Rapid detection method for petroleum hydrocarbon in soil based on visible-near infrared spectrum

Abstract

The invention discloses a method for rapidly detecting petroleum hydrocarbon in soil based on visible-near infrared spectroscopy. The method comprises the following steps: preparing standard samples with the concentration gradient of 0-3000 mg / kg, and establishing a spectrum database; preprocessing the spectral data by adopting Savitzky-Golay filtering and standard normal variable transformation; using a competitive self-adaptive reweighted sampling algorithm to optimize the characteristic wavelength, and constructing a partial least square regression quantitative model; and determining the optimal performance of the CARS algorithm by comparing three feature selection methods of full spectrum, BOSS and CARS, and the like. According to the present invention, the petroleum hydrocarbon content in the soil can be accurately, rapidly and quantitatively detected, the standard detection process is established, the gas chromatography is adopted to carry out comparison verification on the standard sample, the correlation coefficient of the two methods achieves 0.9988, the average relative error is 5.28%, and the accuracy and the reliability of the spectrum detection method are verified.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of environmental pollution detection, and particularly relates to a rapid detection method for petroleum hydrocarbon in soil based on visible-near infrared spectrum. BACKGROUND

[0002] With the widespread use of petroleum products, the problem of petroleum hydrocarbon pollution in soil is becoming increasingly serious. The traditional detection method for petroleum hydrocarbon pollution in soil mainly relies on chemical analysis means such as gas chromatography, and has problems such as long detection period, high cost, complex operation, and the need for a large amount of organic solvent.

[0003] As a kind of rapid and non-destructive detection means, visible-near infrared spectrum technology has great potential in soil pollution detection. However, the existing spectrum detection technology has the following shortcomings in the rapid detection of petroleum hydrocarbon in soil: 1. Lack of special spectrum database and detection model for petroleum hydrocarbon pollution; 2. The spectrum feature extraction method is not accurate enough; 3. The detection precision and stability need to be improved; 4. Lack of standardized detection process.

[0004] Therefore, it is urgent to develop a high-precision spectrum detection method specially for petroleum hydrocarbon contaminated soil. SUMMARY

[0005] The application is proposed to solve the problems in the prior art, and aims to provide a rapid detection method for petroleum hydrocarbon in soil based on visible-near infrared spectrum.

[0006] The application is implemented through the following technical solutions: A rapid detection method for petroleum hydrocarbon in soil based on visible-near infrared spectrum, comprising the following steps: (I) preparing a series of petroleum hydrocarbon concentration standard soil samples; (II) acquiring the diffuse reflectance spectrum signal of the sample prepared in step (I) by using a visible-near infrared spectrometer; (III) preprocessing the diffuse reflectance spectrum signal of the sample acquired in step (II); (IV) respectively adopting full spectrum modeling, BOSS algorithm and CARS algorithm for modeling key wavelength screening, comparing the performances of different algorithms through cross-validation, and determining that the CARS algorithm is the optimal method for screening modeling key wavelength; (V) establishing a quantitative analysis model by using a partial least squares regression algorithm, verifying the performance of the quantitative analysis model by using leave-one-out cross-validation method, and determining the optimal quantitative analysis model; (VI) implementing comparative verification on the sample by using a gas chromatography method; (Ⅶ) Using the finally determined quantitative analysis model to detect the petroleum hydrocarbon content of unknown soil samples.

[0007] In the above technical solution, the specific method for preparing the series of petroleum hydrocarbon concentration standard soil samples is as follows: first, collect a type of clean soil sample, pass it through a 2 mm sieve for use; then, mix the clean soil with petroleum hydrocarbons at a set concentration; finally, place the mixed sample in a sealed glass container and stand at room temperature for 24 hours to obtain the series of petroleum hydrocarbon concentration standard soil samples. In the above technical solution, the concentration gradient of the series of petroleum hydrocarbon concentration standard soil samples is set as 0, 50 mg / kg, 100 mg / kg, 250 mg / kg, 500 mg / kg, 750 mg / kg, 1000 mg / kg, 1500 mg / kg, 2000 mg / kg, 2500 mg / kg, and 3000 mg / kg.

[0008] In the above technical solution, step (Ⅱ) is specifically as follows: in a darkroom environment, use a visible-near infrared spectrometer to perform diffuse reflection measurement.

[0009] In the above technical solution, the specific method for diffuse reflection measurement is as follows: use a contact type optical fiber probe to directly contact the soil sample, continuously measure each sample 6 times, and take the average value as the spectral data of the sample.

