阅读说明：本技术 一种土壤总氮的红外光谱分析方法 (Infrared spectroscopic analysis method for total nitrogen in soil ) 是由 冷庚 刘哲 许文波 贾海涛 罗欣 常乐 于 2021-08-27 设计创作，主要内容包括：本发明属于土壤总氮含量分析技术领域,具体涉及一种土壤总氮的红外光谱分析方法。本发明的分析方法包括如下步骤：步骤1,采集土壤样品的近红外光谱,得到近红外光谱数据；步骤2,通过SG平滑算法对所述近红外光谱数据进行预处理；步骤3,通过预测模型对经过步骤2预处理后的近红外光谱数据进行预测,得到总氮含量结果。本发明在对土壤总氮的红外光谱分析方法中加入了预处理步骤,同时对预处理算法、建模算法、建模参数等进行了优选,能够提高模型预测土壤总氮含量的准确性,在农业、环保和生物研究等领域具有很好的应用前景。(The invention belongs to the technical field of soil total nitrogen content analysis, and particularly relates to an infrared spectrum analysis method of soil total nitrogen. The analysis method of the invention comprises the following steps: step 1, collecting a near infrared spectrum of a soil sample to obtain near infrared spectrum data; step 2, preprocessing the near infrared spectrum data through an SG smoothing algorithm; and 3, predicting the near infrared spectrum data preprocessed in the step 2 through a prediction model to obtain a total nitrogen content result. According to the invention, a pretreatment step is added in the infrared spectrum analysis method of the soil total nitrogen, and a pretreatment algorithm, a modeling algorithm, modeling parameters and the like are optimized, so that the accuracy of predicting the soil total nitrogen content by a model can be improved, and the method has a good application prospect in the fields of agriculture, environmental protection, biological research and the like.)

1. An infrared spectrum analysis method of soil total nitrogen is characterized by comprising the following steps:

step 1, collecting a near infrared spectrum of a soil sample to obtain near infrared spectrum data;

step 2, preprocessing the near infrared spectrum data through an SG smoothing algorithm;

and 3, predicting the near infrared spectrum data preprocessed in the step 2 through a prediction model to obtain a total nitrogen content result.

2. The analytical method of claim 1, wherein: the soil sample was collected from Chengdu plain.

3. The analytical method of claim 1, wherein step 3 comprises the steps of: predicting the preprocessed near infrared spectrum data by adopting a full-wavelength prediction model to obtain a total nitrogen content result; the full-wavelength prediction model is obtained through modeling by a PLSR algorithm or an ANN algorithm.

4. The analytical method of claim 3, wherein: the full-wavelength prediction model is obtained through modeling by a PLSR algorithm.

5. The analytical method of claim 4, wherein: the number of principal components of the PLSR algorithm is set to 8.

6. The analytical method of claim 1, wherein step 3 comprises the steps of:

step 3A, selecting characteristic wavelengths from the preprocessed near infrared spectrum data by adopting a CARS algorithm or a Random-from algorithm to obtain near infrared spectrum data in a characteristic wavelength range;

step 3B, predicting the near infrared spectrum data in the characteristic wavelength range by adopting a characteristic wavelength prediction model to obtain a total nitrogen content result; the characteristic wavelength prediction model is obtained through modeling by a PLSR algorithm.

7. The analytical method of claim 6, wherein: the characteristic wavelength is 1430 + -1 nm, 1431 + -1 nm, 1435 + -1 nm, 1442 + -1 nm, 1445 + -1 nm, 1456 + -1 nm, 1461 + -1 nm, 1465 + -1 nm, 1477 + -1 nm, 1482 + -1 nm, 2873 + -1 nm, 2885 + -1 nm, 2909 + -1 nm and 2920 + -1 nm;

or the characteristic wavelength is 1430 + -1 nm, 1431 + -1 nm, 1435 + -1 nm, 1442 + -1 nm, 1445 + -1 nm, 1456 + -1 nm, 1461 + -1 nm, 1465 + -1 nm, 1477 + -1 nm, 1482 + -1 nm, 2873 + -1 nm, 2885 + -1 nm, 2909 + -1 nm and 2920 + -1 nm.

8. A computer device for infrared spectroscopic analysis of total nitrogen in soil comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the program implements a method for infrared spectroscopic analysis of total nitrogen in soil as claimed in any one of claims 1 to 7.

