CN114202702B - Based on D-fGWinter wheat dynamic harvest index remote sensing estimation method obtained by parameter remote sensing - Google Patents

Abstract

The invention discloses a method based on D-f_GThe invention discloses a method for estimating a dynamic harvest index of winter wheat by parameter remote sensing, which is based on field control experiments of winter wheat and provides dynamic parameters D-f of the ratio between aboveground biomass accumulated at different periods during the period from a flowering period to a mature period and aboveground biomass at corresponding periods_GOn the basis, narrow-band spectral indexes NDSI and D-f constructed by hyperspectral data of crop canopy_GScreening out D-f by the correlation relationship between_GThe estimated sensitive wave band center and the optimal wave band are combined, so that the high spectrum pair D-f of the canopy is utilized_GAccurate estimation of. Finally, based on D-f_GParameter remote sensing information and measured D-f_GAnd the quantitative relation between the measured Dynamic crop Harvest Index (D-HI) and the winter wheat crop Harvest Index realizes the optimal estimation of the winter wheat crop Harvest Index based on canopy hyperspectral remote sensing, so as to further improve the estimation precision and level of the winter wheat Harvest Index and provide theoretical basis and technical reference for the high-precision acquisition of the crop Harvest Index by using a remote sensing satellite technology in a large range.

Description

Remote sensing estimation method of winter wheat dynamic harvest index based on D-fG parameter remote sensing

technical field

The invention relates to a remote sensing estimation method of winter wheat dynamic harvest index based on canopy hyperspectral sensitive band Df _G parameter remote sensing acquisition.

Background technique

Harvest Index (HI), also known as economic coefficient, refers to the ratio of economic yield (grain, fruit) to biological yield when crops are harvested. The index reflects the distribution ratio of assimilation products between grains and vegetative organs. For food crops (such as wheat, corn, etc.), the harvest index is the percentage of grain yield in the above-ground biological yield, where the above-ground biological yield refers to the total dry matter mass above ground (Donald, 1962; Donald and Hamblin, 1976; Pan Xiaohua) et al., 2007). Since crop harvest index is widely used in crop yield simulation and estimation (Fan et al., 2017; Hu et al., 2019; Lorenz et al., 2010), crop variety selection (Hay, 1995; Rivera-Amado et al., 2019) ), crop growth and cultivation environment assessment (Porker et al., 2020; Yang and Zhang, 2010), crop carbon sequestration capacity assessment (Chen et al., 2021; Unkovich et al., 2010), and agricultural response to climate change, etc. It plays an important indicative role. Once its concept is proposed, it has become a research hotspot of scholars at home and abroad (Walter et al., 2018; Ji Xingjie et al., 2010).

At present, the estimation of crop harvest index is mainly studied from the field scale and the regional scale. In the field-scale crop HI estimation, some scholars mainly carry out the simulation estimation of crop harvest index from the perspective of agronomy and crop science and the impact of environmental stress factors (such as high temperature, water deficit, soil nutrient deficiency or excess, etc.) on the formation of crop harvest index. and other aspects for in-depth research (Fletcher and Jamieson, 2009; Kemanian et al., 2007; Soltani et al., 2005). For example, Fletcher and Jamieson (2009) conducted a study on the dynamic simulation of wheat harvest index over time and its influencing factors. The results showed that the change rate of wheat harvest index was closely related to the aboveground biomass of crops in the early stage of grain filling and the growth rate of crops during grain filling. Correlation, and the wheat harvest index shows a curve change with time, which has an important guiding role in the dynamic simulation and estimation of winter wheat harvest index. Kemanian et al. (2007) used wheat, barley and sorghum as research crops. According to the linear or curvilinear relationship between crop HI and the ratio of dry matter accumulation after flowering to the total dry matter mass of the whole growing season (f _G ), a field scale was established. The statistical model between f _G and HI enables accurate simulation and estimation of crop harvest index at the field scale. Meanwhile, Li et al. (2011) conducted a field control experiment of winter wheat under different nitrogen levels in Yucheng City, Shandong Province, China, using the measured data such as the ratio of dry matter accumulation after flowering to the total dry matter mass of the whole growing season (f _G ). The HI estimation method of winter wheat has been studied, and good crop harvest index simulation results have been obtained. The above research results have important reference significance for the estimation of crop harvest index using the _fG parameter, but because the above research only considers the _fG parameter and the harvest index of mature crops, and the dynamics of the _fG parameter and the harvest index are not considered. Changes have an impact on the accuracy of crop harvest index estimation and simulation, thus affecting the stability of harvest index estimation results and the further improvement of estimation accuracy to a certain extent. In addition, some scholars have carried out research on wheat HI estimation based on the proportion of evapotranspiration from flowering to maturity to the total evapotranspiration in the whole growth period (Li et al., 2011; Richards and Townley-Smith, 1987; Sadras and Connor, 1991), The method proposed in this study can effectively estimate the HI of winter wheat under the condition of water deficit, but under the condition of sufficient water and other environmental factors (such as nitrogen stress), the estimation of crop HI still needs to be further studied.

In the estimation of crop HI based on the regional scale, the traditional method adopts the point-to-surface method and the spatial interpolation method to obtain the regional crop harvest index (Ren Jianqiang et al. 2010). Among them, the point-to-surface method is to use the average value of the multi-year harvest index obtained by the fixed-point experiment as the regional harvest index; the spatial interpolation method is to spatially interpolate the actual survey multi-point crop harvest index to obtain the regional spatial distribution of the current year's harvest index. In recent years, with the rapid development of remote sensing technology, remote sensing technology has provided a reliable technical means for accurately obtaining the spatial information of regional crop harvest index by virtue of its advantages of large coverage, rapid and accurate acquisition of surface crop parameter information (Campoy et al. ., 2020; Walter et al., 2018). Among them, scholars at home and abroad have carried out a series of studies on the estimation of crop harvest index based on the time-series vegetation remote sensing information (such as normalized vegetation index and leaf area index, etc.) obtained by remote sensing satellites that can reflect the growth status of crops (Li et al., 2011). ; Moriondo et al., 2007). For example, Moriondo et al. (2007) divided the whole growth period of winter wheat into two stages: germination-blooming and flowering-maturity. According to the average NDVI value of the two periods before and after flowering, a model 1-NDVI _post /NDVI _pre was constructed to estimate the spatial distribution of HI. The method can obtain NDVI data of winter wheat growing season through remote sensing, which has important reference significance for using remote sensing information to obtain regional-scale HI. At the same time, this method has also been further applied by Chinese scholars. For example, Du et al. (2009) used MERIS NDVI time series data to invert and verify the regional winter wheat harvest index in Yucheng, Shandong, and applied the regional winter wheat harvest index to crop yield estimation. Research. Ren Jianqiang et al. (2010) took winter wheat in the Huang-Huai-Hai Plain area of China as the research object, and used the ratio of the NDVI cumulative value at the flowering stage-milk maturity stage and the turning green-before-flowering NDVI cumulative value to characterize the winter wheat harvest index. By establishing the ratio and the measured harvest The inter-index statistical model estimated well the harvest index of winter wheat at the regional scale. The above methods are simple and easy to implement, and the required remote sensing data time series are short and easy to obtain, which is beneficial to the practical application of the method. However, the above methods are only aimed at estimating the harvest index at the mature stage, and none of them can realize the dynamic change process index information of the harvest index change. acquisition, which requires further research.

Therefore, the prior art has shortcomings and needs to be improved.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to provide a remote sensing estimation method for the dynamic harvest index of winter wheat based on the canopy hyperspectral sensitive band Df _G parameter remote sensing acquisition based on the shortcomings of the prior art.

