Construction method of pea high-flux salt tolerance comprehensive scoring model based on unmanned aerial vehicle

Abstract

The invention relates to the technical field of agricultural growth detection and unmanned aerial vehicle remote sensing, in particular to a construction method of a pea high-flux salt tolerance comprehensive scoring model based on an unmanned aerial vehicle. According to the method, the unmanned aerial vehicle carries a sensor to collect pea image data in a target area, structure information of plant height and canopy coverage, texture information and spectral information are extracted according to the sensor image data, and a vegetation index is obtained through spectral information combination; then estimating pea biomass and SPAD through texture information and vegetation indexes by using a machine learning algorithm, and finally constructing a pea salt tolerance comprehensive scoring model PSTS by using four indexes of plant height, canopy coverage, biomass and SPAD. According to the method, the scoring model is constructed through sensor data fusion and a proper algorithm, so that the salt-tolerant pea varieties can be accurately screened in a high-throughput manner.

Description

Technical Field

[0001] The present invention belongs to the technical fields of agricultural growth detection and unmanned aerial vehicle (UAV) remote sensing, and particularly relates to a method for constructing a comprehensive scoring model for high-throughput salt tolerance of peas based on UAVs. Background Art

[0002] The area of saline-alkali land is vast and has great utilization potential. Screening salt-tolerant crop varieties is an important means to improve and utilize saline-alkali land. Peas are one of the important edible legume crops, with a short growth period and certain salt tolerance. Timely and accurately screening salt-tolerant pea crop varieties plays a very important role in improving saline-alkali land. However, the traditional method of screening salt-tolerant crop varieties is time-consuming and laborious, and is easily affected by human factors. With the development of UAV and remote sensing technologies, due to its high portability, the ability to carry a variety of sensors and being less affected by cloud cover, UAVs are increasingly used for monitoring crop phenotypic traits.

[0003] Sensors commonly carried by UAVs include RGB sensors, MS sensors, TIR sensors, hyperspectral sensors and lidar sensors. Lidar and hyperspectral sensors can usually provide more information, but their costs are relatively high, making it difficult to be applied on a large scale. TIR sensors can provide canopy temperature information and are commonly used to monitor the spatio-temporal changes of soil moisture. RGB sensors and MS sensors have been favored by many researchers due to their low costs and simple operations. However, current UAV-based research remains in the estimation of biological phenotypic traits that can be obtained through observation or measurement, lacking a reliable method for identifying salt tolerance in physiological phenotypes based on UAVs. Summary of the Invention

[0004] Aiming at the above technical problems, the present invention provides a method for constructing a comprehensive scoring model for high-throughput salt tolerance of peas based on UAVs. The method is simple and easy to implement, the identification results are more reliable, and it can achieve the evaluation of the salt tolerance of pea varieties in a high-throughput and accurate manner, improving the efficiency of screening salt-tolerant pea germplasm.

[0005] The technical solution of the present invention is as follows:

[0006] The present invention provides a method for constructing a high-throughput comprehensive salt tolerance scoring model of peas based on unmanned aerial vehicles. Pea images of the target area are collected, and the structural information, texture information, and spectral information in the pea images are extracted. The plant height and canopy coverage are estimated through the structural information. Vegetation index information related to biomass and SPAD is obtained through the spectral information, and then the biomass and SPAD of peas are estimated by using machine learning algorithms through the texture information and / or vegetation index information. The salt tolerance coefficients of the plant height, canopy coverage, biomass, and SPAD are calculated, normalized, and then principal component analysis is performed to obtain the principal components and determine the principal component weights. A comprehensive salt tolerance scoring model is constructed based on the principal component weights and the membership function values of the principal components. The magnitude of the salt tolerance factor is positively correlated with the salt tolerance.

[0007] In one embodiment, the pea image acquisition period is from the flowering stage to the filling stage of peas.

[0008] In one embodiment, the unmanned aerial vehicle is equipped with an RGB sensor and / or a multispectral sensor to obtain the pea images of the target area, and the pea images include RGB images and / or MS images.

[0009] In one embodiment, the texture information includes entropy, contrast, inverse variance, variance, correlation, dominance, gray level mean, and energy.

