obtained by processing the multi-source remote sensing data obtained in the step 3, physical parameters, chemical parameters, structural parameters and biological parameters are obtained according to the same time node of the multi-source remote sensing data obtained in the step 3, and the physical parameters, the chemical parameters, the structural parameters and the biological parameters comprise crop data, weather data, soil data and field management data which are collected on the spot; the specific steps for obtaining the prediction image comprise:

step 4.1.4: and performing inverse wavelet transform to obtain a predicted image.

Step 4.2: processing the multi-source remote sensing data collected in the step 3 to obtain a fused image; the specific steps for obtaining the fused image comprise:

step 4.2.3: and performing inverse tower shape transformation on the fused contrast pyramid to obtain a reconstructed image which is a fused image. Step 4.3: and (3) establishing an assimilation algorithm: adopting a 4Dvar assimilation algorithm to obtain assimilation data, wherein the algorithm formula is

The specific operation steps are as follows:

step 4.3.1: taking a mode forecast field as an initial estimation field;

step 4.3.2: initializing the updated field;

step 4.3.3: and (4) forecasting a plurality of steps forwards in the mode, taking the new forecasting field as an initial estimation field updated next time, and then returning to the step 4.3.1 to form a loop.

Step 4.4: step 4.3: and (3) calibrating model parameters: according to the parameters collected in the step 4.1, the uniform distribution U is formed by parameter maximum values_(min，max)Or calibrating the parameters by normal distribution, wherein the specific calibration formula is as follows: logL ═ logL_LAI+logL_yield；

logL_LAf＝-0.5(x-x_obs)^T∑^-1(x-x_obs)-0.5Klog(2π)-0.5log(Δ∑)；

Wherein L represents a likelihood function; x and x_obsTime series vectors formed by model simulation values and observation values respectively corresponding to the LAI; v represents a covariance matrix of LAI observations; k represents a vector dimension; Δ Σ represents a column-column value of Σ; y and Y_obsAn analog value and an observed value respectively representing the yield; σ represents the standard deviation of yield observations.

the cotton aphids are one of the main plant diseases and insect pests of cotton in the cotton field, under the conditions of other factors, meteorological conditions become key factors for determining the generation and diffusion changes of the cotton aphids, and temperature, humidity, rainfall and wind are important factors for influencing the generation and diffusion of the cotton aphids. Therefore, the relationship between aphis gossypii and meteorological factors needs to be studied to reveal the meteorological conditions and change rules of aphis gossypii (i.e. the influencing factor in this embodiment is meteorological factor).

Step 5.1: counting the number of cotton aphids: counting the cotton aphid population and the natural enemy number of cotton aphids of the whole plants and single leaves of the intercropping alfalfa cotton field in statistical sampling points every day according to the requirements of the technical specification of cotton aphids from the three-leaf period of cotton, and recording the spatial position information (including meteorological factors) of the plants;

and step 5.2: image space information statistics: according to the multi-source remote sensing data collected in the step 3, the spatial information characteristics on the image are counted, the evolution of the cotton aphid pest occurrence situation is analyzed, and the spectral reflectivity of the image is obtained according to the multi-source remote sensing data;

step 5.3: and (3) correlation analysis: establishing the relationship between the number of cotton aphids and the spectral reflectivity of the image, the rule of the occurrence time of the cotton aphids and the spectral reflectivity of the image, and the rule of the number of the cotton aphids in different periods and the cotton plant disease spectral index beta, the number of the cotton aphids and the meteorological factor obtained in the step 2.4 according to the data obtained in the step 5.1 and the step 5.2;

The specific construction formula for constructing the cotton aphid diffusion model in the step 5.4 comprises the following steps:

wherein p (x, y, t) is the diffusion coefficient;

wherein q (x, y, t) is the mobility coefficient;

Step 6: constructing a cotton aphid remote sensing forecasting model by adopting a 4DVar assimilation method for the multispectral image cotton aphid identification model and the hyperspectral image cotton aphid identification model obtained in the step 3, the cotton growth model obtained in the step 4 and the cotton aphid diffusion model obtained in the step 5;

step 6.2: and (3) selecting a cotton growth and aphid pest stress model, namely constructing the cotton aphid pest stress model by adopting a partial least square model based on the assimilation coupling system in the step 6.1 and cotton aphid data identified on the image obtained by the multispectral image cotton aphid identification model and the hyperspectral image cotton aphid identification model in the step 3, wherein a formula model is as follows:

X＝TP^T+E

Y＝UQ^T+F

wherein X is a prediction matrix of nxm; y is a response matrix of nxp; t is an n X l X projection matrix; u is a Y projection matrix of nxl; p is an m × l orthogonal load matrix; q is an orthogonal load matrix of p x l; the matrixes E and F are error items and are normal distribution random variables which obey independent same distribution;

where h (x) is the wavelet mother function and σ is the kernel function.