9. An infrared spectroscopic analysis system for total nitrogen in soil, comprising:

infrared spectrum acquisition and/or input means for acquiring and/or inputting near infrared spectrum data;

the computer apparatus of claim 8, configured to analyze the near infrared spectral data to obtain a total nitrogen content result.

10. A computer-readable storage medium characterized by: stored thereon is a computer program for carrying out a method for infrared spectroscopic analysis of total nitrogen in soil according to any one of claims 1 to 7.

Technical Field

The invention belongs to the technical field of soil total nitrogen content analysis, and particularly relates to an infrared spectrum analysis method of soil total nitrogen.

Background

The content of nutrient elements in soil not only affects the growth of vegetation and crops, but also affects the regional ecological quality and the distribution of animal and plant populations. Therefore, the determination of nutrient elements in soil has become an urgent need in the fields of modern environmental science, agricultural science, ecology and the like.

Nitrogen is a basic element required for biochemical reactions in many organisms, is one of four basic elements constituting biomolecules such as DNA and RNA, and is also one of constituent elements of proteins. Nitrogen also plays a role in plant photosynthesis, which is used to produce chlorophyll molecules in related chemical reactions. Soil nitrogen (total nitrogen) can promote the growth and development of leaves, roots and stems of crops, so that the total nitrogen can influence the growth quality of the crops. Therefore, the determination of the total nitrogen content of the soil is very important for crop growth, related scientific research, environmental monitoring, and the like.

At present, the national standard method (HJ 717-.

Therefore, other techniques for analyzing total nitrogen in soil have been developed. The near infrared spectroscopy is a physical analysis method based on a spectroscopic technology, has the advantages of high analysis speed, low cost, no reagent consumption, capability of realizing simultaneous measurement of multiple components, small and portable instrument, suitability for field analysis and the like, and is widely concerned by people. Chinese patent application CN108982406A discloses a method for analyzing total nitrogen content of soil by using near infrared spectrum data, wherein a backward interval partial least square method BIPLS and a competitive adaptive weight sampling method CARS are fused to respectively select a near infrared spectrum characteristic interval and a characteristic variable of soil, and the result of the two algorithms is optimized and fused to determine the near infrared spectrum characteristic interval of the soil; and establishing a prediction model between the characteristic wave band spectrum and the soil nitrogen content by using the PLS algorithm again.

Generally, the chemical composition of soil is complex, and nitrogen element is an element with a low content relative to other elements contained in a large amount. Thus, when measuring near infrared spectral absorption of soil, it has been found that the absorption intensity associated with nitrogen-containing compounds is relatively small and is easily disturbed or superimposed by noise or other peaks. Therefore, if the original spectrogram of the near infrared spectrum is directly analyzed, an accurate soil total nitrogen analysis result is difficult to obtain.

Disclosure of Invention

Aiming at the defects of the prior art, the invention provides an infrared spectrum analysis method of soil total nitrogen, aiming at accurately analyzing the soil total nitrogen content.

An infrared spectrum analysis method of soil total nitrogen comprises the following steps:

step 1, collecting a near infrared spectrum of a soil sample to obtain near infrared spectrum data;

step 2, preprocessing the near infrared spectrum data through an SG smoothing algorithm;

and 3, predicting the near infrared spectrum data preprocessed in the step 2 through a prediction model to obtain a total nitrogen content result.

Preferably, the soil sample is collected from a Cheng Du plain.

Preferably, step 3 comprises the steps of: predicting the preprocessed near infrared spectrum data by adopting a full-wavelength prediction model to obtain a total nitrogen content result; the full-wavelength prediction model is obtained through modeling by a PLSR algorithm or an ANN algorithm.

Preferably, the full-wavelength prediction model is obtained by modeling through a PLSR algorithm.

Preferably, the number of principal components of the PLSR algorithm is set to 10.

Preferably, step 3 comprises the steps of:

step 3A, selecting characteristic wavelengths from the preprocessed near infrared spectrum data by adopting a CARS algorithm or a Random-from algorithm to obtain near infrared spectrum data in a characteristic wavelength range;

step 3B, predicting the near infrared spectrum data in the characteristic wavelength range by adopting a characteristic wavelength prediction model to obtain a total nitrogen content result; the characteristic wavelength prediction model is obtained through modeling by a PLSR algorithm.