[0010] In one embodiment, the vegetation index information related to biomass in the RGB sensor includes WI (Warbeck index), GLI (green leaf area index), VARI (visible light atmospheric impedance vegetation index), ExR (excess red index), ExB (excess blue index), IPCA (principal component analysis index), CIVE (color vegetation index), and the vegetation index information related to SPAD includes INT (color intensity index), WI (Warbeck index), ExG (excess green index), ExB (excess blue index), IPCA (principal component analysis index), MRBVI (modified red-blue vegetation index), VARI (visible light atmospheric impedance vegetation index).

[0011] In one embodiment, the vegetation index information related to biomass in the MS sensor includes CVI (Chlorophyll Vegetation Index), CI (Chlorophyll Index), GDVI (Generalized Difference Vegetation Index), EVI2 (Two-Band Enhanced Vegetation Index), EVI (Enhanced Vegetation Index), GARI (Green Atmospherically Resistant Vegetation Index), GLI (Green Leaf Area Index), BNDVI (Blue Normalized Difference Vegetation Index), GRVI (Green Ratio Vegetation Index), MCARI1 (Modified Chlorophyll Absorption Reflectance Index), MCARI2 (Modified Chlorophyll Absorption Reflectance Index 2), SAVI (Soil-Adjusted Vegetation Index), OSAVI (Optimized Soil-Adjusted Vegetation Index), MCARI1 / OSAVI, and the vegetation index information related to SPAD includes CVI (Chlorophyll Vegetation Index), CCCI (Canopy Chlorophyll Content Index), GNDVI (Green Normalized Difference Vegetation Index), NDRE (Normalized Difference Red Edge Index), PNDVI (Projected Normalized Difference Vegetation Index), GSAVI (Green Soil-Adjusted Vegetation Index), GOSAVI (Green Optimized Soil-Adjusted Vegetation Index), NREI (Normalized Red Edge Index), NNRI (Normalized Near-Infrared Redness Index), MNDI (Modified Normalized Difference Infrared Index), MEVI (Modified Enhanced Vegetation Index), MCARI1 (Modified Chlorophyll Absorption Reflectance Index), MTCI (Terrestrial Chlorophyll Index), GDVI (Generalized Difference Vegetation Index).

[0012] In one embodiment, the Catboost algorithm is used in the machine learning algorithm to estimate the biomass of peas, and the Light GBM algorithm is used to estimate the SPAD of peas.

[0013] In one embodiment, use R 2 and RMSE to evaluate the robustness of the biomass and SPAD of the peas.

[0014] In one embodiment, the comprehensive salt tolerance scoring model is PSTS = ∑(i = 1, n)[U(X ij ) × w i , where PSTS represents the salt tolerance factor, U(X ij ) represents the membership function value, X ij represents the i-th principal component value of the j-th plot, and w i represents the weight of the i-th principal component.

[0015] The present invention also provides a method for screening salt-tolerant pea varieties, using the above-mentioned drone-based high-throughput comprehensive salt tolerance scoring model for peas to evaluate the salt tolerance of peas, and selecting pea varieties with high comprehensive salt tolerance evaluation.

[0016] Advantages of the present invention over the prior art:

[0017] 1. The comprehensive salt tolerance scoring model of the present invention for peas has estimated five salt-tolerant pea varieties, and four salt-tolerant pea varieties have been selected through ground measurement data. Four of the five varieties of the present invention are the same as the 4 varieties selected through ground measurement data, which proves that the scoring model constructed by the present invention through sensor data fusion and appropriate algorithms can evaluate the salt tolerance of pea varieties with high accuracy.

[0018] 2. The model based on the unmanned aerial vehicle constructed by the present invention can achieve high-throughput evaluation of the salt tolerance of pea varieties, and improve the efficiency of screening salt-tolerant pea germplasm. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 Scatter plots of the best results for estimating pea biomass (RM + Catboost) and SPAD (RM + LightGBM) in saline-alkali plots and control plots, where (a) is the scatter plot of estimating pea biomass in the control plot based on RGB + MS data and the Catboost algorithm, (b) is the scatter plot of estimating pea biomass in the saline-alkali plot based on RGB + MS data and the Catboost algorithm; (c) is the scatter plot of estimating pea SPAD in the control plot based on RGB + MS data and the Light GBM algorithm; (d) is the scatter plot of estimating pea SPAD in the saline-alkali plot based on RGB + MS data and the Light GBM algorithm;