The technical scheme of the present invention is as follows:

A remote sensing estimation method of winter wheat dynamic harvest index based on Df _G parameter remote sensing acquisition, comprising the following steps:

A1, according to the measured dynamic biomass data on the ground, construct the ratio dynamic parameter _DfG between the aboveground biomass accumulated in different periods during the flowering period-mature period of the crop and the aboveground biomass in the corresponding period;

Df _G is calculated as follows:

In the formula, ∑W _post is the aboveground biomass (kg/hm ² ) accumulated in different periods between the flowering period and the maturity period of winter wheat;

∑W _whole is the total aboveground biomass corresponding to the sampling period (kg/hm ² ); t is the sampling time, W _t is the weight of the dry matter mass at the t sampling time (kg/hm ² ), and W _a is the dry matter mass at the flowering stage. Weight (kg/hm ² ), Df _{G, t} represents the ratio parameter of t sampling time;

A2. Based on any two canopy hyperspectral narrow-band spectral indices NDSI constructed based on ground crop canopy hyperspectral data, a linear model between NDSI and winter wheat Df _G was established;

A3. Then draw and analyze the two-dimensional graph of the fitting accuracy R ² between NDSI and winter wheat Df _G ;

A4. By determining the maximum value area of R ² and the center of gravity of the maximum value area, the band center sensitive to Df _G of winter wheat is obtained;

A5. Determine the optimal band combination for Df _G estimation;

A6. Df _G remote sensing estimation model based on the relationship between NDSI and Df _G ;

A7. Remote sensing estimation of Df _G ;

A8. Obtain a dynamic harvest index estimation model based on the relationship between Df _G and D-HI;

Remote sensing estimation of A9, D-HI.

In the described method, in step A4, the center of the sensitive band is obtained according to the method of the center of gravity of the maximum value region of R ² , so as to determine the sensitive band estimated by Df _G , that is, the narrow band spectral index NDSI and NDSI corresponding to each band of the canopy hyperspectral spectrum. On the basis of the calculation of the correlation between Df _G , the maximum value area is determined according to the threshold value of the correlation coefficient satisfying the statistical significance requirement _. Large spectral band center and band combination; the specific process is as follows:

First, on the basis of drawing a two-dimensional graph of fitting R ² between NDSI and winter wheat Df _G , the band area with high correlation between NDSI and winter wheat Df _G was determined ^; Traverse all the points that satisfy the significance condition in the neighborhood of this point 8, and mark the set of these points as the R ² maximal value region Ω; finally, calculate the barycenter of the R ² maximal value region as each R ² pole The center of the sensitive band in the large value area; the calculation formula (7) of the center of gravity is as follows:

敏感波段中心坐标。In the formula, f(u, v) is the R ² value with the band coordinates (u, v), Ω is the maximum value area,

The coordinates of the center of the sensitive band.

In the method, in the step A4, the 6 sensitive band centers sensitive to winter wheat Df _G include λ(443nm, 506nm), λ(442nm, 635nm), λ(732nm, 834nm), λ(787nm, 804nm), λ (810 nm, 877 nm) and λ (861 nm, 985 nm).

Described method, described step A6, the Df _G remote sensing estimation model based on NDSI and Df _G relation is:

Df _G,t = m×NDSI _i,j,t +n (5)

Among them, i and j are the hyperspectral bands between 350nm and 1000nm respectively, t is the different sampling time, m and n are the fitting parameters in the linear equation obtained after fitting; according to this formula, the crop flowering period to the t period can be calculated. The ratio parameter Df _G,t between the accumulated above-ground biomass and the above-ground biomass in the t growth period.

Described method, described step A8, obtain the dynamic harvest index estimation model based on Df _G and D-HI relation is:

D-HI _t =HI ₀ +s×Df _G,t (6)

Among them, HI ₀ is the intercept, that is, the value of the dynamic harvest index when the biomass does not change after the flowering period of the crop, that is, when Df _{G, t} is 0, the value of the D-HI harvest index; s is the D-HI Slope constant in linear relationship with Df _G.

Described method, described step A8, according to the dynamic data of aboveground biomass data of winter wheat and the dynamic data of grain yield in the grain filling process of different collection times during the flowering period-maturity period, calculate the Df _G and the dynamic harvest index of 128 sample points in the winter wheat plot. D-HI, on this basis, use formula (6) to fit the correlation between Df _G and dynamic harvest index D-HI, and obtain an estimation model between Df _G and dynamic harvest index D-HI, as follows:

D-HI _t =0.1018+0.8093*Df _G,t

A winter wheat yield estimation method based on a winter wheat dynamic harvest index, wherein the winter wheat dynamic harvest index is obtained by any of the above methods.

The inventive method has the following beneficial effects:

(1) Development from static f _G parameters to dynamic f _G parameters

The previous studies on the estimation of crop harvest index based on _fG only considered the ratio between the cumulative aboveground biomass at the flowering stage and the mature stage of the crop and the aboveground biomass at the mature stage, and this parameter did not consider the dynamic change of grain yield during the crop growth process. Therefore, This parameter is a static parameter. Considering that in the existing research on crop harvest index estimation based on static f _G (Sf _G ), due to the short time period of modeling data and validation data, it may lead to the problem of low stability of the harvest index estimation model. In this case, the present invention further develops the general static parameter Sf _G into a dynamic parameter Df _G (Dynamic f _G ) that considers the ratio between the above-ground biomass accumulated in different growth stages during the flowering period-maturity period and the above-ground biomass in the corresponding period . Since the dynamic Df _G during the flowering period and maturity of the crop is considered, it can take into account the information of the dynamic change process of crop growth to a certain extent, and can increase the number of modeling samples per year, so that the modeling data of shorter years can be used to obtain stability and stability. Crop harvest index estimation model and estimation results with a high level of precision.

(2) A technical method to obtain _fG parameter information based on remote sensing technology is proposed

The general way to obtain _fG is to manually observe the aboveground biomass of crops at the field scale. On the basis of developing static _SfG into dynamic _DfG , the present invention obtains NDSI remote sensing information through canopy hyperspectral, and then uses the center of gravity model to Df _G remote sensing estimation sensitive bands are screened to realize remote sensing estimation of Df _G parameters. The _DfG remote sensing estimation technology method based on the canopy spectral sensitive band center to construct NDSI proposed by the present invention can lay a technical foundation for remote sensing of UAVs and remote sensing acquisition of _DfG parameters using remote sensing satellite technology in a large range, and also provide a basis for field scale based remote sensing. The upscaling application of the crop harvest index estimation method of measured _SfG provides new ideas and technical methods.

(3) The remote sensing estimation technology method of dynamic harvest index based on dynamic f _G parameter remote sensing acquisition is proposed.

In view of the existing traditional crop harvest index estimation based on f _G parameter, it is all realized by ground measured data, but cannot use remote sensing information to realize the upscaling regional application of this method, and the existing researches using remote sensing information to estimate crop harvest index generally only focus on crop maturity. However, the lack of consideration of the dynamic change process of the crop _harvest index affects the further improvement of the estimation accuracy of the crop harvest index to a certain extent. This paper proposes a dynamic crop harvest index remote sensing estimation method that considers crop biomass changes and grain formation processes in different periods of grain filling and maturity, which improves the stability and accuracy of the crop _harvest index remote sensing estimation model to a certain extent. The parametric crop harvest index estimation method cannot use the remote sensing information for the bottleneck of upscaling application, and realizes the high-precision remote sensing estimation of the dynamic harvest index based on the dynamic f _G parameter remote sensing.

