Method for improving plant nitrogen accumulation monitoring precision

Abstract

The invention provides a method for improving the monitoring precision of plant nitrogen accumulation, and belongs to the technical field of agricultural remote sensing information detection and metrology, and the method comprises the following steps: firstly, carrying out data conversion, smooth noise reduction and other preprocessing on obtained near-surface hyperspectral data, and obtaining the near-surface hyperspectral reflectivity of a research area; then, on the basis of the near-surface hyperspectral reflectivity, continuous wavelet transform is carried out, and a wavelet feature database is constructed; applying difference operation on the basis of the wavelet feature database to construct a difference wavelet feature index; and finally, constructing a plant nitrogen accumulation amount estimation model by applying the difference wavelet characteristic index, thereby achieving the effects of eliminating the planting condition difference and improving the nitrogen monitoring precision. The method for eliminating the planting conditions and improving the inversion precision of the nitrogen accumulation amount is simple and efficient in operation step, can be used for eliminating the influence of different planting conditions, and is widely applied to inversion of the nitrogen accumulation amount of near-ground or satellite images and related agronomic parameters and the like.

Description

Technical Field

[0001] The present invention relates to the field of agricultural remote sensing information detection and measurement technology, mainly in the aspects of remote sensing planting condition elimination and nitrogen accumulation monitoring, in particular to a method for eliminating the influence of winter wheat planting conditions and monitoring nitrogen accumulation based on near-ground hyperspectral wavelet characteristics. Background Art

[0002] Winter wheat, one of my country's most important grain crops, is a key factor influencing my country's external grain trade and food security. Currently, many studies use near-ground hyperspectral data to monitor winter wheat growth parameters, such as leaf nitrogen concentration, leaf nitrogen accumulation, leaf chlorophyll content, and plant nitrogen accumulation. Winter wheat plant nitrogen accumulation is susceptible to planting conditions such as sowing time, variety, regional precipitation, and temperature, which reduces the transferability and universality of existing monitoring models. However, most studies only use plot experiments to accurately infer nitrogen accumulation, ignoring the influence of planting conditions, resulting in low model validation accuracy. Therefore, developing a method to eliminate the influence of planting conditions and improve the ability to estimate winter wheat plant nitrogen accumulation is crucial for advancing precision agriculture and digital agriculture in my country.

[0003] Currently, the most commonly used methods for estimating winter wheat plant nitrogen accumulation based on near-ground hyperspectral data are the vegetation index method and the wavelet coefficient method. The vegetation index method constructs a vegetation index using the spectral reflectance of the crop canopy near the ground. The correlation between the vegetation index and plant nitrogen accumulation is then used to invert winter wheat plant nitrogen accumulation. The wavelet coefficient method uses a continuous wavelet transform to enhance the signal at different scales, reduce noise, and isolate wavelet coefficient features in different bands. These wavelet coefficient features are then used to invert plant nitrogen accumulation. Although both the vegetation index and wavelet coefficient methods are based on crop canopy spectral reflectance, their construction principles differ. The former tends to use geometric algorithms to construct a vegetation index to eliminate the influence of spectral noise and planting conditions, while the latter uses a continuous wavelet transform to enhance the reflectance signal related to plant nitrogen accumulation and eliminate the influence of noise. Previous studies have shown that the wavelet coefficient method is superior to the vegetation index method in estimating plant nitrogen accumulation. However, the former requires more spectral information, resulting in spectral signal redundancy, which to some extent affects estimation accuracy. In view of this, it is still unclear how to utilize the advantages of vegetation index and wavelet coefficient method to construct a simple and practical wavelet coefficient index to eliminate the influence of planting conditions and improve the estimation accuracy of nitrogen accumulation in winter wheat plants.