[0020] Figure 2 Intuitive diagrams of the comprehensive salt tolerance scoring of peas based on ground measurement data and the model constructed by the present invention based on unmanned aerial vehicles, where (a) is the display diagram of the comprehensive salt tolerance scoring of peas based on ground measurement data; (b) is the display diagram of the comprehensive salt tolerance scoring of peas based on the estimated data of the model of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0021] The present invention provides a method for constructing a high-throughput salt tolerance comprehensive scoring model of peas based on unmanned aerial vehicles.

[0022] In the present invention, different varieties of peas are sown in the target area, and the sowing depth is 4 - 6 cm, and the row spacing of sowing is 30 - 50 cm. It provides more favorable growth conditions for the growth of peas, improves the emergence rate and seedling quality, and prevents the inconsistent growth of plants caused by sowing differences from affecting the salt tolerance evaluation.

[0023] The growth process of peas includes the seedling stage, flowering stage, pod-setting stage, filling stage and maturity stage. The period from the flowering stage to the filling stage is an important period for pea yield formation. The present invention collects pea images during the flowering stage to filling stage of peas, which can improve the accuracy of screening salt-tolerant varieties and make the evaluation results more objective and true. As an implementation manner, the acquisition period is the flowering stage or the pod-setting stage.

[0024] In the present invention, pea images of a target area are collected by aerial photography using a drone, and the structural information, texture information, and spectral information in the pea images are extracted. As an implementation manner, the flight altitude of the drone is set to 10 - 25 m, the flight speed is 1 - 3 m / s, and the aerial photography is taking pictures at equal time intervals. The growth heights of different pea varieties are also different. At this flight altitude, the drone can take close-up pictures above the pea plants, which is conducive to capturing details. The speed and taking pictures at equal time intervals can completely capture the structural information, texture information, and spectral information of peas in the entire target area. As an implementation manner, the longitudinal overlap rate of the drone aerial photography images is not less than 60%, and the lateral overlap rate is not less than 50%, so as to prevent omission or loss of information on pea varieties in the target area.

[0025] In the present invention, a drone is used to carry a sensor to collect pea images. As an implementation manner, the drone carries an RGB sensor and / or a multispectral sensor to collect the pea images of the target area, and obtain RGB images and / or MS images. Before collecting the multispectral images, a standard reflectance panel image is obtained first for subsequent radiometric calibration.

[0026] In the present invention, the drone image processing is to splice the obtained RGB and MS images. When splicing the MS images, the DN values are converted into reflectance values. As an implementation manner, the splicing process includes geometric correction, image alignment, constructing a dense point cloud, generating a digital surface model, generating a digital terrain model, and obtaining an orthophoto. In one implementation manner, the canopy coverage assessment is represented by the ratio of the vegetation area to the ROI area in the RGB image after removing the background, and the plant height assessment is represented by the maximum value of the crop surface model.

[0027] In the present invention, the plant height and canopy coverage are estimated through the structural information; the vegetation index information related to biomass and SPAD is obtained through the spectral information, and then the biomass and SPAD of peas are estimated by using a machine learning algorithm through the texture information and / or the vegetation index information.

[0028] Among them, the texture information can reflect the canopy structure and structural characteristics. As an implementation manner, the texture information includes entropy, contrast, inverse variance, variance, correlation, dominance, gray mean, and energy. Among them, the vegetation index is composed of the spectral information of different bands and can effectively reflect the growth status of crops. As an implementation manner, the vegetation index information related to biomass in the RGB sensor includes WI, GLI, VARI, ExR, ExB, IPCA, CIVE, and the vegetation index information related to SPAD includes INT, WI, ExG, ExB, IPCA, MRBVI, VARI. As an implementation manner, the vegetation index information related to biomass in the MS sensor includes CVI, CI, GDVI, EVI2, EVI, GARI, GLI, BNDVI, GRVI, MCARI1, MCARI2, SAVI, OSAVI, MCARI1 / OSAVI, and the vegetation index information related to SPAD includes CVI, CCCI, GNDVI, NDRE, PNDVI, GSAVI, GOSAVI, NREI, NNRI, MNDI, MEVI, MCARI1, MTCI, GDVI.