The embodiment shows that the method extracts the spectral data of the cotton canopy at the seedling stage, analyzes the spectrums of healthy cotton plants and cotton aphid harmful cotton plants, and constructs cotton aphid recognition models (a multispectral image cotton aphid recognition model, a hyperspectral image cotton aphid recognition model and a ground hyperspectral cotton aphid recognition model) based on spectral characteristics; constructing a cotton growth model by researching the phenological growth state of cotton; the method has the advantages that the aphid diffusion rule and the influence factors are researched, the aphid diffusion model is built, the cotton aphid remote sensing forecasting model is built by adopting a 4DVar assimilation method, the identification and diffusion of cotton aphids are accurately judged by a remote sensing technology, the pest occurrence condition and the pesticide application condition of cotton in a cotton field can be mastered, and therefore the method has important significance in guiding accurate pesticide application to kill the cotton aphids, saving pesticide application cost, improving ecological environment quality and the like.

The above-mentioned embodiments only express the specific embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present application. It should be noted that, for those skilled in the art, without departing from the technical idea of the present application, several changes and modifications can be made, which are all within the protection scope of the present application.

Claims

1. A method for constructing a cotton aphid remote sensing forecast model is characterized by comprising the following steps: the method comprises the following steps:

step 2: analyzing and processing the cotton canopy ground hyperspectral data obtained in the step 1, screening out a sensitive spectrum section, and constructing a ground hyperspectral aphis gossypii aphid recognition model according to the sensitive spectrum section;

2. The method for constructing the cotton aphid remote sensing forecasting model according to claim 1, characterized in that: in the step 1, high spectral data of the ground of the cotton canopy is obtained through a surface feature spectrometer, the height of an observation probe of the surface feature spectrometer from the cotton canopy is H, and the calculation formula of the height H is as follows:

wherein theta is the angle of view of the observation probe, and L is the row spacing of cotton planting.

3. The method for constructing the cotton aphid remote sensing forecasting model according to claim 1, characterized in that: the specific steps for obtaining the ground hyperspectral aphis gossypii aphid recognition model in the step 2 comprise:

wherein the derivative is taken

And

calculating a formula model;

wherein R is^1st(λ_i) In the spectral band lambda_iThe value of the first derivative of (A) above, R (λ)_i) Is the value of the reflectivity at the i-band, R (λ)_i+1) Is the value of the reflectivity at the (i +1) band; r is^2nd(λ_i) In the spectral region lambda_iA second derivative value of (d);

differential adoption

where ρ' (λ)_i) In the spectral region lambda_iThe first order differential value above; ρ (λ)_i+1) Is the value of the reflectivity at the (i +1) band; ρ (λ)_i-1) Is the value of the reflectivity at the (i-1) wavelength band; ρ' (λ ″)_i) In the spectral region lambda_iSecond order differential value of (d); ρ' (λ)_i+1) Is the first order differential value at the (i +1) band; ρ' (λ)_i-1) Is the first order differential value at (i-1) band; lambda [ alpha ]_iIs the wavelength value of the i wave band; lambda_i+1A wavelength value of (i +1) band;

step 2.4: and (3) index calculation: calculating NNIR index and red edge index R according to the cotton canopy ground hyperspectral data obtained in the step 2.2_(550，760)And disease plant spectral index beta;

wherein, the NNIR index adopts a formula model

A calculation is performed wherein NNIR is the vegetation index, ρ₇₆₀Is the reflectance value at the wavelength band of 760 nm; ρ is a unit of a gradient₆₅₀A reflectance value at a wavelength band of 650 nm; ρ is a unit of a gradient₅₅₀A reflectance value at a wavelength band of 550 nm;

red edge index R_(550，760)Using a formula model

Performing a calculation in which R_(550，760)Is a specific vegetation index; rho₇₆₀Is the reflectance value at a wavelength band of 760 nm; rho₅₅₀A reflectance value at a wavelength band of 550 nm;

disease plant spectral index beta adopts a formula model

Calculation was performed where NNIR is the vegetation index, R_(550，760)Is a specific vegetation index;

step 2.5: insect pest judgment: red edge index R calculated according to step 2.4_(550，760)Judging whether the cotton has insect damage and the degree of the insect damage according to the disease plant spectral index beta;

step 2.7: constructing a model: and (3) constructing a ground hyperspectral cotton aphid recognition model by adopting a wave spectrum angle model according to the characteristic spectrum section of the cotton aphid pests screened out in the step 2.6.