Description of drawings

Figure 1 shows the location of the research area and the layout of the test plot;

Fig. 2 is the hyperspectral curve of winter wheat canopy in different growth stages of the experimental plot;

Figure 3 is a technical roadmap;

Figure 4 is a schematic diagram of the center determination of the remote sensing sensitive band for estimating crop Df _G based on NDSI;

Fig. 5 is the linear relationship model between Sf _G and maturity harvest index G-HI;

Fig. 6 is the verification of the estimation accuracy of crop harvest index at the mature stage based on the measured Sf _G ;

Figure 7 shows the two-dimensional distribution of NDSI in winter wheat under normal horizontal fertilization and irrigation (N2W2) (2020); a. flowering stage (May 10), b. early grain filling (May 18), c. middle grain filling (5 24th of June), d. late stage of grain filling (3rd of June), e. of maturity (19th of June);

Figure 8 is a ^two -dimensional diagram of fitting R2 between NDSI and winter wheat _DfG ;

Fig. 9 is the fitting R ² two-dimensional contour map between NDSI and winter wheat Df _G ;

Figure 10 shows the construction of the model between NDSI and Df _G based on the center of the sensitive band;

Figure 11 is the verification of the Df _G estimation result based on the center of the sensitive band;

Figure 12 shows the establishment of a D-HI estimation model based on Df _G remote sensing parameters;

Fig. 13 is the D-HI estimation result verification based on the hyperspectral sensitive band Df _G parameter;

Fig. 14 is the verification of the mature stage D-HI estimation result based on the hyperspectral sensitive band Df _G parameter;

Sensitive band centers corresponding to each of Figure 10, Figure 11, Figure 13, Figure 14: a (443nm, 506nm), b (442nm, 635nm), c (732nm, 834nm), d (787nm, 804nm), e ( 810nm, 877nm), f (861nm, 985nm);

Detailed ways

The present invention will be described in detail below with reference to specific embodiments.

1 Materials and methods

1.1 Study area and experimental design

The field test of the present invention is set in the agricultural environment comprehensive test demonstration base in Shunyi District, Beijing (116°92'～116°94'E, 40°05'～40°06'N). The study area belongs to the warm temperate semi-humid continental monsoon climate, with an average annual temperature of about 11.5 °C, an average annual precipitation of about 625 mm, an annual sunshine of about 2750 hours, and a frost-free period of about 195 days. The main planting system in this study area is winter wheat-summer maize with two crops a year. The location map of the study area is shown in Figure 1.

In order to obtain observation crops of winter wheat with different growth states and growth states, considering that the growth and yield of winter wheat in the study area are mainly affected by fertilization and irrigation, especially nitrogen and soil moisture have important controlling effects on the growth and yield of winter wheat, therefore, the present invention The design experiment mainly considered two factors, nitrogen fertilizer and water. Winter wheat trials were conducted from October 2019 to June 2020. The wheat variety used in the experiment was Lunxuan 987, and the base fertilizers were superphosphate and potassium sulfate (P ₂ O ₅ was 135kg/hm ² , K ₂ O was 90kg/hm ² ). Four nitrogen supply levels were set in the experiment: N0 (no nitrogen fertilizer application), N1 (nitrogen application rate of 160kg/hm ² ), N2 (nitrogen application rate of 240kg/hm ² ), N3 (nitrogen application rate of 320kg/hm ² ) . The amount of nitrogen fertilizer was applied twice, that is, half of the fertilizer amount was applied at the time of sowing as the base fertilizer and the top-dressing during the greening period, and the top-dressing time was the greening-jointing period of winter wheat. The irrigation water treatment in the whole growth period was divided into four levels: W0 (0mm), W1 (100mm), W2 (150mm), W3 (200mm). Among them, W2 is the total amount of normal irrigation of local winter wheat in the whole growth period. Winter wheat was irrigated for a total of 4 times. The irrigation time was before winter, greening period, jointing period and booting period. On average, each plot was irrigated 0 mm, 25 mm, 37.5 mm and 50 mm per time. Other cultivation and management of winter wheat trials are consistent with the local traditional winter wheat management measures. During the whole growth period of winter wheat, a total of 4 different nitrogen application rates and 4 different water treatments were set, with a total of 16 treatments, each treatment was set with 3 replicates, and a total of 48 experimental plots, each with an area of 30m ² (5m × 6m). ). In the present invention, the sowing time of the control experiment is October 9, 2019, and the heading-flowering period of winter wheat is early May, 2020, grain filling-milk maturity (mid-May to early June, 2020), maturity (2020). mid-June).

1.2 Data acquisition and processing

The ground data collection mainly includes the observation of the above-ground fresh biomass, above-ground dry biomass and winter wheat canopy hyperspectral data of winter wheat in each plot. Data collection work. Finally, the present invention was launched on May 10, 2020 (flowering stage), May 18 (pre-filling stage), May 24 (mid-filling stage), June 3 (late-filling stage) and June 19 (mature stage), 2020. ) carried out a total of 5 ground data collection work. There were 48 plots in the experiment. According to the arrangement of 2 sampling points in each plot, a total of 5 samples were collected. In order to ensure the quality of data collection, 10 wheat plant areas (about 1 row 20 cm), and affixed labels as markers for the collection of winter wheat aboveground fresh biomass and canopy hyperspectral data at each sampling point during the flowering-maturity period of wheat. In order to accurately obtain the aboveground fresh biomass data, aboveground dry biomass data and canopy spectral data of each plot, the data obtained from 2 sample points in each plot were averaged to improve the ground level involved in modeling and model validation. Observation data quality.

(1) Acquisition of aboveground biomass

During each ground observation, the aerial parts of winter wheat with a row length of 20 cm were taken as samples at the marked points in each plot, and then the collected winter wheat samples were sealed in ziplock bags and transported back to the laboratory for analysis. In the laboratory, the total fresh weight of wheat at each sampling point was first weighed and recorded; secondly, the stems, leaves and ears of each sampling point were separated, put into paper bags and the fresh weight of the stems, leaves and ears was weighed; third, the separated The wheat stems, leaves and spikes were placed in an oven at 105 °C for 30 min, and then the samples were dried at 85 °C to a constant weight, and the dry weight weighing results of stems, leaves and spikes at each plot observation point were recorded. Finally, the dry biomass weight of each plot point was obtained.

(2) Obtaining the Dynamic Harvest Index

On the basis of obtaining the dry weight of winter wheat stems, leaves and panicles of 20cm row length at each plot sampling point, the wheat panicles at each sampling point were threshed respectively, and then the grain weight of each sample point was weighed and recorded. Finally, the winter wheat harvest index was calculated at each ground observation time during the grain filling process of winter wheat. During the grain filling process, the crop harvest index is gradually formed and the harvest index also exhibits dynamic changes with time. Therefore, the present invention refers to the winter wheat harvest index at each ground observation time during the grain filling process of winter wheat as the Dynamic Harvest Index (Dynamic Harvest Index, D-HI).

In the formula, t is the ground sampling time from grain filling to maturity, W _z,t , W _{J, t} , W _{Y, t} , W _{S, t} are the t sampling time from grain filling to maturity of winter wheat grains, stems and leaves, respectively. , dry weight of ear, D-HI _t is the dynamic harvest index of t sampling time from grain filling to maturity.

(3) Measurement of canopy hyperspectral data

The canopy hyperspectral measurement mainly uses the American ASD Field Spec 4 spectroradiometer to collect the ground spectrum in the uniform growth area in the 48 cells. The measurement range of the spectrometer is 350-2500nm. Among them, the sampling interval in the wavelength of 350-1000nm is 1.4nm, the sampling interval in the wavelength of 1000nm-2500nm is 2nm, and the data interval after resampling is 1nm. A standard whiteboard is used for calibration before each measurement. The probe is vertically downward during measurement. The viewing angle of the probe of the spectroscopic device is 25°, and the height of the probe from the top of the crop canopy is about 0.5m. Two sampling points are taken for each cell, and five spectral data are read at the optimal time interval of each sampling point, and the average value is taken as the spectral reflectance value of the cell to reduce noise interference and randomness. In the present invention, the crop canopy spectrum collection is carried out under the conditions of local time 10:00-14:00, good weather conditions and sufficient sunlight.

The preprocessing of canopy hyperspectral data mainly includes spectral averaging and spectral smoothing. Among them, the average value of spectral data is processed by ViewSpecPro software, and the average value is taken as the reflectance spectral value of the corresponding sampling point. Spectral smoothing mainly uses the 9-point weighted moving average method in ENVI software to smooth and denoise the spectral reflectance data. Finally, the hyperspectral reflectance data of winter wheat canopy at each sampling point in the observation plot was obtained. Figure 2 shows the canopy hyperspectral curves of winter wheat at different growth stages after the spectra of the 48 experimental plots were averaged and smoothed.