[0029] In the present invention, the machine learning algorithm is one or several of Catboost, Light GBM, GPR, or SVM. Using R 2 and RMSE to evaluate the robustness of the biomass and SPAD of the peas. As an implementation manner, RGB+MS data and the Catboost algorithm are used to estimate the biomass of peas, and RGB+MS data and the Light GBM algorithm are used to estimate the SPAD of peas.

[0030] In the present invention, the salt tolerance coefficient (STC) of the plant height, canopy coverage, biomass, and SPAD is calculated. In one implementation manner, the salt tolerance coefficient refers to the ratio of the trait value after salt treatment to the control trait value. The salt tolerance coefficient is normalized, and principal component analysis is performed on the normalized salt tolerance coefficient to obtain two principal components. Membership function analysis is performed on the two principal components to obtain membership function values, and a comprehensive salt tolerance scoring model is constructed by combining the weights of the two principal components and the membership function values. The salt tolerance factor (PSTS value) is obtained by using the comprehensive salt tolerance scoring model constructed by the present invention, and the salt tolerance of peas can be predicted. Among them, the size of the salt tolerance factor is positively correlated with the salt tolerance.

[0031] In the present invention, the membership function calculation formula is U(X ij )=(X i,j -X i,min ) / (X i,max -X j,min ), i = 1, 2,..., n

[0032] Among them, U represents the membership function value, and X i,j represents the value of the i-th principal component in the j-th plot, and X i,min represents the minimum value of the i-th principal component, and X i,max represents the maximum value of the i-th principal component.

[0033] In the present invention, the calculation formula for the weight of the principal component is: w i = pi / ∑(i = 1, n) pi, i = 1, 2, …, n

[0034] Among them, w i represents the weight of the i-th principal component, and Pi represents the contribution rate of the i-th principal component.

[0035] In the present invention, the calculation formula for the salt tolerance factor of the salt tolerance comprehensive scoring model is:

[0036] PSTS = ∑(i = 1, n) [U(X ij ) × w i

[0037] Among them, PSTS represents the salt tolerance factor, U(X ij ) represents the membership function value, X ij represents the value of the i-th principal component in the j-th plot, and w i represents the weight of the i-th principal component.

[0038] The present invention also provides a method for screening salt-tolerant pea varieties, which uses the above-mentioned UAV-based high-throughput salt tolerance comprehensive scoring model for peas to evaluate the salt tolerance of peas, and selects pea varieties with high comprehensive salt tolerance evaluation.

[0039] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be described in detail below with reference to embodiments, but they should not be construed as limiting the protection scope of the present invention.

[0040] For the materials, reagents, etc. used in the following embodiments, unless otherwise specified, the reagents, consumables, etc. involved in the present invention can be obtained from commercial channels. If the specific usage conditions are not indicated, they are usually carried out under conventional conditions or according to the conditions recommended by the company.

[0041] Example 1

[0042] 1. Research field setting

[0043] ​The research sites were divided into saline-alkali plots and ordinary plots (or control plots). The ordinary plots were located in Lubei District, Tangshan City, Hebei Province (latitude 39.36N, longitude 118.11E), and the saline-alkali plots were in Laoting County, Tangshan City, Hebei Province (latitude 39.43N, longitude 118.9E). Both places belong to the temperate monsoon climate. Except for the different soils, the climate, field plot settings, varieties, cultivation management measures, etc. were all the same.

[0044] Fourteen pea varieties from all over the country were sown in both plots, and a randomized block design was adopted, with three replicates for each variety. Before sowing, the soil was loosened and fertilized. Each plot was set to be 5m long, 2m wide, with a total of 5 rows, and the row spacing was 40cm. Sowing was carried out on March 13, 2023, with 120 seeds sown in each row, and the sowing depth was about 5cm. Insecticides were sprayed every half month, sprinkler irrigation was carried out every 7 - 10 days, and weeds were manually removed regularly. The peas were harvested on June 12, 2023.