[0004] To address the issue of signal redundancy in near-surface hyperspectral wavelet coefficient methods, some researchers have extracted effective wavelet coefficients using principal component analysis (PCA) and then constructed effective wavelet coefficients to estimate crop nitrogen accumulation. However, while this method offers high accuracy, it carries high computational cost and time investment, and can result in loss of wavelet coefficient signal. Vegetation index methods, the simplest approach, effectively reduce noise and error through geometric operations. Difference vegetation index methods primarily enhance essential spectral signals to reduce spectral noise and error. Consequently, some studies have applied texture information to construct vegetation indices for estimating crop nitrogen accumulation. However, this approach relies heavily on imaging drone or satellite imagery, limiting its use with non-imaging near-surface hyperspectral data. Therefore, constructing a new difference vegetation index using wavelet coefficient information is crucial for achieving high-precision inversion of crop nitrogen accumulation. More importantly, there is currently no wavelet-based vegetation index calculation method that can eliminate the influence of winter wheat planting conditions and improve nitrogen accumulation monitoring accuracy. Summary of the Invention

[0005] The technical problem solved by the present invention is to provide a method for improving the monitoring accuracy of plant nitrogen accumulation. Wavelet features are obtained and screened through continuous wavelet transform, and the difference algorithm is applied to calculate the differential wavelet characteristic vegetation index. This method is simple to operate, has a fast calculation speed, and effectively eliminates the influence of planting conditions. It can be applied to the inversion of plant nitrogen accumulation for different winter wheat varieties, nitrogen gradients, and growth periods.

[0006] The technical solutions for achieving the purpose of the present invention are:

[0007] A method for improving the accuracy of monitoring plant nitrogen accumulation includes the following steps:

[0008] S1. According to the key growth period of crops, near-surface hyperspectral data of multiple growth periods are obtained and pre-processed to obtain near-surface hyperspectral reflectance data of the study area;

[0009] S2, calculate the vegetation index VIopt based on the near-surface hyperspectral reflectance data, and apply continuous wavelet transform to construct a database of wavelet coefficients at different scales;

[0010] S3, taking the wavelet coefficient database as input variables for difference, and generating a differenced wavelet characteristic index database;

[0011] S4, screening based on the differential wavelet characteristic index database in S3 to obtain the optimal differential wavelet characteristic index;

[0012] S5. Extract the optimal differenced wavelet characteristic index of each plot as the full sample as the training sample, and use samples according to different planting conditions (variety, nitrogen gradient, growth period, year) as verification samples. The training samples are used to construct a linear regression model of winter wheat nitrogen accumulation, and the verification samples are used to verify the model accuracy.

[0013] Preferably, the key growth periods when acquiring data in S1 mainly include the jointing stage, the booting stage and the flowering stage.

[0014] Preferably, the step of preprocessing the data obtained in S1 includes:

[0015] S11, performing data conversion on the acquired near-ground hyperspectral data to obtain three repeated hyperspectral reflectance data for each cell;

[0016] S12. Use Savitzky-Golay smoothing method to smooth and reduce noise of high spectral reflectance data in S11.

[0017] Preferably, the steps of calculating the optimal vegetation index VIopt, performing continuous wavelet transform, and constructing a database of wavelet coefficients of different scales in S2 include:

[0018] S21. Calculate the optimal vegetation index VIopt based on the smoothed and denoised hyperspectral reflectance obtained in S12. The specific formula is:

[0019] VIopt=(1+0.45)×((R800)^2+1) / (R670+0.45)

[0020] Among them, VIopt is the optimal vegetation index, R800 and R670 are the spectral reflectance values at 800nm and 670nm respectively;

[0021] S22, based on the smoothed and denoised hyperspectral reflectance obtained in S12, continuous wavelet transform is applied to calculate wavelet coefficients of different scales. The specific formula is:

[0022]

[0023] W f (a,b)= <fW f (a,b)= <f

[0024]

[0025] Among them, ψ a,b (λ) is the mother wavelet function after translation and scaling, a is the scale factor, W f (a, b) are wavelet coefficients, also called wavelet features, and f(λ) is the hyperspectral reflectance data;

[0026] S23. Construct a wavelet coefficient database of different scales based on the wavelet coefficients obtained in S22.