[0010] In the above technical solution, the visible-near infrared spectrometer uses a diffuse reflection detection mode, and the spectral signal acquisition parameters of the visible-near infrared spectrometer are as follows: wavelength interval is 200 nm~2500 nm, and spectral resolution is 1 nm.

[0011] In the above technical solution, the pretreatment of step (Ⅲ) specifically includes the following steps: (Ⅲ-ⅰ) Use Savitzky-Golay filtering to eliminate noise; (Ⅲ-ⅱ) Perform standard normal variable transformation processing to eliminate scattering effects.

[0012] In the above technical solution, step (Ⅴ) specifically includes the following steps: (Ⅴ-ⅰ) Construct a data matrix: construct the full-spectrum modeling spectral matrix X 全光谱 , the BOSS algorithm spectral matrix X BOSS , and the CARS algorithm spectral matrix X CARS corresponding to the full-spectrum modeling, the BOSS algorithm, and the CARS algorithm, respectively, using the spectral signal data of the screening modeling key wavelengths screened by the full-spectrum modeling, the BOSS algorithm, and the CARS algorithm; (Ⅴ-ⅱ) respectively establish PLSR quantitative analysis models corresponding to the full-spectrum modeling, the BOSS algorithm, and the CARS algorithm: (V-III) comparing the prediction performance of the PLSR prediction models established by full spectrum modeling, BOSS algorithm and CARS algorithm through leave-one-out cross validation, and determining the optimal quantitative analysis model.

[0013] In the above technical solution, the evaluation index of the prediction performance of the model is correlation coefficient R 2 , root mean square error RMSE and relative prediction deviation RPD; and the optimal quantitative analysis model is the PLSR quantitative analysis model established by the CARS algorithm.

[0014] In the above technical solution, the step (VII) specifically comprises the following steps: (VII-I) sample pretreatment: Passing the soil sample to be measured through a 2mm sieve and loading into a sample measuring device; (VII-II) spectrum data acquisition: Acquiring the diffuse reflection spectrum of the sample to be measured prepared in step (VII-I) by using the same method as in step (II); (VII-III) spectrum pretreatment: Pretreating the diffuse reflection spectrum data acquired in step (VII-II) by using the same method as in step (III); (VII-IV) characteristic wavelength extraction: Extracting the absorbance values of the corresponding wavelengths from the pretreated spectrum by using the modeling key wavelengths screened out by the CARS algorithm to form a characteristic vector; (VII-V) concentration prediction: inputting the characteristic vector into the established CARS-PLSR quantitative analysis model to output the predicted petroleum hydrocarbon concentration value.

[0015] The present application has the following beneficial effects: The present application provides a rapid detection method for petroleum hydrocarbon in soil based on visible-near infrared spectrum, which can accurately and rapidly quantitatively detect the petroleum hydrocarbon content in soil, establish a standardized detection process, and compare and verify the standard samples by using gas chromatography, with a correlation coefficient of 0.9988 and an average relative error of 5.28%, thereby verifying the accuracy and reliability of the spectral detection method.

[0016] The detection method of the present application has strong specificity, is a spectral detection method specially established for petroleum hydrocarbon pollution, has high detection precision, is rapid in detection, with a single sample detection time of less than 1 minute and significantly improved efficiency, is simple to operate without complex sample pretreatment, has a standardized detection process, is low in cost, does not require organic solvents compared with the traditional gas chromatography method, greatly reduces the detection cost, is green and environmentally friendly, is non-destructive detection without secondary pollution, optimizes the algorithm, compares and determines the optimal detection scheme through multiple characteristic selection algorithms, and is reliable in verification, which ensures the accuracy of the detection results. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention.