[0045] 2. UAV data collection

[0046] The flowering stage to the filling stage is an important period for pea yield formation. Therefore, on April 28, 2023 (flowering stage) and May 10, 2023 (pod-setting stage), flights were carried out using DJI Phantom 4 Pro (DJI Technology, Shenzhen, China) and DJI Phantom 4 Multispectral Edition (DJI Technology, Shenzhen, China) UAVs. To ensure data quality, the UAVs collected visible light images (RGB) and multispectral images (MS) at 10:00 - 14:00 at noon. Before obtaining the multispectral images, standard reflectance panel images were obtained for subsequent radiometric calibration. Among them, the UAV flight altitude was set to 15m, the flight speed was 2m / s, and it was set to take pictures at equal time intervals. The longitudinal overlap rate and lateral overlap rate of the two plots were set to 85% and 80% respectively.

[0047] 3. UAV image processing

[0048] The acquired RGB and MS images are mosaicked using Pix4d software. By performing geometric correction, image alignment, and constructing a dense point cloud, a digital surface model is generated, a digital terrain model is generated, and an orthophoto is obtained. When mosaicking the MS images, radiometric calibration is performed using images with known reflectance, and the DN values are converted to reflectance values. Since the resolution of the RGB images is higher than that of the MS images, the structural information and texture information of canopy coverage and plant height are extracted using the RGB images, and the spectral information is extracted using the MS images. After mosaicking the RGB images, they are input into Arc Map pro for cropping, and the soil background is removed using the threshold segmentation method. At the same time, each plot is marked as a specific area and assigned a unique identifier. The corresponding relationship between the pea varieties and identifiers in the two plot areas is shown in Table 1.

[0049] Table 1 Corresponding relationship table of pea varieties and identifiers

[0050]

[0051]

[0052] 3.1 Estimation of pea canopy coverage and plant height

[0053] The canopy coverage (Canopycoverage) is estimated using the ratio of the vegetation area to the ROI area in the RGB image after removing the background, as shown in Equation 1; the plant height is estimated using the maximum value of the crop surface model, as shown in Equation 2:

[0054] Canopy coverage=P canopy / P total (1)

[0055] Where, P canopy represents the number of pea canopy pixels, and P total represents the total number of pixels in the sampling plot.

[0056] CSM=DSM - DTM (2)

[0057] Where, CSM represents the crop surface model, DSM represents the digital surface model, and DTM represents the digital terrain model.

[0058] 3.2 Estimation of pea biomass and SPAD

[0059] Texture information can reflect the canopy structure and structural characteristics. The image after removing the background is input into ENVI 5.3, and the sliding window and sliding wavelength are set to 7×7 respectively. A total of eight texture information, namely entropy, contrast, inverse variance, variance, correlation, dominance, gray mean, and energy, are extracted. Vegetation indices are composed of the spectral information of different bands and can effectively reflect the growth status of crops. This invention mainly uses four machine learning methods, namely Catboost, Light GBM, GPR, and SVM, to measure the texture information and vegetation index information for the estimation of pea biomass and SPAD value. To fairly and completely compare the estimation of pea aboveground biomass (AGB) and yield by different algorithms, 80% of the samples are randomly selected as the training data set for all four algorithms, and the remaining 20% is used as the test data set. Five-fold cross-validation is used to test the accuracy of the estimation, so as to ensure that all samples are independently used for verification.

[0060] In this invention, R 2 , RMSE are used to evaluate the robustness of pea biomass and SPAD. The higher the R 2 value and the lower the RMSE, the better the performance of the estimation. The calculation methods of R 2 , RMSE are shown in Formulas 3 and 4 as follows:

[0061]

[0062] Among them, n represents the total number of samples; X i and respectively represent the measured value and the estimated AGB or SPAD of each sample; represents the average value of the measured AGB or SPAD.