[0027] Preferably, the step of constructing the differential wavelet characteristic index in S3 includes:

[0028] S31, using the wavelet coefficients of different scales obtained in S2 as input data;

[0029] S32. Calculate the differential wavelet characteristic index using the difference method. The specific formula is as follows:

[0030] DWF(a,λ1,λ2)=WF(a,λ1)-WF(a,λ2)

[0031] Among them, WF(a,λ1) and WF(a,λ2) are the wavelet coefficients corresponding to the λ1 and λ2 bands at the a-th scale, respectively, and DWF(a,λ1,λ2) is the difference wavelet characteristic index at the a-th scale.

[0032] Preferably, the step of screening the optimal difference wavelet characteristic index in S4 includes:

[0033] S41, using the different-scale difference wavelet characteristic indices obtained in S3 and the nitrogen accumulation of winter wheat plants in each experimental plot as input data;

[0034] S42. Optimal differential wavelet characteristic index was screened out by calculating the determination coefficient between differential wavelet characteristic indexes at different scales and nitrogen accumulation of wheat plants in each experimental plot.

[0035] Preferably, the steps of constructing a linear model and model verification in S5 are:

[0036] S51, according to the optimal difference wavelet characteristic index data obtained in S4, extract its average value in each cell as a sample;

[0037] S52, selecting all the data in S51 as training samples and combining them with the nitrogen accumulation data of the plot plants to construct a linear regression model;

[0038] S53. Based on the model constructed in S52, the data under different planting conditions (variety, nitrogen gradient, growth period, year) in S51 were selected as validation samples, and the coefficient of determination R was used. 2 , RMSE evaluation and verification results, the specific formula is as follows:

[0039]

[0040] Among them, y i and y′ i are the test value and model prediction value of plant nitrogen accumulation, is the average nitrogen accumulation of the test plants, and n is the number of test samples.

[0041] Compared with the prior art, the technical solution adopted by the present invention has the following technical effects:

[0042] 1. The method of improving the monitoring accuracy of plant nitrogen accumulation of the present invention is a method for eliminating the influence of winter wheat planting conditions and monitoring nitrogen accumulation based on near-ground hyperspectral wavelet characteristics. It eliminates the influence of planting conditions and improves the monitoring accuracy of winter wheat nitrogen accumulation based on near-ground hyperspectral wavelet characteristics. While eliminating the influence of planting conditions, the estimation accuracy of winter wheat nitrogen accumulation is guaranteed at the same time.

[0043] 2. The method of the present invention for eliminating the influence of planting conditions and improving the monitoring accuracy of winter wheat nitrogen accumulation based on near-ground hyperspectral wavelet features is insensitive to the influence of spectral resolution and has good universality.

[0044] 3. The method of the present invention, which is based on near-ground hyperspectral wavelet features to eliminate the influence of planting conditions and improve the monitoring accuracy of winter wheat nitrogen accumulation, is applied to the detection, measurement and related standardization research of agricultural growth information. It is simple and effective and can be integrated into intelligent agricultural machinery and equipment such as drones and unmanned vehicles, facilitating large-scale promotion and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a schematic diagram of near-ground hyperspectral acquisition of winter wheat according to the present invention;

[0046] Figure 2 is a flow chart of the present invention;

[0047] Figure 3 The present invention is a regression model for predicting nitrogen accumulation in winter wheat using other methods and the present invention, wherein (a) is a nitrogen accumulation prediction model based on the vegetation index VIopt, (b) is a nitrogen accumulation prediction model based on the wavelet coefficient WF(4,770) at the fourth scale of 770 nm, and (c) is a leaf nitrogen concentration prediction model based on the differential wavelet characteristic index at the fourth scale of 560 nm and 770 nm.