1.3 Research methods

1.3.1 Basic Concepts

(1) Dynamic Harvest Index

The general harvest index (Harvest Index, HI) only considers the percentage of grain yield in the total shoot dry matter at the mature stage (Donald and Hamblin 1976), which is the maximum crop harvest index, that is, the final harvest index. In order to improve the estimation accuracy of the crop harvest index and improve the stability of the crop harvest index model, the present invention not only considers the harvest index at the maturity stage, but also considers the gradual formation of the crop harvest index and the dynamic information of the harvest index that changes with time. Dynamic Harvest Index (D-HI). In order to distinguish the proposed D-HI index, the present invention defines the general harvest index as G-HI (General Harvest Index). For food crops (such as wheat, corn, etc.), the dynamic harvest index refers to the dynamic change process of the gradually increasing crop grain yield as a percentage of the dry matter mass of the crop from the beginning of the formation of the crop grain and from the gradual grain filling to the mature process. The index increased with the gradual increase of crop growth and development time after grain formation until reaching the maximum harvest index.

(2) Dynamic f _G parameter

The general crop flowering stage-maturity ratio between the above-ground biomass and the mature above-ground biomass ratio parameter f _G is a static parameter (Static f _G , Sf _G ) and is only used in the calculation of f _G at the mature stage, and lacks the flowering stage- Research on the dynamic process of _fG parameters between maturation periods, which leads to the unstable relationship between the _fG static parameters obtained by the short-year experiment and the crop harvest index, which reduces the estimation of the crop harvest index to a certain extent. precision. In order to improve the stability and estimation accuracy of the crop harvest index estimation model, the present invention proposes a dynamic _fG index based on the original static _fG parameter, that is, considering the aboveground biomass accumulated in different periods during the flowering period and the mature period and the corresponding period Aboveground biomass ratio dynamic parameter Df _G (Dynamic f _G ). The calculation method of this indicator Df _G is as follows:

In the formula, ∑W _post is the aboveground biomass accumulated in different periods during the flowering period and mature period of winter wheat (kg/hm ² ); ∑W _whole is the total aboveground biomass corresponding to the sampling period (kg/hm ² ); t is the sampling period Time, W _t is the weight of dry matter mass at t sampling time (kg/hm ² ), W _a is the weight of dry matter mass at flowering stage (kg/hm ² ), Df _G,t is the ratio parameter of t sampling time.

(3) Canopy Hyperspectral Narrow Band Spectral Index NDSI

A large number of crop remote sensing monitoring studies have shown that there is a strong correlation between the normalized vegetation index (NDVI) and crop aboveground biomass and crop grain yield (Ren Jianqiang et al., 2015; Feng Meichen et al., 2010). The crop harvest index remote sensing estimation has also been applied to a certain extent and obtained good research results. Therefore, the present invention also uses the most commonly used normalized vegetation index to carry out the dynamic harvest index remote sensing estimation research. The formula for calculating NDVI is:

NDVI=(ρ _nir -ρ _red )/(ρ _nir +ρ _red ) (3)

where nir and red represent the near-infrared and red bands, respectively, and ρ _nir and ρ _red represent the spectral reflectance in the near-infrared and red bands, respectively. When the near-infrared band reflectance and the red light band reflectivity are not limited to the near-infrared region and red light region of the electromagnetic spectrum, but are combined for any two bands of the hyperspectral spectrum, the normalized spectral index of the narrow band of the hyperspectral spectrum can be used. , NDSI) to represent, as follows:

NDSI=(ρ _i -ρ _j )/(ρ _i +ρ _j ) (4)

In the formula, i and j are the wavelengths corresponding to the hyperspectral bands, respectively, and ρ _i and ρ _j are the spectral reflectances corresponding to the i and j wavelengths, respectively. Among them, the NDSI value range is [-1,1]. In order to facilitate the study of the correlation between the crop canopy hyperspectral narrow-band spectral index and f _G , considering that the crop canopy spectrum is greatly affected by the atmosphere and water vapor at 1350-1415 nm and 1800-1950 nm, and the present invention is mainly aimed at visible light-near Therefore, the present invention performs remote sensing sensitive band screening for Df _G estimation and remote sensing estimation of dynamic crop harvest index in the 350-1000 nm band range (including 650 bands).

1.3.2 Overall technical route

Firstly, according to the measured biomass data on the ground, the ratio parameters between the above-ground biomass accumulated during the flowering-maturity period and the above-ground biomass in the mature period were constructed as the static parameter (Sf _G ) and the above-ground biomass accumulated in different periods during the flowering-maturity period. The dynamic parameter (Df _G ) of the ratio of aboveground biomass to the corresponding period. Then, based on any two canopy hyperspectral narrowband spectral indices (NDSI) constructed from ground crop canopy hyperspectral data, a linear model between NDSI and winter wheat Df _G was established; then the relationship between NDSI and winter wheat Df _G was drawn and analyzed. Fitting precision R ² two-dimensional map; on this basis, the band center sensitive to winter wheat f _G was obtained by determining the R ² maximum value area and the center of gravity of the maximum value area; finally, the dynamic harvest index D-HI and The best model for parameter Df _G. Secondly, according to the measured biomass data on the ground, a harvest index estimation model based on _SfG was constructed. Finally, the reserved harvest index validation data set is used for verification, and the accuracy of the crop harvest index estimation model constructed by the two methods is compared and analyzed. The specific technical route is shown in Figure 3.

1.3.3 Estimation of crop dynamic harvest index based on hyperspectral sensitive band Df _G parameters

1.3.3.1 Construction of crop dynamic harvest index estimation model

The invention proposes a dynamic harvest index D-HI remote sensing estimation method based on Df _G remote sensing information. Taking the dynamic parameter Df _G as the intermediate variable, firstly determine the model between NDSI and Df _G , and then determine the band center sensitive to winter wheat f _G according to its model; then determine the statistical relationship model between Df _G and the dynamic harvest index D-HI; finally, According to the selected sensitive band center, the corresponding Df _G can be determined, then the crop harvest index can be estimated, and the accuracy can be verified. The calculation method of the crop dynamic harvest index is as follows:

Df _G,t = m×NDSI _i,j,t +n (5)

D-HI _t =HI ₀ +s×Df _G,t (6)

Equation (5) is mainly used for model building of NDSI and _DfG . Among them, i and j are the hyperspectral bands between 350nm and 1000nm respectively, t is the different sampling time, m and n are the fitting parameters in the linear equation obtained after fitting. According to this formula, the ratio parameter Df _G,t between the above-ground biomass accumulated from the flowering period to the t period and the above-ground biomass of the t growth period was calculated.

Equation (6) is mainly used for model building between Df _G and D-HI. Among them, HI ₀ is the intercept, that is, the value of the dynamic harvest index when the biomass does not change after the flowering period of the crop, that is, when Df _{G, t} is 0, the value of the D-HI harvest index; s is the D-HI Slope constant in linear relationship with Df _G. According to this formula, the dynamic harvest index in the t growth period can be calculated.

1.3.3.2 Determination of the center of the canopy hyperspectral sensitive band estimated by Df _G remote sensing

The invention utilizes remote sensing technology to obtain information on the ratio dynamic parameter _DfG between the aboveground biomass accumulated in different periods during the flowering period and the mature period of the crop and the aboveground biomass in the corresponding period, so as to utilize the correlation between the measured _DfG and the crop dynamic harvest index to realize Accurate estimation of crop harvest index based on _DfG remote sensing information. In the research, based on the obtained winter wheat canopy hyperspectral and the measured Df _G parameters of winter wheat, the remote sensing estimation of winter wheat Df _G based on the hyperspectral remote sensing narrow-band spectral index NDSI was carried out. Due to the large number of hyperspectral data bands and the high correlation between the bands, the redundancy of spectral information increases. In order to improve the accuracy of the Df _G parameter remote sensing estimation model, the present invention needs to use the narrow band spectral index NDSI which is sensitive to the Df _G parameter. Filter by band center and spectral band.