[0063] This invention extracts eight texture information (Contrast, Correlation, Dissimilarity, Entropy, Homogeneity, Mean, Variance, Energy) and seven vegetation index information with high correlation with biomass and SPAD from the RGB images collected by UAV data (among them, those related to biomass are WI, GLI, VARI, ExR, ExB, IPCA, CIVE; those related to SPAD are INT, WI, ExG, ExB, IPCA, MRBVI, VARI), and respectively estimates pea biomass and SPAD based on four machine learning methods (Catboost, Light GBM, GPR, SVM). The RGB estimation results are shown in Tables 2 and 3 as follows:

[0064] Table 2 Results of estimating pea biomass by RGB sensor

[0065] Catboost LightGBM GPR SVM Ordinary plot <![CDATA[R 2 > 0.68 0.50 0.58 0.50 RMSE 2.35 2.77 2.68 2.95 Saline-alkali plot <![CDATA[R 2 > 0.54 0.48 0.59 0.48 RMSE 1.01 0.88 0.92 1.07

[0066] Table 3 Results of estimating pea SPAD by RGB sensor

[0067] Catboost LightGBM GPR SVM Ordinary plot <![CDATA[R 2 > 0.31 0.20 0.29 0.22 RMSE 3.53 3.32 3.49 3.81 Saline-alkali plot <![CDATA[R 2 > 0.19 0.38 0.26 0.25 RMSE 2.97 2.41 2.78 2.87

[0068] The present invention extracts 14 vegetation index information with relatively high correlations with biomass and SPAD from MS images based on UAV data collection (among them, those related to biomass are CVI, CI, GDVI, EVI2, EVI, GARI, GLI, BNDVI, GRVI, MCARI1, MCARI2, SAVI, OSAVI, MCARI1 / OSAVI; those related to SPAD are CVI, CCCI, GNDVI, NDRE, PNDVI, GSAVI, GOSAVI, NREI, NNRI, MNDI, MEVI, MCARI1, MTCI, GDVI) and estimates pea biomass and SPAD based on four machine learning methods (Catboost, LightGBM, GPR, SVM) respectively. The MS estimation results are shown in Table 4 and Table 5 below:

[0069] Table 4 Results of estimating pea biomass by MS sensor

[0070] Catboost LightGBM GPR SVM Ordinary plot <![CDATA[R 2 > 0.58 0.56 0.64 0.51 RMSE 2.64 2.62 2.48 2.92 Saline-alkali plot <![CDATA[R 2 > 0.72 0.62 0.63 0.59 RMSE 0.78 0.76 0.87 0.88

[0071] Table 5 Results of estimating pea SPAD by MS sensor

[0072] Catboost LightGBM GPR SVM Ordinary plot <![CDATA[R 2 > 0.50 0.49 0.51 0.40 RMSE 2.85 3.03 2.86 3.16 Saline-alkali plot <![CDATA[R 2 > 0.40 0.55 0.42 0.26 RMSE 2.47 2.11 2.37 2.68

[0073] Using the fusion data of RGB and MS sensors, the biomass and SPAD values of peas are estimated based on four machine learning methods (Catboost, LightGBM, GPR, SVM) respectively, and the estimation results are shown in Table 6 and Table 7 below:

[0074] Table 6 Results of estimating pea biomass by RGB+MS sensor

[0075] Catboost LightGBM GPR SVM Ordinary plot <![CDATA[R 2 > 0.67 0.57 0.68 0.50 RMSE 2.41 2.62 2.33 2.95 Saline-alkali plot <![CDATA[R 2 > 0.73 0.63 0.65 0.57 RMSE 0.78 0.83 0.84 0.97

[0076] Table 7 Results of estimating pea SPAD by RGB+MS sensor

[0077] Catboost LightGBM GPR SVM Ordinary plot <![CDATA[R 2 > 0.59 0.64 0.53 0.48 RMSE 2.79 2.59 2.73 2.97 Saline-alkali plot <![CDATA[R 2 > 0.43 0.57 0.41 0.35 RMSE 2.44 2.07 2.39 2.62

[0078] From the above research, it can be seen that the estimation results of the two calculation methods of Catboost and Light GBM are better. Among them, the estimation results of RGB+MS data and Catboost algorithm for evaluating pea biomass in saline-alkali plots and control plots and RGB+MS data and LightGBM algorithm for evaluating pea SPAD in saline-alkali plots and control plots are the best, and their scatter plots are asFigure 1 as shown

[0079] 3.3 Establishment of PSTS

[0080] To evaluate the salt tolerance of peas, a comprehensive scoring model for pea salt tolerance was constructed based on four indicators: plant height, biomass, canopy coverage, and SPAD. The salt tolerance factor (PSTS value) was calculated. The larger the PSTS value, the higher the salt tolerance.