Preferably, the characteristic wavelength is 1430 + -1 nm, 1431 + -1 nm, 1435 + -1 nm, 1442 + -1 nm, 1445 + -1 nm, 1456 + -1 nm, 1461 + -1 nm, 1465 + -1 nm, 1477 + -1 nm, 1482 + -1 nm, 2873 + -1 nm, 2885 + -1 nm, 2909 + -1 nm and 2920 + -1 nm;

or the characteristic wavelength is 1430 + -1 nm, 1431 + -1 nm, 1435 + -1 nm, 1442 + -1 nm, 1445 + -1 nm, 1456 + -1 nm, 1461 + -1 nm, 1465 + -1 nm, 1477 + -1 nm, 1482 + -1 nm, 2873 + -1 nm, 2885 + -1 nm, 2909 + -1 nm and 2920 + -1 nm.

The invention also provides computer equipment for infrared spectrum analysis of soil total nitrogen, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the infrared spectrum analysis method of soil total nitrogen when executing the program.

The invention also provides an infrared spectroscopic analysis system for soil total nitrogen, which comprises:

infrared spectrum acquisition and/or input means for acquiring and/or inputting near infrared spectrum data;

the computer equipment is used for analyzing the near infrared spectrum data to obtain a total nitrogen content result.

The present invention also provides a computer-readable storage medium characterized in that: on which a computer program is stored for implementing the above-mentioned method for infrared spectroscopic analysis of total nitrogen of soil.

The invention improves the infrared spectrum analysis method of the soil total nitrogen, adds the pretreatment step of infrared spectrum data before the prediction is carried out through the model, and simultaneously, the invention also optimizes a pretreatment algorithm, a model modeling method and the like. By comparing parameters such as correlation coefficient, decision coefficient, root mean square error and relative analysis error of the prediction result, the method provided by the invention has better accuracy when used for predicting the total nitrogen content of the soil.

Obviously, many modifications, substitutions, and variations are possible in light of the above teachings of the invention, without departing from the basic technical spirit of the invention, as defined by the following claims.

The present invention will be described in further detail with reference to the following examples. This should not be understood as limiting the scope of the above-described subject matter of the present invention to the following examples. All the technologies realized based on the above contents of the present invention belong to the scope of the present invention.

Drawings

FIG. 1 is raw data of near infrared spectrum of a soil sample;

FIG. 2 is near infrared spectral data after processing by SG smoothing algorithm;

FIG. 3 is near infrared spectral data after processing by moving average;

FIG. 4 is near infrared spectral data after processing by the first derivative method;

FIG. 5 is near infrared spectral data after processing by the second derivative method;

FIG. 6 is near infrared spectral data after treatment by standard normal variation;

FIG. 7 is a graph showing the results of the ANN network in Experimental example 2;

FIG. 8 shows the prediction result of the MLR modeling algorithm in Experimental example 2;

FIG. 9 shows the predicted result of PCR as the modeling algorithm in Experimental example 2;

FIG. 10 shows the predicted results of PLSR as a modeling algorithm in Experimental example 2;

FIG. 11 shows the result of the SVR model in Experimental example 2;

FIG. 12 shows the predicted result of the modeling algorithm ANN in Experimental example 2;

FIG. 13 shows the predicted result of CARS as the characteristic wavelength selection algorithm in Experimental example 3;

FIG. 14 shows the result of predicting the characteristic wavelength selection algorithm as Random-from in Experimental example 3.

Detailed Description

It should be noted that, in the embodiment, the algorithm of the steps of data acquisition, transmission, storage, processing, etc. which are not specifically described, as well as the hardware structure, circuit connection, etc. which are not specifically described, can be implemented by the contents disclosed in the prior art.

Example 1 analysis of total nitrogen in soil by full wavelength prediction model

An infrared spectrum analysis method of soil total nitrogen comprises the following steps:

step 1, collecting a near infrared spectrum of a soil sample to obtain near infrared spectrum data.

The specific method comprises the following steps: and drying the soil sample at 105 ℃ for 10 hours. And grinding potassium bromide into powder in a mortar, adding the dried soil sample into the mortar for grinding together, and then putting the ground soil sample into a tablet machine for tabletting. After the tablet pressing is finished, the finished product is put in a Fourier infrared spectrometer for detection (the device needs to be started up for 10 minutes for preheating). Each sample was measured 3 times and averaged. The results of near infrared spectroscopy of a batch of samples collected from Chengdu plain (cropland in Chong State demonstration area) according to this method are shown in FIG. 1.