Since in the two-dimensional map of R ² between NDSI and Df _G , the maximum value region of R ² is not uniformly distributed, the center of gravity of the maximum value point of R ² and the maximum value region of R ² may not completely coincide, resulting in the extreme value of R ² The band corresponding to the large value point does not necessarily coincide with the center of the optimal band. Therefore, in order to ensure that the NDSI constructed by using the selected band center can obtain a high-precision Df _G estimation result, thereby making the crop harvest index estimation result more stable, the present invention obtains the sensitive band center according to the R ² maximum value area center of gravity method, thereby Determine the sensitive band for _DfG estimation, that is, on the basis of the correlation calculation between the narrow-band spectral index NDSI and _DfG corresponding to each band of the canopy hyperspectrum, determine the maximum value region according to the threshold that the correlation coefficient meets the requirements of statistical significance, On this basis, the center of gravity of the region with the maximum value of the correlation coefficient is calculated, so as to obtain the spectral band center and band combination with greater correlation between the NDSI and Df _G parameters. The specific process is as follows:

First, on the basis of drawing a two-dimensional graph of fitting R ² between NDSI and winter wheat Df _G , the band area with high correlation between NDSI and winter wheat Df _G was determined ^; Traverse all the points that satisfy the significance condition in the neighborhood of this point 8, and mark the set of these points as the R ² maximal value region Ω; finally, calculate the barycenter of the R ² maximal value region as each R ² pole The center of the sensitive band in the area of large value. The schematic diagram of determining the center of the Df _G sensitive band of winter wheat is shown in Figure 4. The calculation formula (7) of the center of gravity is as follows:

敏感波段中心坐标。In the formula, f(u, v) is the R ² value with the band coordinates (u, v), Ω is the maximum value area,

The coordinates of the center of the sensitive band.

1.3.4 Estimation of crop harvest index (G-HI) at maturity based on measured Sf _G

In order to compare with the estimation accuracy of the crop harvest index before the improvement, the present invention uses the proportional coefficient (Sf _G ) of the dry matter accumulation from flowering to maturity of crops proposed by Kemanian et al. Harvest index (G-HI) linear relationship model is used to estimate the harvest index G-HI of mature crops based on the measured Sf _G. The linear model form is as follows:

G-HI=HI ₀ +k×Sf _G (8)

In the formula, G-HI is the crop harvest index at the harvest period, HI ₀ is the intercept in the linear relationship; k is the slope in the linear relationship; Sf _G is the dry matter accumulation during the flowering-maturity period of the crop in the entire growth period. proportion of the total cumulative amount.

1.3.5 Accuracy Evaluation of Crop Harvest Index Estimation Model

In order to evaluate the estimation accuracy of f _G (including Sf _G and Df _G ) and the estimation accuracy of harvest index (including G-HI and D-HI) in the estimation process of crop harvest index, the present invention selects coefficient of determination (R ² ), Root mean square error (RMSE), normalized root mean square error (NRMSE) and mean relative error (MRE) were used to test the accuracy of the harvest index estimation model. Among them, R ² represents the degree of fitting between the simulated value and the measured value, and the value ranges from 0 to 1. The higher the R ² is, the closer it is to 1, the better the fitting effect of the model is. On the contrary, the lower the R ² is, the closer it is to 0, the worse the fitting effect of the model is. RMSE indicates the degree of deviation between the simulated value and the measured value. The larger the root mean square error value, the greater the deviation between the simulated value and the actual measured value, that is, the worse the simulation effect; on the contrary, the smaller the root mean square error value, the better the simulation effect. . NRMSE refers to the ratio of the root mean square error to the mean of the measured values. MRE refers to the average value of the sum of the absolute values of the relative errors, indicating the average deviation of the simulated value from the measured value. When NRMSE and MRE<10%, the accuracy of simulation results is judged to be excellent; when 10%<NRMSE and MRE<20%, the accuracy of simulation results is judged to be good; when 20%<NRMSE and MRE<30%, it is judged that the simulation results are accurate The accuracy of the results is general; when NRMSE and MRE>30%, the accuracy of the simulation results is judged to be poor, and the judgment standard gives priority to the size of the NRMSE value. The specific formula is as follows:

分别为x_i，y_i的均值，n 为样本数。In the formula, x _i is the measured value of f _G (such as Sf _G and Df _G ) or HI (such as G-HI and D-HI); y _i is the winter wheat f _G (such as Sf _G and Df _G ) or HI (such as G-HI and D-HI) estimates;

are the mean of x _i and y _i respectively, and n is the number of samples.

2 Results and Analysis

2.1 Estimation of crop harvest index (G-HI) at maturity based on measured Sf _G

In the study, based on the aboveground biomass data and wheat grain yield per unit area of winter wheat at the maturity stage on June 19, the average measured harvest index of winter wheat at the maturity stage of each plot was calculated; Using biomass as the benchmark, the Sf _G parameter was calculated to obtain the ratio of dry matter accumulation in winter wheat to the total accumulation in the entire growth period during the flowering-maturity period. On this basis, according to formula (8), a model was constructed based on the proportion of dry matter accumulation in the flowering-maturity period of winter wheat to the total accumulation in the entire growth period, _SfG , and the harvest index at maturity. In the study, according to the principle of 2:1 ratio between the modeling data set and the validation data set, the measured Sf _G and G-HI of one of the winter wheat field control experiment repeated treatment groups were randomly used as the validation data set (16 data samples in total). , and the remaining 2 experiments were repeated with the measured Sf _G and G-HI of the treatment group as the modeling data set (32 data samples in total). Among them, it can be seen from Figure 5 that the coefficient of determination of the linear model constructed by Sf _G and harvest index HI is 0.5058. The RMSE is 0.0603, and the NRMSE and MRE are 11.78% and 11.31%, respectively, as shown in Figure 6.

2.2 Remote sensing estimation of crop harvest index based on hyperspectral sensitive band Df _G parameters

2.2.1 Calculation results of hyperspectral NDSI of winter wheat canopy under different control experimental treatments

Based on the preprocessing of the crop canopy hyperspectral data collected from different experimental treatment plots for each ground observation, the narrow band spectral index (NDSI) of any combination of two bands in each experimental treatment plot for each ground observation is calculated and plotted according to formula (4). ). Among them, in the hyperspectral range of 350-1000nm, there are a total of 650×650 combinations between any two bands and related NDSI values. The present invention only shows the canopy hyperspectral NDSI results calculated from 5 ground observation data of normal level fertilization and irrigation treatment (N2W2) in the winter wheat control experiment. The five NDSI distribution maps shown in Fig. 7 are the NDSI calculation results of different growth stages of winter wheat during the normal level of fertilization and irrigation (N2W2) in the control experiment of winter wheat at flowering, pre-filling, late-filling, late-filling and mature stages. Among them, the abscissa (λ ₁ ) and the ordinate (λ ₂ ) are the hyperspectral wavelengths of the crop canopy, and the wavelength range is 350-1000 nm. The points corresponding to the two-dimensional space formed by the horizontal and vertical axes are any two bands λ ₁ , The calculated NDSI value of the reflectivity corresponding to λ ₂ .

2.2.2 NDSI-based Df _G estimation canopy hyperspectral sensitive band center determination

(1) Two-dimensional distribution map of R ² based on the correlation between NDSI and Df _G

The present invention first calculates and obtains data indicators such as NDSI and Df _G of each cell on May 18, May 24, June 3, and June 19. The calculation of Df _G is based on May 10. The aboveground biomass of winter wheat at flowering stage was used as the benchmark. On this basis, a statistical relationship model between NDSI and Df _G was constructed. Among them, 48 sub-districts conducted four ground observations in total, and finally 192 sub-district ground observation data samples were accumulated, and the data indicators included NDSI and Df _G . According to the principle that the ratio of modeling data set and verification data set is 2:1, one of the winter wheat field control experiments was randomly calculated to obtain NDSI and Df _G as the verification data set (a total of 64 plots of ground observation data samples), and the rest Two experiments were repeated for processing groups to obtain NDSI and Df _G as the modeling data set (a total of 128 cell ground data samples).

