Fruit maturity comprehensive evaluation and shelf life prediction method based on near infrared spectrum technology

Abstract

The invention discloses a fruit maturity comprehensive evaluation and shelf life prediction method based on a near infrared spectrum technology, relates to the technical field of agricultural product detection, and solves the problems that an existing nondestructive detection method is single, low in accuracy, incapable of meeting requirements and inaccurate in shelf life prediction. A quality and near infrared spectrum change model of fruits in the whole biological process in the actual production process is established, the optimal harvesting time is accurately determined through a fruit maturity comprehensive evaluation model, the fruit quality is guaranteed, and loss is reduced; in addition, by establishing a fruit shelf life prediction model, the shelf life of the fruits is judged, and proper circulation channels and selling time are selected, so that it is guaranteed that the consumers receive the fruits with the best quality. The method has the effects of improving the accuracy of timely harvesting of the fruits, realizing judgment of the optimal harvesting time of the fruits and field grading and sorting of the fruits, and accurately predicting the shelf life after harvesting.

Description

Technical Field

[0001] The present application relates to the technical field of agricultural product detection, and in particular to a method for comprehensive evaluation of fruit maturity and shelf life prediction based on near-infrared spectroscopy technology. Background Art

[0002] Nondestructive testing technology is a technology that uses changes in physical quantities (such as sound, light, electricity, and magnetism) caused by abnormal internal structure or defects of materials to inspect or test internal and surface defects of materials, parts, equipment, etc., as well as certain physical properties, organizational status, and chemical composition of materials without destroying or damaging the performance, internal structure, or appearance of the object being tested.

[0003] Among them, near-infrared spectroscopy is commonly used in non-destructive testing of fruits. Near-infrared spectroscopy refers to electromagnetic waves with wavelengths between visible light and mid-infrared light, ranging from 780nm to 2526nm. In the absorption band of the near-infrared region, the near-infrared spectrum is formed by measuring the absorption information such as the frequency doubling and sum frequency of hydrogen-containing groups (-OH, -NH and -CH functional groups) inside the molecule. The position of the wavelength and the intensity of the absorption band in the infrared absorption spectrum reflect the characteristics of the molecular structure and can be used to identify the chemical groups, structural composition, and purity identification of unknown substances for qualitative and quantitative analysis.

[0004] At present, most of the existing non-destructive testing methods for the intrinsic quality of fruits are mainly based on a single physical and chemical index for evaluation, and only static measurements at a certain moment after the fruit is picked are selected. This testing and modeling method extracts a predicted sample set from all samples at one time when performing set division, which cannot meet the needs of non-destructive testing of the intrinsic quality of the entire time period from fruit picking to the consumer's table required for industrial development; the growth, development and ripening process of fruits is comprehensively characterized by multiple indicators such as hardness, pH value, SSC (soluble solids), DM (dry matter content), etc. The existing technology detection accuracy is not enough and cannot meet the needs.

[0005] In addition, fruits harvested too early have insufficient accumulation of soluble solids (SSC) and dry matter (DM), and lack flavor and have low quality after ripening. Fruits harvested too late have aged pulp fibers, poor edible quality, and a short shelf life. Therefore, harvesting too early or too late will affect the commercial value of fruits. Summary of the invention

[0006] The present application provides a method for comprehensive evaluation of fruit maturity and prediction of shelf life based on near-infrared spectroscopy technology, the purpose of which is to improve the accuracy of timely fruit harvesting, to achieve the determination of the optimal fruit harvesting period and the field grading and sorting of fruits, as well as the prediction of post-harvest shelf life.