[0018] For those skilled in the art, other related figures can be obtained from the above figures without any creative effort. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0020] like Figure 1 As shown, a rapid detection method for petroleum hydrocarbons in soil based on visible-near-infrared spectroscopy includes the following steps: (I) Preparation of a series of soil samples with petroleum hydrocarbon concentration standards; The specific method for preparing a series of petroleum hydrocarbon concentration standard soil samples is as follows: First, collect a type of clean soil sample and pass it through a 2 mm sieve for later use; then mix the clean soil with petroleum hydrocarbons according to the set concentration; finally, put the mixed sample into a sealed glass container and let it stand at room temperature for 24 hours to obtain a series of petroleum hydrocarbon concentration standard soil samples. The concentration gradients of the series of petroleum hydrocarbon concentration standard soil samples were set as follows: 0, 50 mg / kg, 100 mg / kg, 250 mg / kg, 500 mg / kg, 750 mg / kg, 1000 mg / kg, 1500 mg / kg, 2000 mg / kg, 2500 mg / kg and 3000 mg / kg. (II) Obtain the diffuse reflectance spectral signal of the sample prepared in step (I) using a visible-near infrared spectrometer; The specific method for obtaining the diffuse reflectance spectral signal is as follows: in a dark room environment, a visible-near-infrared spectrometer is used to measure diffuse reflectance. The visible-near-infrared spectrometer adopts a diffuse reflectance detection mode. The spectral signal acquisition parameters of the visible-near-infrared spectrometer are: wavelength range of 200 nm to 2500 nm, and spectral resolution of 1 nm. The specific method for measuring diffuse reflectance is as follows: a contact fiber optic probe is used to directly contact the sample, and each sample is measured 6 times continuously. The average value is taken as the diffuse reflectance spectral signal data of the sample. (III) Preprocess the diffuse reflectance spectral signal data of the sample obtained in step (II); The preprocessing specifically includes the following steps: (III-i) Noise is eliminated using Savitzky-Golay (SG) filtering; The specific processing procedure is that the window width of the SG filter is set to 9 data points, the polynomial order is set to 2, and the original diffuse reflection spectrum signal data of the sample is smoothed; the SG filter eliminates high-frequency noise by least square polynomial fitting in a moving window, while maintaining the shape and intensity information of the spectral peak; (III-ii) Standard Normal Variate (SNV) processing is performed to eliminate scattering effects; The specific processing procedure is that the spectral data of each sample after step (III-i) is processed for centering and standardization to eliminate multiplicative scattering effects and additive baseline drift caused by factors such as particle size and density difference; The calculation formula of the normal variable transformation processing is: x_corrected=(x-mean(x)) / std(x), In the formula, x is the original spectrum vector, mean(x) and std(x) are the mean and standard deviation of the spectrum, respectively; (IV) A variety of feature selection algorithms (full spectrum modeling, BOSS algorithm, CARS algorithm) are used for modeling key wavelength screening, and the performance of different algorithms is compared through cross-validation to determine that the CARS algorithm is the optimal method for screening modeling key wavelengths; The specific processing procedure is: The full spectrum modeling is specifically: all spectrum data points in the wavelength range of 300-2500 nm are used for full spectrum modeling; full spectrum modeling does not perform wavelength screening, and all 2301 wavelength variables are retained as modeling inputs; full spectrum modeling is used as a control benchmark to evaluate the effect of the feature selection algorithm; The BOSS algorithm is specifically: first, B times Bootstrap sampling (B=1000) is performed on the training set (75% of the samples are selected); the number of samples in each sampling is the same as that of the original training set; a partial least squares regression model is established for each Bootstrap sample, and the regression coefficient of each wavelength variable is calculated; the frequency and average coefficient size of each wavelength in B times modeling are counted; according to the comprehensive score of the selection frequency and the coefficient size, a soft threshold method is used to select the modeling key wavelength; The CARS algorithm is specifically: combining Monte Carlo sampling and the competition mechanism of partial least squares regression coefficients, the most representative modeling key wavelength is selected through an adaptive reweighted sampling strategy; The implementation steps of the algorithm include: firstly, setting the number of Monte Carlo sampling N=50; in each sampling, weighted random sampling is performed according to the current wavelength weight, and the sampling ratio gradually decreases from 80% to 2%; a PLS model is established for the wavelength subset obtained by sampling, and the cross-validation root mean square error (RMSECV) is calculated; the wavelength weight is updated according to the PLS model performance and the size of the regression coefficient, and the wavelength with good performance and large coefficient obtains a higher weight; after N iterations, the wavelength subset corresponding to the minimum RMSECV is selected as the optimal feature, that is, the modeling key wavelength; (V) A partial least squares regression algorithm is used to establish a quantitative analysis model, and a leave-one-out cross-validation method is used to verify the performance of the quantitative analysis model; The specific modeling process is as follows: (V-ⅰ) Construct a data matrix: The spectral signal data of the modeling key wavelengths screened by full spectrum modeling, the BOSS algorithm and the CARS algorithm are used to construct a full spectrum modeling spectral matrix X 全光谱 (n 全光谱 ×p 全光谱 , wherein n 全光谱 is the number of samples of the full spectrum modeling, p 全光谱 is the number of wavelength variables of the full spectrum modeling), a BOSS algorithm spectral matrix X BOSS (n BOSS ×p BOSS , wherein n BOSS is the number of samples of the BOSS algorithm, p BOSS is the number of wavelength variables of the BOSS algorithm) and a CARS algorithm spectral matrix X CARS (n CARS ×p CARS , wherein n CARS is the number of samples of the CARS algorithm, p CARS is the number of wavelength variables of the CARS algorithm), and the petroleum hydrocarbon concentration values are constructed into a response matrix Y (n×1); (V-ⅱ) The PLSR quantitative analysis models corresponding to the full spectrum modeling, the BOSS algorithm and the CARS algorithm are respectively established: The specific modeling process is as follows: the preprocessed spectral matrix X i (i=full spectrum, BOSS, CARS) and the response matrix Y are simultaneously decomposed into the product form of the score matrix and the loading matrix: , wherein T and U are the score matrices of X i and Y, P and Q are the corresponding loading matrices, E and F are the residual matrices; PLSR determines the latent factors by maximizing the covariance between T and U, so as to ensure that the extracted factors can not only explain the variance of X i , but also have the maximum correlation with Y; (V-III) Leave-one-out cross-validation optimization and performance evaluation: the prediction performance of the PLSR prediction model established by the three wavelength selection methods was compared through leave-one-out cross-validation; The specific processing process is: the search range of the number of principal components is set to 1-20; for each number of principal components, leave-one-out cross-validation is performed, that is, each time one sample is left out as a validation sample, and the remaining samples are used to establish a PLSR model to predict the concentration of the left-out sample; the root mean square error (RMSECV) of all samples is calculated; the number of principal components corresponding to the minimum RMSECV is selected as the optimal value; The evaluation index of the prediction performance of the model is the correlation coefficient R 2 , the root mean square error RMSE and the relative prediction deviation RPD; Considering the prediction accuracy, model complexity and calculation efficiency, the CARS-PLSR quantitative analysis model based on the CARS optimization algorithm feature selection is determined as the best detection model; (VI) The sample is subjected to comparative verification by gas chromatography to evaluate the accuracy of the spectral quantitative analysis method, specifically including the following steps: (VI-I) The series of standard samples prepared in step (I) are subjected to petroleum hydrocarbon content detection by gas chromatography (GC); (VI-II) The gas chromatography detection results are compared and verified with the CARS-PLSR spectral prediction model results; (VI-III) The accuracy and reliability of the spectral detection method are evaluated; (VII) Model application method The specific application steps of the established CARS-PLSR quantitative analysis model for unknown soil sample petroleum hydrocarbon content detection are: (VII-I) Sample pretreatment: the soil sample to be tested is passed through a 2 mm sieve and loaded into a sample tester; (VII-II) Spectral data acquisition: the diffuse reflectance spectrum of the sample to be tested is acquired by using the same spectrometer parameters (wavelength 200-2500 nm, resolution 1 nm); (VII-III) Spectral pretreatment: the spectrum of the sample to be tested is sequentially subjected to SG filtering and SNV transformation to ensure consistency with the pretreatment method of the modeling sample; (VII-IV) Feature wavelength extraction: according to the key wavelengths selected by the CARS algorithm, the absorbance values of the corresponding wavelengths are extracted from the pretreated spectrum to form a feature vector; (VII-V) Concentration prediction: the feature vector is substituted into the established CARS-PLSR quantitative analysis model to output the predicted petroleum hydrocarbon concentration value.