And 2, preprocessing the near infrared spectrum data through an SG smoothing algorithm. The SG smoothing algorithm, Savitzky-Golsy convolution smoothing, can be implemented by the existing software MATLAB. The result of infrared spectrum data acquired in step 1 after being preprocessed by SG smoothing algorithm is shown in fig. 2.

And 3, predicting the preprocessed near infrared spectrum data by adopting a full-wavelength prediction model to obtain a total nitrogen content result.

The full-wavelength prediction model is obtained through modeling by a PLSR algorithm, and the number of main components of the PLSR algorithm is set to be 10. The PLSR algorithm, i.e. Partial least squares regression algorithm (Partial least squares regression), belongs to the prior art. And carrying out model training set and verification set division on the samples in the modeling process. Training and validation sets were as follows 7: 3, namely 70 percent of soil samples are taken as a training set, and 30 percent of soil samples are taken as a verification set. The near infrared spectrum of the soil sample required by modeling is collected by the method in the step 1 and is preprocessed by the method in the step 2, and the total nitrogen content of the soil sample required by modeling is detected by the existing national standard method (HJ 717 2014).

Example 2 analysis of total Nitrogen in soil by characteristic wavelength prediction model

An infrared spectrum analysis method of soil total nitrogen comprises the following steps:

step 1, collecting a near infrared spectrum of a soil sample to obtain near infrared spectrum data.

The specific method comprises the following steps: and drying the soil sample at 105 ℃ for 10 hours. And grinding potassium bromide into powder in a mortar, adding the dried soil sample into the mortar for grinding together, and then putting the ground soil sample into a tablet machine for tabletting. After the tablet pressing is finished, the finished product is put in a Fourier infrared spectrometer for detection (the device needs to be started up for 10 minutes for preheating). Each sample was measured 3 times and averaged. The results of near infrared spectroscopy of a batch of samples collected from Chengdu plain (cropland in Chong State demonstration area) according to this method are shown in FIG. 1.

And 2, preprocessing the near infrared spectrum data through an SG smoothing algorithm. The SG smoothing algorithm, Savitzky-Golsy convolution smoothing, can be implemented by the existing software MATLAB. The result of infrared spectrum data acquired in step 1 after being preprocessed by SG smoothing algorithm is shown in fig. 2.

And step 3A, selecting characteristic wavelengths from the preprocessed near infrared spectrum data by adopting a CARS algorithm, and obtaining the near infrared spectrum data in a characteristic wavelength range. The CARS algorithm, i.e. the competitive adaptive weighting algorithm, belongs to the prior art.

The characteristic wavelengths selected in this embodiment are 1430nm, 1431nm, 1435nm, 1442nm, 1445nm, 1456nm, 1461nm, 1465nm, 1477nm, 1482nm, 2873nm, 2885nm, 2909nm and 2920 nm.

And 3B, predicting the near infrared spectrum data in the characteristic wavelength range by adopting a characteristic wavelength prediction model to obtain a total nitrogen content result.

The characteristic wavelength prediction model is obtained through modeling by a PLSR algorithm. The number of principal components of the PLSR algorithm is set to 8. And carrying out model training set and verification set division on the samples in the modeling process. Training and validation sets were as follows 7: 3, namely 70 percent of soil samples are taken as a training set, and 30 percent of soil samples are taken as a verification set. The near infrared spectrum of the soil sample required by modeling is collected by the method in the step 1 and is preprocessed by the method in the step 2, and the total nitrogen content of the soil sample required by modeling is detected by the existing national standard method (HJ 717 2014).

In order to explain the technical effects of the present invention, the following further describes the technical solution of the present invention by experimental examples.

In the following experimental examples, the main statistical parameters for comparing the merits of different methods include: relative Predictive Development (RPD), coefficient of determination R²A Mean Square Error (MSE), a Root Mean Square Error (RMSE), and a correlation coefficient R.

Experimental example 1 comparison of preprocessing methods and selection of the number of principal components in PLSR Algorithm modeling

1. Pretreatment method

This experimental example was carried out by changing the pretreatment method in step 2 based on the method of example 1, and the pretreatment methods used were: SG smoothing algorithm, moving average method, first derivative method, second derivative method and Standard Normal Variation (SNV). The pre-processed spectral data are shown in FIGS. 2-6. The pretreatment methods all belong to the prior art and can be realized by MATLAB software.