Finally, using Matlab software, a two-dimensional graph of the fitting accuracy R ² of NDSI and Df _G was obtained (as shown in Figure 8), where the abscissa (λ ₁ ) and the ordinate (λ ₂ ) are the interval in the wavelength range of 350-1000 nm For the crop canopy hyperspectral wavelength of 1 nm, the horizontal and vertical axes form a total of 650×650 points corresponding to the two-dimensional space of R ² , and each point in the two-dimensional space of R ² corresponds to a combination of two bands (λ ₁ , λ ₂ ) The fitting accuracy R ² between the constructed NDSI value and the measured Df _G. Among them, each R ² two-dimensional space point is the coefficient of determination of the linear model constructed by the NDSI constructed by the combination of two bands of a certain wavelength in 128 cells and the corresponding cell Df _G. It can be seen from Figure 8 that the fitting accuracy R ² is axisymmetrically distributed with the line connecting the two points (350, 350) and (1000, 1000), from which we can obtain the region and related band information where NDSI has a high correlation with Df _G. On the upper side of the symmetry axis between the two points (350, 350) and (1000, 1000), it can be seen from the dark red part of the dotted box in Figure 8 that there are two main areas where NDSI and winter wheat Df _G are highly correlated. , the region includes a two-dimensional region ranging from 370 to 550 nm on the horizontal axis and 480 to 720 nm on the vertical axis, 750 to 970 nm on the horizontal axis and 770 to 1000 nm on the vertical axis, where R ² reaches more than 0.75, which is the _DfG estimate of canopy hyperspectral sensitivity The determination of the band center has laid the foundation.

(2) Df _G estimation canopy hyperspectral sensitive band center determination

In the study, by looking up the correlation coefficient significance test table, when the sample size n=128, at the 0.05 significance level, when R ² >0.030, NDSI and Df _G showed a significant correlation; at the 0.01 significance level, When R ² >0.051, NDSI and Df _G showed a very significant correlation. In order to ensure the accuracy and reliability of the determination of the center of the sensitive band, the present invention selects a two-dimensional region of R ² that conforms to R ² >0.051 for research, finds the maximum value point of R ² >0.051 in Figure 8, and traverses this point 8 All points with R ² >0.051 in the neighborhood, mark the set of these points as the maximum value area, and display the distribution area of R ² with R ² =0.1 as the gradient. In order to more intuitively display the distribution range of sensitive bands, here The results for R ² ≥ 0.6 are displayed, as shown in FIG. 9 .

In order to improve the accuracy of estimating Df _G at the center of the selected hyperspectral sensitive band, and further improve the accuracy of estimating the harvest index, the present invention selects the R ² two-dimensional region with R ² >0.8 to determine the center of the sensitive band. Ω _A to Ω _F are the maximum value region of R ² satisfying R ² >0.8, and the specific results are as follows: Ω _A ranges from 400 to 500 nm on the horizontal axis and 480 to 540 nm on the vertical axis; Ω _B ranges from 370 to 550 nm on the horizontal axis and the vertical axis between 550 and 720 nm; the range of Ω _C is between 720 and 750 nm on the horizontal axis and between 720 and 980 nm on the vertical axis; the range of Ω _D is between 770 and ₈₁₀ nm on the horizontal axis and between 790 and 820 nm on the vertical axis; The range of Ω _F is between 750-970 nm on the horizontal axis and 950-1000 nm on the vertical axis. After determining the sensitive areas of NDSI and Df _G , calculate the center (λ ₁ , λ ₂ ) of each R ² maximum area according to formula (7), where λ ₁ is the abscissa of the center of the maximum area, λ ₂ The ordinate of the center of the maximum value area. Finally, the center of gravity of the maximal region of Ω _A ~ Ω _F is obtained as A (443nm, 506nm), B (442nm, 635nm), C (732nm, 834nm), D (787nm, 804nm), E (810nm, 877nm) and F (861 nm, 985 nm).

2.2.3 Df _G remote sensing estimation of NDSI based on the canopy spectral sensitive band center

The canopy hyperspectral narrow-band spectral index NDSI was calculated according to the Df _G screened out in 2.2.2 to estimate the canopy hyperspectral sensitive band center. Among them, the center results of 6 sensitive bands include λ(443nm, 506nm), λ(442nm, 635nm), λ(732nm, 834nm), λ(787nm, 804nm), λ(810nm, 877nm) and λ(861nm, 985nm) . Then, according to formula (5), the NDSI data of 128 plots of ground observation data samples and the winter wheat Df _G were used to build a linear model, and the remaining 64 plots of ground observation data samples were used as verification data sets for accuracy verification. The statistical relationship between NDSI and Df _G constructed based on the center of the sensitive band and the specific results of Df _G estimation accuracy are shown in Table 1, Figure 10 and Figure 11. It can be seen from Table 1 that the NDSI fitting Df _G constructed by the selected 6 Df _G estimated canopy hyperspectral sensitive band centers all reached a very significant level at the level of P < 0.01, and the model determination coefficient R ² was between 0.8138 and 0.8138. between 0.8613. The established statistical model between NDSI and Df _G is verified by the reserved validation data set, and it can be seen that the statistical models between NDSI and Df _G constructed by the selected 6 Df _G estimated canopy hyperspectral sensitive band centers have good performance. The Df _G estimation effect and the estimated Df _G have reached a high level of accuracy. Among them, the RMSE is between 0.0267 and 0.0411, the NRMSE is between 9.27% and 14.27%, and the MRE is between 8.81% and 14.26%. Among them, the NDSI constructed by the selected sensitive band centers (732nm, 834nm) has the highest accuracy in estimating the _DfG of winter wheat, and its determination coefficient R ² reaches 0.9522, and the sums of NRMSE and MRE are 9.27% and 8.81%, respectively; The NDSI constructed by the sensitive band center (443nm, ^506nm ) has a high accuracy in estimating the _DfG of winter wheat. , 985nm) constructed NDSI to estimate _DfG of winter wheat with relatively low accuracy, its coefficient of determination R ² was 0.8981, and the sum of NRMSE and MRE were 14.27% and 14.26%, respectively.

Table 1 Statistical relationship between NDSI and Df _G based on the center of sensitive band and Df _G estimation accuracy

Note: In the fitting equation, x is the NDSI constructed by the bands λ ₁ and λ ₂ , and y is the fitted winter wheat Df _G . N is the number of samples.

** indicates a very significant correlation at the p<0.01 level.

2.2.4 Establishment and verification of the D-HI remote sensing estimation model based on the acquisition of Df _G parameters in the hyperspectral sensitive band

Based on the correlation between NDSI and Df _G , the present invention uses canopy hyperspectral to screen out Df _G parameters to estimate the sensitive band center; then, the NDSI index constructed by the sensitive band center is used to realize accurate Df _G remote sensing estimation. On this basis, based on the measured statistical relationship model between Df _G and D-HI, the remote sensing estimation of dynamic harvest index D-HI is realized by using Df _G remote sensing parameter information. Finally, the D-HI remote sensing estimation model is verified using the reserved verification data.

2.2.4.1 Establishment of D-HI estimation model based on Df _G remote sensing parameters

According to the dynamic data of aboveground biomass of winter wheat and the dynamic data of grain yield during the period of flowering and maturity at different collection times, the Df _G and dynamic harvest index D-HI of 128 sample points in the winter wheat plot were calculated. Formula (6) fits the correlation between Df _G and dynamic harvest index D-HI, and obtains an estimation model between Df _G and dynamic harvest index D-HI, as follows:

D-HI _t =0.1018+0.8093*Df _G,t

The research shows that there is a significant linear relationship between the measured winter wheat Df _G and the crop dynamic harvest index in the present invention, wherein the coefficient of determination of the linear model constructed by the dynamic Df _G and the dynamic harvest index D-HI reaches 0.9679 (Fig. 12). It lays a good foundation for the estimation of dynamic harvest index based on dynamic Df _G.