[0007] The present application provides a method for comprehensive evaluation of fruit maturity and prediction of shelf life based on near infrared spectroscopy technology, which adopts the following technical solutions: A fruit maturity comprehensive evaluation method based on near infrared spectroscopy technology comprises the following steps: S1: Under room temperature conditions, the near-infrared spectrometer was preheated and stabilized, and the acquisition parameters were set. The acquisition wavelength range of the near-infrared spectrometer was 600-1100nm, and the number of scans was 50 times; S2: Select several fruits and scan them respectively, and output the average value of 50 spectral values ​​of each point scanned 50 times as the spectral value of the point; S3: Collect the absorbance values ​​of the measuring points of the fruits from picking to ripening, select the original absorbance values ​​of the fruits at different time after picking, calculate the average absorbance of the fruits at different wavelengths, establish a coordinate system, and draw a graph of the average absorbance change at different time after picking the fruits; S4: Perform an outlier test on the fruit spectral data obtained by the measurement. After removing the singular points, use the sample algorithm to divide the measured data of fruit hardness, pH value, SSC, and DM quality into a calibration set and a prediction set. Use the calibration set data to train the model, use the prediction set data to test the model performance, and display the distribution characteristics of the sample set data through the comprehensive set. S5: preprocess the spectra using a preprocessing algorithm to remove noise in the spectra; S6: Analyze the distribution and contribution rate of the principal component eigenvalues ​​of the indicators of different varieties of fruits, calculate the score coefficient matrix of the components, obtain the principal component expression, and construct the comprehensive evaluation index F of the maturity of different varieties of fruits: F=a*Zx1+b*Zx2+c*Zx3+d*Zx4; In the formula, Zx1, Zx2, Zx3, and Zx4 are the hardness value, pH value, SSC value, and DM value after standardization, respectively; a, b, c, and d represent different fruit varieties; S7: The constructed formula for comprehensive evaluation of fruit maturity was used to calculate the F value of each variety, and a box plot was used to conduct a comprehensive evaluation of fruit maturity using the F values ​​of samples with different maturity levels.

[0008] Preferably, the preprocessing algorithm includes multivariate scatter correction, and the multivariate scatter correction includes the following calculation formula: .

[0009] Where: X represents the spectral data matrix of the modeling set samples, Xi is the spectral data of the i-th sample, ai and bi are the slope and intercept of the linear regression of the i-th spectrum Xi and the average spectrum X.

[0010] Preferably, the preprocessing algorithm includes a standard normal transformation, and the standard normal transformation includes the following calculation formula:

[0011] Where: X represents the spectral data matrix of the modeling set samples, Xi is the spectral data of the i-th sample, and X is the average spectrum.

[0012] Preferably, the preprocessing algorithm includes SG convolution derivative, and the SG convolution derivative includes the following calculation formula:

[0013] In the formula: dm / dxm represents the operator of the m-th order derivative; y(x) represents the value of the original spectral signal (or data point) at the position; ∑ represents summation, which is used to perform weighted summation on the data in the window; k represents the index in the window; N represents the window size, that is, the length of the filter; m represents the order of the derivative; Ck(m) represents the convolution coefficient, which represents the weighted coefficient calculated after the polynomial fitting, and these coefficients are determined according to the window size N and the derivative order m; y(xk) represents the value of the original signal at point xk, where xk is a neighborhood point centered on the current point x.

[0014] Preferably, the preprocessing algorithm includes convolution smoothing, and convolution smoothing includes the following calculation formula:

[0015] Where: After the window width is determined, the convolution weight ωk is obtained by looking up the table to obtain the corresponding ωk.

[0016] Preferably, the preprocessing algorithm includes normalization, and the normalization includes the following calculation formula:

[0017] Where: , n is the number of wavelength points, k=1, 2, ..., m.

[0018] Preferably, the preprocessing algorithm includes maximum-minimum normalization, and the maximum-minimum normalization includes the following calculation formula:

[0019] Where: X is the original data value, which represents a single data point that needs to be normalized; Xnorm is the normalized data value, that is, the value after maximum-minimum normalization processing; Xmin is the minimum value in the data set, that is, the minimum value in the original data; Xmax is the maximum value in the data set, that is, the maximum value in the original data.