2. Principal component quantity in PLSR algorithm modeling

According to the modeling step of the PLSR algorithm, the number of principal components needs to be determined in the modeling process, which is a very important step of the algorithm, and according to the relevant knowledge of matrix theory, the component arrangement order represents the size of the relevant data information quantity extracted by the scoring factor, and the proportion of the components at the front to the data information quantity is larger. Therefore, it is important to determine the number of components, and too many components may cause loss of the dimension reduction means, and bring much useless noise, thereby affecting the data prediction model and causing overfitting. Too few components lose much useful information in the spectrum, and the prediction effect and accuracy of the model are seriously affected.

In the experimental example, the selection is performed by a method of drawing the variance percentage explained in the variables as a function of the number of the components, and the range of the number of the primarily selected preferred principal components is 6-10. For the different pretreatment methods, the number of main components should be in the range of 6 to 10.

3. Comparison of results

RMSE and R of soil total nitrogen prediction result of optimal model established by each pretreatment method²As shown in table 1. Wherein the number of principal components shown in the table is by RMSE and R²After comparison, the most preferred results for each pretreatment method.

TABLE 1 Total nitrogen modeling results of soil by different pretreatment methods

As can be seen from the above table, when the SG smoothing algorithm is selected for preprocessing, and the number of the principal components is set to 10, the RMSE of the verification set of the established model is lower than that of the models established by other preprocessing methods, and the R of the verification set is lower than that of the models established by other preprocessing methods²Compared with the models established by other preprocessing methods, the method is closer to 1. This indicates that the prediction model is more accurate. Therefore, the SG smoothing algorithm is selected in the preprocessing, and the number of the main components is preferably set to 10.

Experimental example 2 comparison of full-wavelength prediction model modeling algorithms

1. Modeling algorithm

In this experimental example, on the basis of the method of example 1, the modeling algorithm in step 3 was changed, and the prediction accuracy of the modeling result was compared. The modeling algorithms compared are: the method comprises the following steps of a multiple linear regression algorithm (MLR), principal component regression modeling analysis (PCR), partial least squares regression algorithm (PLSR), support vector machine regression modeling analysis (SVR) and artificial neural network Algorithm (ANN), wherein the specific algorithms of the methods are all the prior art and can be realized in MATLAB software.

Specifically, the method comprises the following steps:

(1) multiple linear regression algorithm (MLR)

The spectral data were used as input variables and the total nitrogen content as output. The sample is divided into a training set and a testing set, and the proportion is 7: 3. the specific implementation of the multiple linear regression model algorithm is realized by calling a regression function in matlab.

(2) Principal component regression modeling analysis (PCR)

According to the experimental steps of the principal component regression algorithm, firstly, data are subjected to standardization processing; and the second step is to determine the number of the main components, and finally, to perform regression analysis by using the determined number of the main components, and the specific experimental steps call a zscore function, a pca function and a regression function in the matlab for processing. According to the variance percentage change trend chart of the component number, when the component number reaches more than 9, the variance ratio reaches nearly 100 percent; the mean square error tends to be smooth and stable after the number of the components is 9, and the number of the principal components is preferably 10 in the comprehensive view.

(3) Partial least squares regression algorithm (PLSR)

According to the experimental steps of the partial least square algorithm, the first step is to carry out standardization processing on data; the second step is to determine the quantity of the components to be extracted, and since the Y matrix only has one variable of the total nitrogen content, the Y matrix does not need to be subjected to variable extraction; the third step is to perform regression analysis using the determined amounts of the components. According to the results of experimental example 1, 10 major components were measured. In the specific experimental steps, a zcore function, a Plregress function and the like are used for calculation and analysis in matlab.

(4) Support vector machine regression modeling analysis (SVR)

The algorithm is realized through a libsvm library of Matlab, wherein a kernel function selects a radial basis function, the penalty factor value is 9.38, and the gamma value is 5.

(5) Artificial neural network Algorithm (ANN)

The algorithm is implemented in Matlab. The sample set division mode adopts a random division method, and the training set, the verification set and the test set are respectively according to the following steps of 7: 1.5: 1.5. The number of hidden layers is default to 10, and an output layer is added, and the specific network structure is shown in fig. 7. In the figure, input represents the variable input and represents the near infrared spectrum matrix in this experiment, 1272 is the number of variables and represents the number of spectral wavelengths in this experiment. w is a weight vector, b is a bias coefficient, and output is a predicted value of the nitrogen content of the soil sample. The optimal number of training sessions is 10.