[0048] Figure 4 Figure 3 is a scatter plot of the nitrogen accumulation prediction model verified by other nitrogen accumulation monitoring methods and this method under different planting conditions, among which (a) is a scatter plot of the model verified based on the vegetation index VIopt, (b) is a scatter plot of the model verified based on the wavelet coefficient WF(4,770) at the 4th scale of 770nm, and (c) is a scatter plot of the model verified based on the differenced wavelet characteristic index at the 4th scale of 560nm and 770nm wavelet coefficients. DETAILED DESCRIPTION

[0049] The embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limiting the present invention.

[0050] The present invention implements a field experiment of winter wheat based on different growth periods, varieties, nitrogen fertilizer application rates and years to eliminate the influence of planting conditions and improve the inversion of winter wheat nitrogen accumulation. The present invention uses a handheld hyperspectral instrument (ASD, HandHeld 2, USA) to obtain near-ground hyperspectral data of the study area as experimental data. The specific near-ground hyperspectral data acquisition status is as follows: Figure 1 The specific parameters of the handheld hyperspectrometer and the sampling period of the acquired data are shown in Table 1.

[0051] Table 1 Handheld hyperspectral instrument parameters and data sampling period

[0052] Sensor indicators and sampling period parameter sensor HandHeld2 Wavelength range 325-1075nm Wavelength accuracy 1nm Field of view 25° Spectral resolution <3.0nm@700nm Equivalent Noise Radiation (NEdL) 5x10-9W / cm2 / nm / sr@700nm Memory 500MB Dimensions (HxWxD) 90x140x215mm Sampling growth period Jointing stage, booting stage, flowering stage

[0053] like Figure 2 As shown, the method of the present invention specifically comprises the following steps:

[0054] S1. According to the key growth period of crops, near-ground hyperspectral data of multiple growth periods are obtained and preprocessed to obtain near-ground hyperspectral reflectance data of the study area.

[0055] The experimental plots were selected for winter wheat during the key growth stages of jointing, booting, and flowering. Spectra were collected using a handheld hyperspectral instrument at a distance of 50 cm directly above the wheat canopy at 12:00 PM Beijing time on a clear, cloudless day. Three sample points were evenly spaced in each plot, for a total of five spectra collected at each point. Radiation correction was performed using a standard reflectance whiteboard with 100% reflectance.

[0056] The collected data preprocessing process mainly includes:

[0057] S11. Apply ViewSpecPros to the acquired near-ground hyperspectral data TM The software performed data conversion separately to obtain the hyperspectral reflectance data of three replicates for each plot;

[0058] S12, using Savitzky-Golay smoothing method in ViewSpecPros TM The high spectral reflectance data of S11 were smoothed and denoised in the software.

[0059] S2. Calculate the vegetation index VIopt based on the near-surface hyperspectral reflectance data and apply continuous wavelet transform to construct a database of wavelet coefficients at different scales. Specifically, the following steps are included:

[0060] S21. In ENVI 5.3 software, the optimal vegetation index VIopt is calculated by using the hyperspectral reflectance obtained in S12 after smoothing and noise reduction. The specific formula is:

[0061] VIopt=(1+0.45)×((R800)^2+1) / (R670+0.45)

[0062] Among them, VIopt is the optimal vegetation index, R800 and R670 are the spectral reflectance values at 800nm and 670nm respectively;

[0063] S22. In the ENVI 5.3 IDL programming environment, the hyperspectral reflectance obtained by S12 after smoothing and denoising is applied to calculate the wavelet coefficients of different scales. The specific formula is:

[0064]

[0065] W f (a,b)= <fW f (a,b)= <f

[0066]

[0067] Among them, ψ a,b (λ) is the mother wavelet function after translation and scaling, a is the scale factor, W f (a, b) are wavelet coefficients, also called wavelet features, and f(λ) is the hyperspectral reflectance data;

[0068] S23. Apply the wavelet coefficients obtained in S21 in ENVI 5.3 software to construct a wavelet coefficient database of different scales (4th, 5th, 6th, 7th, etc.).