[0020] Preferably, the normalization includes partial least squares (PLS), and the partial least squares is used to establish a regression model, decompose the independent variable and dependent variable data matrix, introduce the dependent variable information into the decomposition process of the independent variable data, so that the principal component of the independent variable is directly associated with the content of the measured component of the sample. The steps of the partial least squares method include: The content matrix of the measured components of n samples Y = (yi)n×1 and the spectral reflectance matrix of n samples at p wavelength points X = (Xij)n×p are decomposed into principal components:

[0021] Where: U and Q are the score matrix and load matrix of the measured component content matrix Y, respectively; T and P are the score matrix and load matrix of the spectral matrix X, respectively; F and E are the measurement errors of the spectral matrix and the measured component content matrix, respectively. A regression model between U and T is established: U = TB + E B=TTU(TTT)-1 Where: E is the random error matrix, B is the regression coefficient matrix. When making predictions, the content of the unknown sample's measured component can be obtained from the unknown sample's spectral matrix according to the formula Ynew = Tnew BQT; Ynew is the new content matrix of the measured component, Tnew is the new spectral matrix. In the modeling process, the number of principal components is determined, and the interactive verification method is used to determine the number of principal components. The optimal number of principal components is determined with the lowest predicted root mean square error value.

[0022] Preferably, the regression model of the partial least squares method is as follows: Y = bX + e Where b is the vector of regression coefficients, e is the model residual, and X is the model variable. The quality of all models is quantitatively compared using the corrected correlation coefficient, corrected root mean square error, and average relative error to determine the detection model.

[0023] A method for predicting shelf life based on near infrared spectroscopy technology, including a comprehensive evaluation method for fruit maturity based on near infrared spectroscopy technology, further comprising: After removing abnormal fruits from the samples, the samples are graded and assigned values ​​according to the number of days required for the fruits to reach full maturity; The samples were divided into a calibration set and a prediction set, and the partial least squares method was used to establish a fruit storage potential detection model. During the detection process, if non-integer situations occurred, the threshold was used for discrimination.

[0024] In summary, the beneficial effects of this application are: 1. The method adopted in this application is to collect near-infrared spectral data of samples in different time periods according to the physiological changes of fruits after picking, so that the sample set contains independent subsets to improve the quality of modeling; 2. This application also introduces a modeling method for post-harvest time factors, and establishes a non-destructive detection model for dynamic quality of post-harvest fruits based on key intrinsic quality indicators such as soluble solids content, hardness, pH value, SSC, DM, etc. of fruits. A comprehensive evaluation index F for fruit maturity during the entire commercial harvest period is established based on the fruit hardness, pH value, SSC, and DM of fruits. The optimal harvest time of fruits is determined by combining the different interval distributions of F values, thereby improving the reliability and accuracy of maturity detection. The shelf life is predicted by combining the different days from fruit harvest to full maturity, and the optimal time for fruit to leave the warehouse is predicted. 3. The present application can provide a fast and accurate non-destructive detection method for fruit maturity, which has the effect of improving the accuracy of timely fruit harvesting, realizing the determination of the optimal fruit harvesting period and the field grading and sorting of fruits, as well as the prediction of the post-harvest shelf life. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a schematic diagram of using a preprocessing algorithm to preprocess a spectrum in this embodiment 1 of the present application; Figure 2 It is a schematic diagram of establishing a regression model using partial least squares method in this embodiment 1 of the present application; Figure 3A is a schematic diagram of F values ​​of samples of different maturity in Example 1 of the present application (I); Figure 3B is a schematic diagram (II) of the F values ​​of samples of different maturity in Example 1 of the present application; Figure 3C is a schematic diagram (III) of the F values ​​of samples of different maturity in Example 1 of the present application; Figure 4 This is a schematic diagram of the second embodiment of the present application after the samples are classified and assigned values. DETAILED DESCRIPTION

[0026] The present application is further described in detail below in conjunction with the accompanying drawings. Example 1

[0027] The embodiment of the present application discloses a method for comprehensive evaluation of fruit maturity based on near infrared spectroscopy technology, comprising the following steps: S1: Under room temperature conditions, preheat the near-infrared spectrometer for 30 minutes and then set the acquisition parameters. The acquisition wavelength range of the near-infrared spectrometer is 600-1100nm, the integration time is 100ms, and the number of scans is 50 times.