[0081] First, calculate the salt tolerance coefficient (STC) of each trait. The salt tolerance coefficient is the ratio of the trait value after salt treatment to the control trait value, and the calculation formula is shown in Formula 5. Then, normalize the salt tolerance coefficient, perform principal component analysis on the normalized salt tolerance coefficient, and obtain two principal components FAC1 and FAC2 with eigenvalues greater than 1. The contribution rates of the two principal components in each plot are shown in Table 8. According to the contribution rates of the principal components, calculate their weights using Formula 6 as shown in Table 9. Then, perform membership function analysis (see Formula 7) on the two principal components to obtain their corresponding membership function values as shown in Table 10. Finally, construct a comprehensive scoring model for salt tolerance as shown in Formula 8 based on the weights and membership function values, and calculate the salt tolerance factor (PSTS value) of peas as shown in Table 11.

[0082] STC = trait value after salt treatment / normal trait value (5)

[0083] w i = pi / ∑(i = 1, n)pi, i = 1, 2, …, n (6)

[0084] Among them, w i represents the weight of the i-th principal component, and Pi represents the contribution rate of the i-th principal component.

[0085] U(X ij ) = (X i,j - X i,min ) / (X i,max - X j,min ), i = 1, 2, …, n (7)

[0086] U represents the membership function value, X i,j represents the value of the i-th principal component in the j-th plot, X i,min represents the minimum value of the i-th principal component, and X i,max represents the maximum value of the i-th principal component.

[0087] PSTS = ∑(i = 1, n)[U(X ij ) × w i (8)

[0088] Among them, U represents the membership function value, and X ij represents the value of the i-th principal component in the j-th plot.

[0089] Table 8 Contribution rates of two principal components in each plot

[0090]

[0091]

[0092] Table 9 Weights of two principal components

[0093] Principal component <![CDATA[w i > FAC1 0.59 FAC2 0.41

[0094] Table 10 Membership function values of two principal components in each plot

[0095]

[0096] Comparative Example 1 Ground measurement data collection

[0097] On the same day as the UAV collection, the plant height, SPAD, and biomass data of the materials in the two plots were measured. For each plot, 5 plants were randomly selected, and the average value of the distance from the ground to the top of the plant was measured as the plant height data. 10 plants were randomly selected, and the chlorophyll content of the middle and upper leaves was measured using a SPAD-502plus chlorophyll meter. Finally, 3 plants were randomly selected, the roots were removed, and the above-ground biomass was measured by weighing.

[0098] Test Example 1

[0099] The comprehensive salt tolerance scoring model of peas based on UAV measurement data in the present invention was used to evaluate the salt tolerance factors (PSTS values) of different pea varieties in each plot of saline-alkali land. In Comparative Example 1, the salt tolerance factors of different pea varieties in each plot of saline-alkali land were evaluated based on ground measurement data using the PSTS construction method of the present invention. The calculation results of the salt tolerance factors of the present invention and Comparative Example 1 are shown in Table 11:

[0100] Table 11 Salt tolerance factors evaluated by ground measurement in Comparative Example 1 and UAV in the present invention

[0101]

[0102] The intuitive graph obtained from the calculation results in Table 11 is as Figure 2 shown. From Figure 2It can be seen that the five varieties with the best salt tolerance marked by the UAV evaluation of the present invention are 20231-2, 0804-5, Zhongwan 03 (repeated twice), and Tangwan No. 3; the five varieties with the best salt tolerance marked by the ground measurement data of Comparative Example 1 are 0804-5, 20231-2, Longwan 15, Zhongwan 03, and Tangwan No. 3. It is observed that four of the varieties with relatively high salt tolerance selected based on the UAV estimation of the present invention and the ground measurement data of Comparative Example 1 are the same, with high accuracy. This also proves that the model obtained through the sensor data fusion and appropriate algorithms of the present invention can accurately achieve high-throughput evaluation of the salt tolerance of pea varieties, improving the efficiency of screening salt-tolerant pea germplasms.

[0103] The above are only the embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification of the present invention, or directly or indirectly applied in other related technical fields, shall be similarly included in the patent protection scope of the present invention.