2.2.4.2 Validation of D-HI estimation model based on Df _G remote sensing parameters

Based on the establishment of the D-HI estimation model based on the Df _G remote sensing parameters, the NDSI corresponding to the center of each remote sensing sensitive band is calculated using the spectral information in the reserved 64 sets of verification data. Among them, each band center can obtain 64 NDSI substitution data. On this basis, the above NDSI is substituted into the statistical model between the corresponding NDSI and Df _G in Table 1, so as to obtain 64 Df _G remote sensing parameters corresponding to the center of each sensitive band. Then, the above Df _G remote sensing parameters are substituted into the D-HI remote sensing estimation model based on the Df _G parameters in Figure 12, so as to obtain 64 D-HI estimation results at the center of each sensitive band, and the accuracy of the D-HI remote sensing estimation results is carried out. verify. Among them, the reserved 64 D-HI data were used to evaluate the accuracy of the remote sensing estimation results of D-HI under different sensitive band conditions.

(1) Overall accuracy verification of winter wheat dynamic harvest index

The present invention is in 6 remote sensing sensitive bands such as λ(443nm, 506nm), λ(442nm, 635nm), λ(732nm, 834nm), λ(787nm, 804nm), λ(810nm, 877nm) and λ(861nm, 985nm). Under the condition of _DfG estimated by NDSI constructed by the center, the accuracy of the dynamic crop harvest index results of remote sensing estimation of _DfG parameters in each canopy hyperspectral sensitive band center was verified. Finally, the overall verification results of the dynamic harvest index of winter wheat at different sampling periods from flowering to maturity are shown in Figure 13 and Table 2. It can be seen from Table 2 that under the condition of _DfG estimated by NDSI constructed by the 6 remote sensing sensitive band centers, the _DfG parameter remote sensing information obtained by the canopy hyperspectral can realize the accurate estimation of the dynamic crop harvest index. From the results of the overall accuracy evaluation index of D-HI estimation, it can be seen that under the condition of the center of the selected 6 canopy hyperspectral sensitive bands, the D-HI estimation results based on the Df _G parameter of the hyperspectral sensitive band have reached a high level of accuracy. The fitting accuracy R ² is between 0.9169 and 0.9584, the RMSE is between 0.0380 and 0.0507, the NRMSE is between 10.83% and 14.45%, and the MRE is between 9.62% and 13.99%. Among them, the D-HI estimation result based on the hyperspectral sensitive band center λ(732nm, 834nm) estimated the Df _G parameter with the highest accuracy, with NRMSE and MRE of 10.83% and 9.62%, respectively; secondly, based on the hyperspectral sensitive band center λ(443nm) , 506nm), the D-HI estimation results for estimating _DfG parameters are highly accurate, with NRMSE and MRE of 11.60% and 10.24%, respectively; D-HI estimation of _DfG parameters based on the center λ of the hyperspectral sensitive band (861nm, 985nm) The accuracy of the measurement results is relatively low, with NRMSE and MRE of 14.45% and 13.99%, respectively.

Table 2 Overall accuracy verification of D-HI estimation model based on Df _G remote sensing parameters

(2) Accuracy verification of the D-HI estimation model for winter wheat at different growth stages from grain filling to maturity

In addition to using the reserved 64 sets of verification data to analyze the overall accuracy of remote sensing estimation of the dynamic harvest index of winter wheat in different sampling periods between the flowering period and the maturity period, the present invention also targets the data on May 18, May 24, June 3, and June respectively. The D-HI estimation model results of the pre-grouting, mid-grouting, and post-grouting corresponding to different sampling dates such as the 19th were evaluated for accuracy. The specific results are shown in Tables 3-6. Among them, among the 64 sets of data reserved, there are 16 sets of verification data corresponding to each sampling date.

A. Verification of the accuracy of the D-HI estimation model for winter wheat at different grain filling stages

In the early stage of winter wheat filling (Table 3), the fitting accuracy between the harvest index estimated by remote sensing and the measured crop harvest index R ² was between 0.3128 and 0.5819, the RMSE was between 0.0273 and 0.0393, and the NRMSE was between 13.21% and 19.04%. MRE was between 11.50% and 17.94%. Among them, the Df _G parameter estimated based on the hyperspectral sensitive band center λ(732nm, 834nm) has the highest accuracy of D-HI estimation results in the pre-grouting period, with NRMSE and MRE of 13.21% and 11.50%, respectively; secondly, based on the hyperspectral sensitive band center λ (443nm, 506nm) The D-HI estimation results of Df _G parameters are highly accurate, with NRMSE and MRE of 13.82% and 11.92%, respectively; D-HI estimation of Df _G parameters based on the hyperspectral sensitive band center λ (861nm, 985nm) The accuracy of HI estimation results is relatively low, with NRMSE and MRE of 19.04% and 17.94%, respectively.

In the mid-filling stage of winter wheat (Table ⁴ ), the fitting accuracy R2 between the harvest index estimated by remote sensing and the measured crop harvest index was between 0.3597 and 0.4391, the RMSE was between 0.0321 and 0.0437, and the NRMSE was between 11.27% and 15.36%. MRE was between 9.05% and 13.18%. Among them, the Df _G parameter estimated based on the hyperspectral sensitive band center λ(732nm, 834nm) has the highest accuracy in the pre-grouting D-HI estimation results, with NRMSE and MRE of 11.27% and 9.05%, respectively; secondly, based on the hyperspectral sensitive band center λ (443nm, 506nm) D-HI estimation results of Df _G parameters are highly accurate, with NRMSE and MRE of 11.69% and 9.17%, respectively; D-HI estimation of Df _G parameters based on the hyperspectral sensitive band center λ (861nm, 985nm) The accuracy of HI estimation results is relatively low, with NRMSE and MRE of 15.36% and 13.18%, respectively.

At the late grain filling stage of winter wheat (Table 5), the fitting accuracy R2 between the harvest index estimated by remote sensing and the measured crop harvest index was between 0.4928 and ^0.5964 , the RMSE was between 0.0396 and 0.0565, and the NRMSE was between 9.92% and 14.16%. MRE was between 8.66% and 13.40%. Among them, the Df _G parameter estimated based on the hyperspectral sensitive band center λ(732nm, 834nm) has the highest accuracy of D-HI estimation results in the early stage of grouting, with NRMSE and MRE of 9.92% and 8.66%, respectively; secondly, based on the hyperspectral sensitive band center λ (443nm, 506nm) D-HI estimation results of Df _G parameters are highly accurate, with NRMSE and MRE of 10.92% and 9.95%, respectively; D-HI estimation of Df _G parameters based on the hyperspectral sensitive band center λ (861nm, 985nm) The accuracy of HI estimation results is relatively low, with NRMSE and MRE of 14.16% and 13.40%, respectively.

B. Comparison of accuracy of D-HI remote sensing estimation model for mature winter wheat with traditional methods

At the maturity stage of winter wheat (Table 6 and Figure 14), the fitting accuracy R2 between the harvest index estimated by remote sensing and the measured crop harvest index was between 0.2724 and ^0.6762 , the RMSE was between 0.0492 and 0.0601, and the NRMSE was between 9.62% and 11.74%. The MRE was between 9.27% and 11.43%. Among them, the mature-stage D-HI estimation results based on the hyperspectral sensitive band center λ(732nm, _834nm ) have the highest accuracy, with NRMSE and MRE of 9.62% and 9.27%, respectively; secondly, based on the hyperspectral sensitive band center λ (443nm, 506nm) D-HI estimation results of Df _G parameters are highly accurate, with NRMSE and MRE of 10.33% and 9.92%, respectively _. The accuracy of HI estimation results is relatively low, with NRMSE and MRE of 11.74% and 11.43%, respectively. By comparing with the traditional G-HI estimation results of mature stage based on measured Sf _G in section 2.1, the method proposed in the present invention estimates the harvest index of the mature stage based on the hyperspectral sensitive band center λ (732nm, 834nm) to estimate the Df _G parameter Compared with the traditional G-HI estimation accuracy of measured Sf _G , the accuracy is improved by 0.0111, the NRMSE is improved by 2.16%, and the MRE is improved by 2.04%, which shows the effectiveness of the present invention using remote sensing technology to improve the traditional G-HI estimation method of measured Sf _G .