[0069] S3, taking the wavelet coefficient database as input variables for difference, and generating a difference wavelet characteristic index database. The steps of constructing the difference wavelet characteristic index include:

[0070] S31, using the wavelet coefficients of different scales obtained in S2 as input data of the spectrum and performing calculations in ENVI 5.3 software;

[0071] S32. Calculate the differential wavelet characteristic index using the difference method. The specific formula is as follows:

[0072] DWF(a,λ1,λ2)=WF(a,λ1)-WF(a,λ2)

[0073] Among them, WF(a,λ1) and WF(a,λ2) are the wavelet coefficients corresponding to the λ1 and λ2 bands at the a-th scale, respectively, and DWF(a,λ1,λ2) is the difference wavelet characteristic index at the a-th scale.

[0074] S4: Screening based on the differential wavelet characteristic index database in S3 to obtain the optimal differential wavelet characteristic index. The specific steps include:

[0075] S41, inputting the different scale difference wavelet characteristic indices obtained in S3 and the nitrogen accumulation of winter wheat plants in each experimental plot into matlab7.0 software as input data;

[0076] S42. The optimal differential wavelet characteristic index was screened out by calculating the determination coefficient between the differential wavelet characteristic index of different scales and the nitrogen accumulation of wheat plants in each experimental plot in MATLAB 7.0 software.

[0077] S5. Extract the optimal differenced wavelet characteristic index of each plot as the full sample as the training sample, and select statistical samples as validation samples based on different planting conditions such as variety, nitrogen application gradient, growth period, and year. The training samples are used to construct a linear regression model of winter wheat nitrogen accumulation; the validation samples are used to verify the model accuracy. Specifically, it includes:

[0078] S51. Based on the optimal differenced wavelet characteristic index data obtained in S4, the average value in each cell is extracted in MATLAB 7.0 software as the overall sample. In this embodiment, the number of samples = 155;

[0079] S52, selecting all the data in S51 as training samples and combining them with the nitrogen accumulation data of the plot plants to construct a linear regression model;

[0080] S53. Based on the model constructed in S52, the data under different planting conditions (variety, nitrogen gradient, growth period, year) in S51 were selected as validation samples, and the coefficient of determination R was used. 2 , RMSE evaluation and verification results, the specific formula is as follows:

[0081]

[0082] Among them, y i and y′ i are the test value and model prediction value of plant nitrogen accumulation, is the average value of nitrogen accumulation of the test plants, and n is the number of test samples, which varies under different planting conditions.

[0083] The following compares the different performances of the method of the present invention with the VIopt method and the wavelet coefficient WF(4,770) method at the 4th scale 770nm in eliminating the influence of planting conditions and improving the monitoring accuracy of winter wheat nitrogen accumulation in the study area.

[0084] Comparison of the accuracy of monitoring winter wheat nitrogen accumulation using the method of the present invention and the prior art Figure 3 As shown, the validation scatter plots of the models constructed by all methods are as follows Figure 4 The effects of the three methods of constructing models on spectral resolution are shown in Table 2.

[0085] Table 2 Ability of the present invention and prior art to predict nitrogen accumulation at different spectral resolutions

[0086] Spectral resolution (nm) Vegetation Index VIopt Wavelet coefficient WF(4,770) Methods of this study 1 0.73 0.77 0.82 2 0.72 0.73 0.81 4 0.72 0.72 0.81 8 0.64 0.67 0.80 16 0.51 0.65 0.79 32 0.46 0.61 0.74

[0087] from Figure 3 、 Figure 4 As shown in Table 2, the differential wavelet characteristic index generated by the method of the present invention is superior to other methods in terms of the ability to model and verify the nitrogen accumulation of winter wheat and the effect on spectral resolution. The method of the present invention can eliminate the influence of planting conditions and improve the monitoring accuracy of winter wheat nitrogen accumulation. It can be used for precise monitoring of nitrogen accumulation and other agronomic parameters in a large-scale and multi-range manner on various intelligent monitoring platforms. At the same time, it can also be applied to the detection and measurement of other agricultural growth information materials, as well as related standardization research.