[0021] Example 1 A rapid detection method of petroleum hydrocarbon in soil based on visible-near infrared spectrum, comprising the following steps: (I) preparing a series of petroleum hydrocarbon concentration standard soil samples; (I-i) collecting clean forest soil with water content of 19.52% and bulk density of 0.001 g / cm 3 , and passing through a 2 mm sieve for use; (I-ii) selecting commercially available 0# diesel as the test petroleum hydrocarbon; (I-iii) accurately weighing a plurality of dry soil samples obtained in step (I-i), each of which is 100 g, and adding 0, 5 mg, 10 mg, 25 mg, 50 mg, 75 mg, 100 mg, 150 mg, 200 mg, 250 mg and 300 mg of commercially available 0# petroleum hydrocarbon to the dry soil samples, respectively, fully mixing and sealing for 24 h, thereby obtaining a series of petroleum hydrocarbon concentration standard soil samples; (II) acquiring the diffuse reflectance spectrum signal of the sample by using a visible-near infrared spectrometer; The visible-near infrared spectrometer is a Leisun optical LSFS-241205-026 visible-near infrared spectrometer; The specific spectrum measurement is as follows: in a dark environment, a contact optical fiber probe is vertically contacted with the surface of the sample, each sample is repeatedly measured 6 times, and the average value is taken as the spectrum data of the sample; (III) performing SG filtering and SNV transformation on the diffuse reflectance spectrum signal of the sample obtained in step (II), and then establishing a PLSR model by using full spectrum, BOSS and CARS algorithms, respectively; The same standard sample data set is used to establish PLSR models of full spectrum, BOSS and CARS three methods, respectively: Full spectrum-PLSR model: 1. wavelength range used: 200-2500 nm (all 2297 wavelength points) 2. cross-validation results: R 2 =0.9926, RMSE=66.09 mg / kg, RPD=4.00 BOSS-PLSR model: 1. number of selected wavelengths: 30 important wavelengths 2. cross-validation results: R 2 =0.9970, RMSE=42.12 mg / kg, RPD=5.21 CARS-PLSR model: 1. number of selected wavelengths: 40 optimal wavelengths 2. cross-validation results: R 2= 0.9974, RMSE = 39.32 mg / kg, RPD = 6.15 The results show that the CARS algorithm greatly reduces the number of variables while maintaining high accuracy, and the model performance is optimal.