Claims

1. A method for constructing a high-throughput salt tolerance comprehensive scoring model for peas based on drones, characterized in that: The following steps are involved: Collect pea images of the target area, and extract structural information, texture information and spectral information from the pea images; Plant height and canopy coverage were estimated through structural information; vegetation index information related to biomass and SPAD was obtained through spectral information, and then machine learning algorithms were used to estimate pea biomass and SPAD through texture information and / or vegetation index information; The salt tolerance coefficients of the plant height, canopy coverage, biomass and SPAD were calculated, the salt tolerance coefficients were normalized, and principal component analysis was performed to obtain the principal components and determine the principal component weights. A comprehensive salt tolerance scoring model was constructed based on the principal component weights and the membership function values ​​of the principal components. The size of the salt tolerance factor was positively correlated with salt tolerance.

2. The method for constructing a comprehensive scoring model for salt tolerance according to claim 1, characterized in that: The pea image acquisition period is from the pea flowering period to the pea filling period.

3. The method for constructing a comprehensive scoring model for salt tolerance according to claim 1, characterized in that: The unmanned aerial vehicle is equipped with an RGB sensor and / or a multispectral sensor to obtain a pea image of the target area, and the pea image includes an RGB image and / or an MS image.

4. The method for constructing a comprehensive scoring model for salt tolerance according to claim 1, wherein: The texture information includes entropy, contrast, inverse variance, variance, correlation, dominance, grayscale mean and energy.

5. The method for constructing a comprehensive scoring model for salt tolerance according to claim 3, wherein: The vegetation index information related to biomass in the RGB sensor includes the Warbeck index, the green leaf area index, the visible atmospheric impedance vegetation index, the super red index, the super blue index, the principal component analysis index, and the color vegetation index. The vegetation index information related to SPAD includes the color intensity index, the Warbeck index, the super green index, the super blue index, the principal component analysis index, the improved red and blue vegetation index, and the visible atmospheric impedance vegetation index.

6. The method for constructing a comprehensive scoring model for salt tolerance according to claim 3, characterized in that: The vegetation index information related to biomass in the MS sensor includes chlorophyll vegetation index, chlorophyll index, generalized difference vegetation index, dual-band enhanced vegetation index, enhanced vegetation index, green anti-atmospheric vegetation index, green leaf area index, blue normalized difference vegetation index, green ratio vegetation index, improved chlorophyll absorption reflectance index, improved chlorophyll absorption reflectance index 2, soil adjusted vegetation index, optimized soil adjusted vegetation index, improved chlorophyll absorption reflectance index 1 / optimized soil adjusted vegetation index, and the vegetation index information related to SPAD includes chlorophyll vegetation index, canopy chlorophyll content index, green normalized difference vegetation index, normalized difference red edge index, projected normalized difference vegetation index, green soil adjusted vegetation index, green optimized soil adjusted vegetation index, normalized red edge index, normalized near-infrared redness index, improved normalized difference infrared index, improved enhanced vegetation index, improved chlorophyll absorption reflectance index, terrestrial chlorophyll index, and generalized difference vegetation index.

7. The method for constructing a comprehensive scoring model for salt tolerance according to claim 1, characterized in that: The machine learning algorithm uses the Catboost algorithm to estimate the pea biomass, and uses the Light GBM algorithm to estimate the pea SPAD.

8. The method for constructing a comprehensive scoring model for salt tolerance according to claim 1, characterized in that: Using R 2 The robustness of the pea biomass and SPAD was evaluated by RMSE.

9. The method for constructing a comprehensive scoring model for salt tolerance according to claim 1, characterized in that: The salt tolerance comprehensive scoring model is: PSTS = ∑ (i = 1, n) [U (X ij )×w i ], where PSTS represents the salt tolerance factor, U(X ij ) represents the membership function value, X ij represents the value of the i-th principal component of the j-th cell, w i represents the weight of the i-th principal component.

10. A method for screening salt-tolerant pea varieties, characterized in that: The salt tolerance of peas is evaluated using the drone-based high-throughput salt tolerance comprehensive scoring model for peas as described in any one of claims 1 to 9, and pea varieties with high comprehensive salt tolerance evaluation are selected.

Images