[0028] S2: Select several fruit samples, scan the equatorial part of the fruit samples (referring to the area with the largest diameter of the cross section of the fruit samples) respectively, and output the average value of 50 spectral values ​​of each point scanned 50 times as the spectral value of the point.

[0029] S3: Collect the absorbance values ​​of the measuring sites of the fruit samples from picking to ripening, select the original absorbance values ​​of the fruit samples at different time after picking, calculate the average absorbance of the fruit samples at different wavelengths, establish a coordinate system through GrapHpad8 software (a scientific research data analysis and drawing software), and draw a graph of the average absorbance changes of the fruit samples at different times after picking.

[0030] S4: Perform an outlier test on the spectral data of the fruit samples obtained by measurement. After removing the singular points, use the sample algorithm (i.e., the Kennard-Stone algorithm, an algorithm for sample selection) to divide the measured data of the hardness, pH value, SSC, and DM quality of the fruit samples into a calibration set and a prediction set. Use the calibration set data to train the model, use the prediction set data to test the performance of the model, and use the comprehensive set to display the distribution characteristics of the sample set data.

[0031] S5: By comparing the original spectra (see Figure 1 -A) using multivariate scatter correction (see Figure 1 -B), standard normal transformation (see Figure 1 -C), SG convolution derivative (see Figure 1 -D), SG convolution smoothing (see Figure 1 -E), vector normalization (see Figure 1 -F), maximum-minimum normalization (see Figure 1 -G), and a combination of SG convolution smoothing and SG convolution derivatives (see Figure 1 -H), SG convolution smoothing and multivariate scatter correction and SG convolution derivative combination (see Figure 1 -I), multivariate scatter correction and combination of maximum-minimum normalization and SG convolution derivative (see Figure 1 -J) Preprocessing algorithm to remove noise from the spectrum.

[0032] S6: SPSS Statistics 26 (data analysis technology) was used to conduct Pearson correlation analysis on the firmness, pH value, SSC, DM, and actual sampling time of fruit samples to analyze the correlation between the four indicators and the maturity of fruit samples; The principal component analysis (PAC) method was used to analyze the distribution and contribution rate of the principal component eigenvalues ​​of the four indicators of the three varieties of fruit samples; the regression method was used to calculate the score coefficient matrix of the components, and the principal component expression was obtained to construct the comprehensive evaluation index F of the maturity of the three varieties of fruit samples. The principal component eigenvalue distribution and contribution rate of the indicators of the fruit samples of different varieties were analyzed, and the score coefficient matrix of the components was calculated to obtain the principal component expression, and the comprehensive evaluation index F of the maturity of the fruit samples of different varieties was constructed: F=a*Zx1+b*Zx2+c*Zx3+d*Zx4; Wherein, Zx1, Zx2, Zx3, and Zx4 are the hardness value, pH value, SSC value, and DM value after standardized treatment, respectively; a, b, c, and d represent different fruit varieties, respectively.

[0033] S7: The maturity of fruits that have just reached the commercial harvest standard is defined as 65% ripe, and then one maturity is added every fixed number of days, recorded as 70%, 75%, 80%, 85%, and 90% ripe; the F value of each sample is calculated using the constructed comprehensive evaluation index formula for fruit maturity, and a box plot is used to conduct a comprehensive evaluation of the fruit maturity of the F values ​​of samples with different maturity levels, see the attached manual Figure 3A , Figure 3B and Figure 3C .