[0022] (IV) Comparison of the prediction performance of the three models by leave-one-out cross-validation The results are as follows: Full spectrum-PLSR model: R 2 = 0.9926, RMSE = 66.09 mg / kg; BOSS-PLSR model: R 2 = 0.9970, RMSE = 42.12 mg / kg; CARS-PLSR model: R 2 = 0.9974, RMSE = 39.32 mg / kg; The CARS algorithm selects 40 optimal wavelengths, and the model performance is the best.

[0023] (V) Comparison and verification of standard samples by gas chromatography to evaluate the accuracy of spectral quantitative analysis method.

[0024] Select standard samples with different concentration gradients, and detect the content of petroleum hydrocarbon by CARS-PLSR spectral model and gas chromatography (refer to "Determination of Petroleum Hydrocarbons (C10-C40) in Soil and Sediment - Gas Chromatography" (HJ 1021-2019)): GC detection conditions: 1. Gas chromatograph equipped with hydrogen flame ionization detector (FID) 2. Detection method: "Determination of Petroleum Hydrocarbons (C10-C40) in Soil and Sediment - Gas Chromatography" (HJ 1021-2019) The comparison results are shown in Table 1: Table 1 Statistical analysis found that the correlation coefficient of spectral method and GC method was R 2 = 0.9988, the average relative error was 5.28%. The results show that the spectral quantitative method based on CARS optimization algorithm has good consistency with the traditional gas chromatography method, and verifies the accuracy and reliability of the spectral quantitative method.

[0025] The applicant states that the above description is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and it should be understood by those skilled in the art that any changes or replacements within the technical scope disclosed by the present application can be easily thought out by those skilled in the art, and all of them fall within the protection scope and disclosure scope of the present application.

Claims

1. A method for rapid detection of petroleum hydrocarbons in soil based on visible-near infrared spectroscopy, characterized in that: It comprises the following steps: (I) preparing a series of petroleum hydrocarbon concentration standard soil samples; (II) obtaining the diffuse reflectance spectrum signal of the sample prepared in step (I) by using a visible-near infrared spectrometer; (III) pre-processing the diffuse reflectance spectrum signal of the sample obtained in step (II); (IV) respectively adopting full spectrum modeling, BOSS algorithm and CARS algorithm to screen the key wavelength, comparing the performance of different algorithms through cross validation, and determining that CARS algorithm is the optimal method for screening the key wavelength for modeling; (V) adopting partial least squares regression algorithm to establish a quantitative analysis model and verifying the performance of the quantitative analysis model by using leave-one-out cross validation method, and determining the optimal quantitative analysis model; (VI) implementing comparison and verification on the sample by using gas chromatography method; (VII) applying the finally determined quantitative analysis model to detect the petroleum hydrocarbon content of unknown soil samples.