4. The method for constructing the cotton aphid remote sensing forecasting model according to claim 3, characterized by comprising the following steps: the specific steps for obtaining the multispectral image aphis gossypii identification model and the hyperspectral image aphis gossypii identification model in the step 3 comprise:

step 3.1: acquisition of remote sensing data of the unmanned aerial vehicle: acquiring an unmanned aerial vehicle cotton canopy multispectral data set of a normal cotton plant in a cotton seedling stage and a cotton plant stressed by aphids by using an unmanned aerial vehicle according to the time node in the step 1;

step 3.4: satellite remote sensing data processing: preprocessing the satellite high-resolution remote sensing data obtained in the step 3.3, completing geographical Registration by using an ENVI Registration tool through a second-order polynomial method to obtain satellite multispectral image data and satellite panchromatic image data, and preprocessing the satellite hyperspectral remote sensing data obtained in the step 3.3 to obtain satellite hyperspectral image data;

step 3.5.1: and (3) selecting a multispectral Aphis gossypii characteristic spectrum segment and an index: calculating the multispectral image data of the satellite according to the formula listed in the step 2.4NNIR with multispectral image data of unmanned aerial vehicle_mulIndices (F ═ 1.26; df ═ 4, 12;^ap-0.0368), red-edge index R_mul(550，760)(F＝1.57；df＝4，12；^aP-0.0460) and spectrum index beta of diseased plant_mul(F＝9.73；df＝4，12；^bP ═ 0.0008), where F is the ratio of the two mean squares, df is the factor degree of freedom, P determines the factor significance,^aP＜0.05；^bP＜0.001；

step 3.5.2 construction

A second-order polynomial regression model, wherein y is a cotton aphid pest interpretation index; phi, phi,

Gamma is the yield coefficient; b is an adjustment coefficient;

step 3.6: constructing a multispectral image cotton aphid recognition model: performing homologous fusion on the satellite multispectral image data obtained in the step 3.4 and the satellite panchromatic image data to obtain fused super-resolution multispectral image data, applying the cotton aphid insect pest multispectral identification model obtained in the step 3.5 to the fused super-resolution multispectral image to comprehensively construct a multispectral image cotton aphid identification model, wherein the construction method of the multispectral image cotton aphid identification model comprises the following steps of:

step 3.6.1: performing homologous fusion on the satellite multispectral image data and the satellite panchromatic image data to obtain fused super-resolution multispectral image data;

step 3.6.2: calculating NNIR (NNIR) of satellite multispectral image data based on formulas in step 2.4 and step 3.5.1_mulIndex, red edge index R_mul(550, 760) and spectrum index beta of diseased plants_mulSynchronously confirming the values of F, df and P, and keeping the values within a confidence interval that P is less than 0.001;

step 3.7: constructing a hyperspectral image cotton aphid recognition model: and 3, performing principal component analysis on the satellite hyperspectral image data obtained in the step 3.4, using the satellite hyperspectral image data and the image data subjected to data dimensionality reduction as low-pass parameters, using the panchromatic waveband of the fused super-resolution multispectral image data obtained in the step 3.6 as high-pass parameters for fusion to obtain fused high-resolution hyperspectral image data, performing data assimilation operation on the cotton canopy ground hyperspectral data and the fused high-resolution hyperspectral image data, and constructing a hyperspectral image aphis gossypii glover identification model by adopting a depth learning algorithm.

5. The method for constructing the cotton aphid remote sensing forecasting model according to claim 1, characterized in that: the specific steps of constructing the cotton growth model in the step 4 comprise:

step 4.3: and (3) calibrating model parameters: uniformly distributed U formed by parameter maximum values according to the parameters acquired in the step 4.1_(min.max)Or calibrating the parameters by normal distribution, wherein the specific calibration formula is as follows: logL ═ logL_LAI+logL_yield；

logL_LAI＝-0.5(x-x_obs)^T∑^-1(x-x_obs)-0.5Klog(2π)-0.5log(Δ∑)；

Wherein L represents a likelihood function; x and x_obsTime series vectors formed by model simulation values and observation values respectively corresponding to the LAI; Σ represents a covariance matrix of LAI observations; k represents a vector dimension; Δ Σ represents a determinant value of Σ; y and Y_obsAn analog value and an observed value respectively representing the yield; σ represents the standard deviation of the yield observation;

step 4.4: and (3) establishing an assimilation algorithm: carrying out 4Dvar assimilation algorithm on the model parameters obtained in the step 4.1 and the observation data obtained in the step 4.2 to obtain assimilation data, wherein the algorithm formula is

step 4.5: constructing a cotton growth model: and (4) constructing a cotton growth model according to the WOFOST model of Waherty root in the Netherlands by combining the assimilation data obtained in the step 4.4.