In general, the dynamic harvest index at different filling stages (pre-filling, mid-filling and late-filling stage) and the harvest index at maturity obtained by the D-HI remote sensing estimation method based on the hyperspectral sensitive band Df _G parameter obtained by the present invention all reach relatively high levels. High-level accuracy results, and the highest accuracy of the dynamic harvest index obtained in different growth periods is in the early stage of grain filling < middle stage of grain filling < late stage of grain filling < mature stage. The above precision evaluation results fully demonstrate the feasibility of the D-HI remote sensing estimation method proposed in the present invention. It is of great significance to accurately obtain the dynamic harvest index information and the harvest index at maturity by considering the dynamic growth information of crops. In addition, compared with the common remote sensing estimation results of crop harvest index at the mature stage (Moriondo et al., 2007; Ren Jianqiang et al., 2010;), the estimation results of the harvest index at maturity obtained by the method of the present invention and the results of common remote sensing estimation methods have the same Consistent, for example, from the D-HI estimation accuracy of Df _G parameter estimation using the hyperspectral sensitive band center, the sensitive band center λ (732nm, 834nm) with the highest D-HI estimation accuracy is mainly located in the red and near-infrared bands , and the NDSI composed of the above-mentioned centers is basically the same as the NDVI band combination used in the traditional estimation of crop harvest index, which proves that the D-HI remote sensing estimation method based on the hyperspectral sensitive band Df _G parameter obtained by the present invention is different from the previous crop estimation method using NDVI. Harvest index estimates have some consistency.

Table 3 D-HI estimation model based on Df _G remote sensing parameters for pre-filling harvest index accuracy verification (May 18)

Note: N represents the number of samples of D-HI measured before grouting in the validation dataset. ** means extremely significant correlation at p<0.01 level, * means significant correlation at p<0.05 level.

Table 4 Accuracy verification of the harvest index in the mid-filling stage of the D-HI estimation model based on Df _G remote sensing parameters (May 24)

Note: N represents the number of samples of D-HI measured in the mid-filling stage in the validation dataset. ** means extremely significant correlation at p<0.01 level, * means significant correlation at p<0.05 level.

Table 5 Accuracy verification of harvest index in the later stage of grain filling based on D-HI estimation model based on Df _G remote sensing parameters (June 3)

Note: N represents the number of samples of D-HI measured at the later stage of grouting in the validation dataset. ** means extremely significant correlation at p<0.01 level, * means significant correlation at p<0.05 level.

Table 6 Accuracy verification of harvest index at maturity of D-HI estimation model based on Df _G remote sensing parameters (June 19)

Note: N represents the number of samples of D-HI measured at maturity in the validation dataset. ** means extremely significant correlation at p<0.01 level, * means significant correlation at p<0.05 level.

It should be understood that, for those skilled in the art, improvements or changes can be made according to the above description, and all these improvements and changes should fall within the protection scope of the appended claims of the present invention.

Claims (7)

1. a winter wheat dynamic harvest index remote sensing estimation method based on Df _G parameter remote sensing acquisition, is characterized in that, comprises the following steps: A1, according to the measured dynamic biomass data on the ground, construct the ratio dynamic parameter _DfG between the aboveground biomass accumulated in different periods during the flowering period-mature period of the crop and the aboveground biomass in the corresponding period; Df _G is calculated as follows:

In the formula, ∑W _post is the aboveground biomass accumulated in different periods during the flowering period and mature period of winter wheat (kg/hm ² ); ∑W _whole is the total aboveground biomass corresponding to the sampling period (kg/hm ² ); t is the sampling period time, W _t is the weight of the dry matter mass at the t sampling time (kg/hm ² ), W _a is the weight of the dry matter mass at the flowering stage (kg/hm ² ), and Df _G,t is the ratio parameter of the t sampling time; A2. Based on any two canopy hyperspectral narrow-band spectral indices NDSI constructed based on ground crop canopy hyperspectral data, a linear model between NDSI and winter wheat Df _G was established; A3. Draw and analyze the two-dimensional graph of fitting accuracy R ² between NDSI and winter wheat Df _G ; A4. By determining the maximum value area of R ² and the center of gravity of the maximum value area, the band center sensitive to Df _G of winter wheat is obtained; A5. Determine the optimal band combination for Df _G estimation; A6. Df _G remote sensing estimation model based on the relationship between NDSI and Df _G ; A7. Remote sensing estimation of Df _G ; A8. Obtain the dynamic harvest index estimation model based on the relationship between Df _G and the dynamic harvest index D-HI; Remote sensing estimation of A9, D-HI. 2. The method according to claim 1, wherein, in the step A4, the center of the sensitive band is obtained according to the R ² maximal value area centroid method, thereby determining the sensitive band estimated by Df _G , that is, in each canopy hyperspectral spectrum. On the basis of the correlation calculation between the narrow-band spectral indices NDSI and Df _G corresponding to each band, the maximum value region is determined according to the threshold value of the correlation coefficient satisfying the requirement of statistical significance. On this basis, the center of gravity of the maximum value region of the correlation coefficient is calculated, Thus, the spectral band center and band combination with greater correlation between NDSI and Df _G parameters are obtained; the specific process is as follows: First, on the basis of drawing a two-dimensional graph of fitting R ² between NDSI and Df _G of winter wheat, determine the band area with high correlation between NDSI and Df _G of winter wheat ^; Traverse all the points that satisfy the significance condition in the neighborhood of this point 8, and mark the set of these points as the R ² maximal value region Ω; finally, calculate the barycenter of the R ² maximal value region as each R ² pole The center of the sensitive band in the large value area; the calculation formula (7) of the center of gravity is as follows:

In the formula, f(u, v) is the R ² value with the band coordinates (u, v), Ω is the maximum value area,

The coordinates of the center of the sensitive band. 3. method according to claim 1, is characterized in that, described step A4, 6 sensitive band centers sensitive to winter wheat Df _G comprise λ (443nm, 506nm), λ (442nm, 635nm), λ (732nm, 834 nm), λ (787 nm, 804 nm), λ (810 nm, 877 nm) and λ (861 nm, 985 nm). 4. method according to claim 1, is characterized in that, described step A6, the Df _G remote sensing estimation model based on NDSI and Df _G relation is: Df _G,t = m×NDSI _i,j,t +n (5) Among them, i and j are the hyperspectral bands between 350nm and 1000nm respectively, t is the different sampling time, m and n are the fitting parameters in the linear equation obtained after fitting; according to this formula, the crop flowering period to t period can be calculated and obtained. The ratio parameter Df _G,t between the accumulated aboveground biomass and the aboveground biomass in the t growth period. 5. method according to claim 1, is characterized in that, described step A8, obtains the dynamic harvest index estimation model based on Df _G and D-HI relation is: D-HI _t =HI ₀ +s×Df _G,t (6) Among them, HI ₀ is the intercept, that is, the value of the dynamic harvest index when the biomass does not change after the flowering period of the crop, that is, when Df _{G, t} is 0, the value of the D-HI harvest index; s is the D-HI Slope constant in linear relationship with Df _G. 6. method according to claim 5, is characterized in that, described step A8, calculates winter wheat plot 128 according to the dynamic data of aboveground biomass data of winter wheat and the grain yield dynamic data in filling process of different collection time during flowering stage-maturity stage. Df _G and dynamic harvest index D-HI of each sample point, on this basis, use formula (6) to fit the correlation between Df _G and dynamic harvest index D-HI, and obtain Df _G and dynamic harvest index D -Inter-HI estimation model, as follows: D-HI _t =0.1018+0.8093*Df _G,t . 7. A winter wheat yield estimation method based on a winter wheat dynamic harvest index, characterized in that, the winter wheat dynamic harvest index is obtained by any one of the methods of claims 1-6.

Images