2. The method for rapid detection of petroleum hydrocarbons in soil based on visible-near infrared spectroscopy according to claim 1, characterized in that: The specific method for preparing the series of petroleum hydrocarbon concentration standard soil samples is as follows: first, collecting a type of clean soil sample, and passing it through a 2 mm sieve for use; then, mixing the clean soil with petroleum hydrocarbon at a set concentration; finally, putting the mixed sample into a sealed glass container, and standing it at room temperature for 24 hours to obtain the series of petroleum hydrocarbon concentration standard soil samples.

3. The method for rapid detection of petroleum hydrocarbons in soil based on visible-near infrared spectroscopy according to claim 2, characterized in that: The concentration gradient of the series of petroleum hydrocarbon concentration standard soil samples is set as 0, 50 mg / kg, 100 mg / kg, 250 mg / kg, 500 mg / kg, 750 mg / kg, 1000 mg / kg, 1500 mg / kg, 2000 mg / kg, 2500 mg / kg and 3000 mg / kg.

4. The method for rapid detection of petroleum hydrocarbons in soil based on visible-near infrared spectroscopy according to claim 1, characterized in that: The step (II) specifically comprises the following: in a darkroom environment, diffuse reflectance measurement is performed by using a visible-near infrared spectrometer.

5. The method for rapid detection of petroleum hydrocarbons in soil based on visible-near infrared spectroscopy according to claim 4, characterized in that: The specific method for the diffuse reflectance measurement is as follows: a contact type optical fiber probe is used to directly contact the soil sample, each sample is continuously measured for 6 times, and the average value is taken as the spectrum data of the sample.

6. The method for rapid detection of petroleum hydrocarbons in soil based on visible-near infrared spectroscopy according to claim 4, characterized in that: The visible-near infrared spectrometer adopts a diffuse reflectance detection mode, and the spectrum signal acquisition parameters of the visible-near infrared spectrometer are as follows: wavelength interval is 200 nm~2500 nm, and spectrum resolution is 1 nm.

7. The method for rapid detection of petroleum hydrocarbons in soil based on visible-near infrared spectroscopy according to claim 1, characterized in that: The pre-processing of the step (III) specifically comprises the following steps: (III-ⅰ) adopting Savitzky-Golay filtering to eliminate noise; (III-ⅱ) performing standard normal variable transformation processing to eliminate scattering effect.

8. The method for rapid detection of petroleum hydrocarbons in soil based on visible-near infrared spectroscopy according to claim 1, characterized in that: The step (V) specifically comprises the following steps: (V-i) Constructing data matrix: the spectral signal data of the screening modeling key wavelengths screened by full spectrum modeling, BOSS algorithm and CARS algorithm are used to construct full spectrum modeling spectral matrix X 全光谱 , BOSS algorithm spectral matrix X BOSS and CARS algorithm spectral matrix X CARS , and the petroleum hydrocarbon concentration value is constructed into response matrix Y; (V-ⅱ) respectively establishing PLSR quantitative analysis models corresponding to full spectrum modeling, BOSS algorithm and CARS algorithm: (V-ⅲ) comparing the prediction performance of the PLSR prediction models established by full spectrum modeling, BOSS algorithm and CARS algorithm through leave-one-out cross validation, and determining the optimal quantitative analysis model.

9. The method for rapid detection of petroleum hydrocarbons in soil based on visible-near infrared spectroscopy according to claim 8, characterized in that: The evaluation index of the prediction performance of the model is correlation coefficient R 2 , root mean square error RMSE and relative prediction deviation RPD; the optimal quantitative analysis model is a PLSR quantitative analysis model established by CARS algorithm.

10. The method for rapid detection of petroleum hydrocarbons in soil based on visible-near infrared spectroscopy according to claim 1, characterized in that: The step (VII) specifically comprises the following steps: (VII-ⅰ) sample pre-processing: passing the soil sample to be measured through a 2 mm sieve, and putting it into a sample container; (VII-ⅱ) spectrum data acquisition: obtaining the diffuse reflectance spectrum of the sample to be measured prepared in step (VII-ⅰ) by using the same method in step (II); (VII-ⅲ) spectrum pre-processing: The diffuse reflectance spectrum data obtained in step (VII-1) is pretreated by the same method as in step (III); (Ⅶ-ⅳ) Feature wavelength extraction: The absorbance values of the corresponding wavelengths are extracted from the pretreated spectrum to form a feature vector by using the key wavelengths screened by the CARS algorithm; (Ⅶ-ⅴ) Concentration prediction: The feature vector is substituted into the established CARS-PLSR quantitative analysis model to output the predicted petroleum hydrocarbon concentration value.

Images