6. The method for constructing the cotton aphid remote sensing forecasting model according to claim 5, wherein the method comprises the following steps: the specific steps of obtaining a predictive image in step 4.1 include:

step 4.1.2: performing translation invariance wavelet transformation on the multi-source remote sensing data to obtain corresponding wavelet transformation low-frequency and high-frequency component coefficients with different resolutions and different directions;

7. The method for constructing the cotton aphid remote sensing forecasting model according to claim 5, characterized by comprising the following steps: the specific steps of obtaining the fusion image in the step 4.2 comprise:

step 4.2.2: respectively performing fusion processing on each decomposition layer of the image pyramid, adopting a weighted average operator for the low-frequency part of the decomposed image, and using a large operator for the high-frequency part to process to obtain a contrast pyramid of the fused image;

step 4.2.3: and performing inverse tower shape transformation on the fused contrast pyramid to obtain a reconstructed image which is a fused image.

8. The method for constructing the cotton aphid remote sensing forecasting model according to claim 3, characterized in that: the specific steps for constructing the cotton aphid diffusion model in the step 5 comprise:

step 5.1: data acquisition and processing: determining an influence factor of cotton aphid diffusion, acquiring and recording the number of cotton aphids, the number of natural enemies of the cotton aphids and relevant data of the influence factor for multiple times according to the time node in the step 1, and analyzing the relevant data of the influence factor, the number of the cotton aphids and the data of the natural enemies of the cotton aphids to obtain a rule of occurrence time of the cotton aphids, a rule of quantity of the cotton aphids at different periods and a rule of influence of the influence factor on the quantity of the cotton aphids;

step 5.2: image space information statistics: according to the multi-source remote sensing data collected in the step 3, the spatial information characteristics on the image are counted, the evolution of the cotton aphid pest occurrence situation is analyzed, and the spectral reflectivity of the image is obtained according to the multi-source remote sensing data;

step 5.3: and (3) correlation analysis: establishing a relation between the number of cotton aphids and the spectral reflectivity of the image, a rule of occurrence time of the cotton aphids and the spectral reflectivity of the image, a rule of the number of the cotton aphids in different periods and the cotton plant disease spectral index beta, the number of the cotton aphids and the influence factor obtained in the step 2.4 according to the data obtained in the step 5.1 and the step 5.2;

9. The method for constructing the cotton aphid remote sensing forecasting model according to claim 8, characterized in that: the formula model for constructing the cotton aphid diffusion model in the step 5.4 is as follows:

let u (x, y, t) be the number of Aphis gossypii at time (x, y) t, and K (u) be the population dynamics, then there is the formula:

k (u) a (x, y, t) u + b (x, y, t), wherein b (x, y, t) is the cotton aphid diffusivity;

wherein p (x, y, t) is the diffusion coefficient;

wherein e (x, y, t) is the cotton aphid birth rate; c (x, y, t) is the migration rate of Aphis gossypii; d (x, y, t) is the cotton aphid death rate;

wherein q (x, y, t) is the mobility coefficient;

10. The method for constructing the cotton aphid remote sensing forecasting model according to claim 1, characterized in that: the specific steps of constructing the cotton aphid remote sensing forecasting model in the step 6 comprise:

step 6.1: heterogeneous data assimilation: carrying out heterogeneous data assimilation on the multispectral image aphis gossypii glover identification model and the hyperspectral image aphis gossypii glover identification model obtained in the step 3, the cotton growth model obtained in the step 4 and the cotton aphid diffusion model obtained in the step 5 based on a 4DVar assimilation method to obtain an assimilation coupling system;

step 6.2: selecting a cotton growth and aphid pest stress model: based on the assimilation coupling system in the step 6.1 and the cotton aphid data identified on the image obtained from the multispectral image cotton aphid identification model and the hyperspectral image cotton aphid identification model in the step 3, a partial least square model is adopted to construct a cotton aphid pest stress model, and a formula model is as follows:

X＝TP^T+E

Y＝UQ^T+F

wherein X is a prediction matrix of nxm; y is a response matrix of nxp; t is an n X l X projection matrix; u is an n multiplied by l Y projection matrix; p is an m × l orthogonal load matrix; q is an orthogonal load matrix of p x l; the matrixes E and F are error terms and are normal distribution random variables which are independent and distributed in the same way;

step 6.3: constructing a cotton aphid remote sensing forecasting model: based on the assimilation coupling system obtained in the step 6.1 and the cotton aphid pest stress model obtained in the step 6.2, the system analyzes the relationship among the cotton aphid stress state, the cotton aphid diffusion state, cotton spectrum information and multi-source remote sensing data information in the cotton chemical growth process, and adopts a least square twin support vector machine method to construct a cotton aphid remote sensing forecasting model, wherein a formula model is as follows:

where h (x) is the wavelet mother function and σ is the kernel function.