[0088] It should be noted that, in this article, the terms "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or apparatus.

[0089] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for improving the accuracy of monitoring nitrogen accumulation in plants, characterized in that: The following steps are involved: S1. According to the key growth period of crops, near-surface hyperspectral data of multiple growth periods are obtained and pre-processed to obtain near-surface hyperspectral reflectance data of the study area; S2, applying continuous wavelet transform based on near-surface hyperspectral reflectance data to construct a database of wavelet coefficients at different scales; S3, taking the wavelet coefficient database as input variables for difference, and generating a differenced wavelet characteristic index database; S4, screening based on the differential wavelet characteristic index database in S3 to obtain the optimal differential wavelet characteristic index; S5. Extract the optimal differenced wavelet characteristic index of each plot as the full sample as the training sample, and use the samples under different planting conditions as the verification samples. The training samples are used to construct a linear regression model of winter wheat nitrogen accumulation, and the verification samples are used to verify the model accuracy.

2. A method for improving the monitoring accuracy of plant nitrogen accumulation according to claim 1, characterized in that: The key growth stages for obtaining data in S1 include the jointing stage, the booting stage, and the flowering stage.

3. The method for improving the monitoring accuracy of plant nitrogen accumulation according to claim 1, characterized in that: The planting conditions in S5 include variety, nitrogen application gradient, growth period and year.

4. A method for improving the monitoring accuracy of plant nitrogen accumulation according to claim 3, characterized in that: The steps of constructing the linear model and model verification in S5 are: S51, according to the optimal difference wavelet characteristic index data obtained in S4, extract its average value in each cell as a sample; S52, selecting all the data in S51 as training samples and combining them with the nitrogen accumulation data of the plot plants to construct a linear regression model; S53. Based on the model constructed in S52, the data under different planting conditions in S51 were selected as validation samples and the coefficient of determination R was applied. 2 , root mean square error RMSE evaluation verification results.

5. The method for improving the monitoring accuracy of plant nitrogen accumulation according to claim 1, characterized in that: The steps of preprocessing near-ground hyperspectral data in S1 include: S11, performing data conversion on the acquired near-ground hyperspectral data to obtain three repeated hyperspectral reflectance data for each cell; S12. Use Savitzky-Golay smoothing method to smooth and reduce noise of high spectral reflectance data in S11.

6. A method for improving the monitoring accuracy of plant nitrogen accumulation according to claim 5, characterized in that: Constructing a database of wavelet coefficients at different scales includes the following steps: S21, calculating the optimal vegetation index VIopt based on the smoothed and denoised hyperspectral reflectance obtained in S12; S22, calculating wavelet coefficients of different scales based on the smoothed and denoised hyperspectral reflectance obtained in S12 by applying continuous wavelet transform; S23. Construct a wavelet coefficient database of different scales based on the wavelet coefficients obtained in S22.

7. The method for improving the monitoring accuracy of plant nitrogen accumulation according to claim 1, characterized in that: In S2, the vegetation index VIopt is calculated based on the near-ground hyperspectral reflectance data, and a continuous wavelet transform is applied to construct a wavelet coefficient database of different scales.

8. The method for improving the monitoring accuracy of plant nitrogen accumulation according to claim 1, characterized in that: The step of constructing the differential wavelet characteristic index in S3 includes: S31, using the wavelet coefficients of different scales obtained in S2 as input data; S32. Apply the difference method to calculate the differenced wavelet characteristic index.

9. The method for improving the monitoring accuracy of plant nitrogen accumulation according to claim 1, characterized in that: The step of screening the optimal difference wavelet characteristic index in S4 includes: S41, using the wavelet coefficients of different scales obtained in S3 and the nitrogen accumulation of winter wheat plants in each experimental plot as input data; S42. By calculating the determination coefficient between wavelet coefficients of different scales and the nitrogen accumulation of wheat plants in each experimental plot, the optimal differential wavelet characteristic index is screened out.

Images