2. Modeling results

The predicted results for MLR, PCR, PLSR, SVR, and ANN are shown in FIGS. 8-12, respectively.

The prediction results of the specific modeling algorithms on the test set are shown in table 2.

TABLE 2 modeling results of different spectral modeling algorithms

Comparing the correlation coefficient R and the determination coefficient R in the table²And the evaluation indexes such as root mean square error RMSE and relative analysis error RPD. The indexes of the PLSR algorithm and the ANN algorithm are superior to those of the other three algorithms. Therefore, the PLSR algorithm and the ANN algorithm can be considered to be superior. In addition, although the ANN algorithm correlation coefficient 0.933 is higher than the PLSR correlation coefficient 0.883, the PLSR RMS error 55.97mg/kg is significantly smaller than the ANN RMS error 85.13mg/kg, the RMS error represents the deviation between the predicted and measured values of the model, and the PLSR relative analysis error is also higher than the ANN. In summary, the model established by the PLSR algorithm has the best prediction performance.

Experimental example 3 comparison of characteristic wavelength selection algorithms

1. Characteristic wavelength selection algorithm

In this experimental example, on the basis of the method in example 2, the characteristic wavelength selection algorithm in step 3 is changed, and the prediction accuracy of the modeling result is compared. The compared characteristic wavelength selection algorithms are: a competition adaptive re-weighting algorithm (CARS) and a Random-frog algorithm (Random-frog), and the specific algorithms of the methods are all the prior art.

Specifically, the method comprises the following steps:

(1) competition adaptive reweighting algorithm (CARS)

Implemented in Matlab using the carspls function. The model iteration is stopped at an iteration number of 27. According to the result of the iterative model at this time, 10 characteristic bands are extracted, which are respectively: 1431nm, 1442nm, 1445nm, 1477nm, 1482nm, 2650nm, 2758nm, 2838nm, 2858nm, 2870nm, 2915nm and 2920 nm.

(2) Random frog-leaping algorithm (Random-frog)

Implemented in Matlab using randomFrog function. The selected characteristic wavelengths are 1430nm, 1431nm, 1435nm, 1442nm, 1445nm, 1456nm, 1461nm, 1465nm, 1477nm, 1482nm, 2873nm, 2885nm, 2909nm and 2920nm respectively.

2. Predicted results

The results of the prediction using the characteristic wavelengths selected by CARS and Random-from are shown in fig. 13 and 14.

Specifically, for the prediction of the verification set, when the characteristic wavelength selection algorithm adopts CARS, the correlation coefficient R is 0.889, the correlation coefficient value is high, and the coefficient R is determined²The root mean square error was 65.94mg/kg, 0.791. The RPD value was 5.76, greater than 2.5, indicating that the model can be used for quantitative predictive analysis.

When Random-from is selected as the characteristic wavelength, the correlation coefficient R is 0.871, the correlation coefficient value is high, and the coefficient R is determined²The root mean square error was 69.17mg/kg, which was 0.759. The RPD value was 5.31, greater than 2.5, indicating that the model can be used for quantitative predictive analysis.

The results show that both characteristic wavelength selection algorithms CARS and Random-from can be used, wherein CARS is superior.

Experimental example 4 comparison of full wavelength prediction model and characteristic wavelength prediction model

In this experimental example, the performance of the full-wavelength prediction model and the performance of the characteristic-wavelength prediction model are compared, and specifically, the prediction models of example 1 and example 2 are compared.

Example 1 predicted R ═ 0.883, RMSE ═ 55.97mg/kg, and RPD 6.46 for the validation set; example 2 the validation set predicted R0.889, RMSE 65.94mg/kg, and an RPD of 5.76.

The R values of the two examples are almost consistent, while the RMSE and RPD values are better than those of example 1, so that the full-wavelength prediction model (example 1) has better prediction performance after the pretreatment step is added.

As can be seen from the above examples and experimental examples, the method of the invention adds a pretreatment step in the infrared spectroscopic analysis method of the total nitrogen in the soil, and simultaneously optimizes a pretreatment algorithm, a modeling algorithm, modeling parameters and the like. The method can improve the accuracy of the model for predicting the total nitrogen content of the soil, thereby having good application prospect in the fields of agriculture, environmental protection, biological research and the like.