[0034] Furthermore, the preprocessing algorithm includes multivariate scattering correction, which assumes that there is a linear relationship between the change in the spectrum and the content of the measured component of the sample. It can effectively eliminate the scattering effect caused by uneven sample particle size distribution and different particle sizes, and enhance the spectral information related to the measured component.

[0035] The multivariate scatter correction includes the following calculation formula: Where: X represents the spectral data matrix of the modeling set samples, Xi is the spectral data of the i-th sample, ai and bi are the slope and intercept of the linear regression of the i-th spectrum Xi and the average spectrum X.

[0036] Furthermore, the preprocessing algorithm includes the standard normal transformation, which can also be used to correct the spectral error caused by scattering between samples. The algorithm assumes that the absorbance values ​​of each wavelength point in each spectrum should be distributed in a certain way, and then corrects each spectrum. The standard normal transformation includes the following calculation formula:

[0037] Where: X represents the spectral data matrix of the modeling set samples, Xi is the spectral data of the i-th sample, and X is the average spectrum.

[0038] Furthermore, the preprocessing algorithm includes SG convolution derivative, and the SG convolution derivative includes the following calculation formula:

[0039] In the formula: dm / dxm represents the operator of the m-th order derivative; y(x) represents the value of the original spectral signal (or data point) at the position; ∑ represents summation, which is used to perform weighted summation on the data in the window; k represents the index in the window; N represents the window size, that is, the length of the filter; m represents the order of the derivative; Ck(m) represents the convolution coefficient, which represents the weighted coefficient calculated after the polynomial fitting, and these coefficients are determined according to the window size N and the derivative order m; y(xk) represents the value of the original signal at point xk, where xk is a neighborhood point centered on the current point x.

[0040] Furthermore, the preprocessing algorithm includes convolution smoothing, which is also called window moving polynomial least squares fitting smoothing method, which can be used to eliminate high-frequency random errors. Convolution smoothing includes the following calculation formula:

[0041] Where: After the window width is determined, the convolution weight ωk is obtained by looking up the table to obtain the corresponding ωk.

[0042] Furthermore, the preprocessing algorithm includes normalization, which is divided into area normalization, maximum normalization and vector normalization. Vector normalization is commonly used in spectral analysis, which can correct the spectral changes caused by small optical path differences. Normalization includes the following calculation formula:

[0043] Where: , n is the number of wavelength points, k=1, 2, ..., m.

[0044] Furthermore, the preprocessing algorithm includes maximum-minimum normalization, and the maximum-minimum normalization includes the following calculation formula:

[0045] Where: X is the original data value, which represents a single data point that needs to be normalized; Xnorm is the normalized data value, that is, the value after maximum-minimum normalization processing; Xmin is the minimum value in the data set, that is, the minimum value in the original data; Xmax is the maximum value in the data set, that is, the maximum value in the original data.

[0046] In addition, normalization includes partial least squares (PLS), which is used to establish a regression model. In PLS analysis, the independent variable and dependent variable data matrices are decomposed at the same time, and the dependent variable information is introduced into the decomposition process of the independent variable data, so that the principal component of the independent variable is directly associated with the content of the measured component of the sample. The steps of partial least squares include: First, the content matrix of the measured components of n samples Y = (yi)n×1 and the spectral reflectance matrix of n samples at p wavelength points X = (Xij)n×p are decomposed into principal components:

[0047] Where: U and Q are the score matrix and load matrix of the measured component content matrix Y, respectively; T and P are the score matrix and load matrix of the spectral matrix X, respectively; F and E are the measurement errors of the spectral matrix and the measured component content matrix, respectively. A regression model between U and T is established: U = TB + E B=TTU(TTT)-1 Where: E is the random error matrix, B is the regression coefficient matrix. When making predictions, the content of the unknown sample's measured component can be obtained from the unknown sample's spectral matrix according to the formula Ynew = Tnew BQT; Ynew is the new content matrix of the measured component, Tnew is the new spectral matrix. In the modeling process, the number of principal components is determined, and the interactive verification method is used to determine the number of principal components. The optimal number of principal components is determined with the lowest predicted root mean square error value.

[0048] Use partial least squares method to build models respectively. The regression model of partial least squares method is as follows: Y = bX + e Where b is the vector of regression coefficients, e is the model residual, and X is the model variable. The quality of all models is quantitatively compared using the corrected correlation coefficient, corrected root mean square error, and average relative error to determine the detection model. Example 2

[0049] A shelf life prediction method based on near infrared spectroscopy technology, comprising: After removing abnormal fruits from the samples, the samples were divided into three levels and assigned values ​​according to the shelf life (the number of days required for the fruit to reach full maturity). There were 75 samples that reached full maturity in 5 to 7 days and were assigned a value of 1; 71 samples that reached full maturity in 2 to 4 days and were assigned a value of 2; and 45 fully ripe samples were assigned a value of 3.

[0050] The samples were divided into a calibration set and a prediction set, and the partial least squares method was used to establish a fruit storage potential detection model. During the detection process, when non-integer values ​​appeared, the threshold (the integer value closest to the actual value) was used for discrimination.

[0051] Furthermore, the discrimination rules of the fruit storage potential detection model are as follows: when the predicted value is less than 1.5, it is judged to be fully ripe in 5 to 7 days; when the predicted value is greater than or equal to 1.5 and less than 2.5, it is judged to be a fully ripe sample in 2 to 4 days; when the predicted value is greater than or equal to 2.5, it is judged to be a fully ripe sample.

[0052] The method of this application introduces a time factor to establish a quality and near-infrared spectral change model for the entire biological process of non-storage refrigerated fruits from picking to full maturity under environmental conditions during the actual production process. The optimal harvesting time is accurately determined through a comprehensive evaluation model for fruit maturity to ensure fruit quality and reduce losses. The dynamic intrinsic quality non-destructive detection model can realize real-time non-destructive monitoring of the intrinsic quality of fruits during the growth period and post-harvest period, making it easy to promptly remove fruits with lower intrinsic quality and sell or store fruits with higher comprehensive quality, thereby increasing the commodity rate and commodity value of fruits, and achieving timely post-harvest and quality grading of fruits.

[0053] By establishing a fruit shelf life prediction model, fruit merchants can determine the shelf life of fruits and choose appropriate distribution channels and sales time to ensure that consumers receive the best quality fruits. Fruits with a longer shelf life can choose low-cost logistics transportation or send the fruits to the farthest market in the country; fruits with a shorter shelf life can be sent to inland areas with convenient transportation and fast logistics; fruits that are about to end their shelf life can be transported to local fruit markets for sale, thereby achieving accurate distribution of fruits through distribution channels.

[0054] This method establishes a dynamic non-destructive detection model for the core quality of fruits in this period, which can quickly and accurately collect data information on fruit hardness, pH value, SSC and DM; and transmits it to the cloud platform via Bluetooth, providing producers with real-time monitoring, prediction and early warning services for fruit development, thereby improving the level of intelligent management of fruit orchards.

[0055] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A fruit maturity comprehensive evaluation method based on near infrared spectroscopy technology, characterized in that: The following steps are involved: S1: Under room temperature conditions, the near-infrared spectrometer was preheated and stabilized, and the acquisition parameters were set. The acquisition wavelength range of the near-infrared spectrometer was 600-1100nm, and the number of scans was 50 times; S2: Select several fruits and scan them respectively, and output the average value of 50 spectral values ​​of each point scanned 50 times as the spectral value of the point; S3: Collect the absorbance values ​​of the measuring points of the fruits from picking to ripening, select the original absorbance values ​​of the fruits at different time after picking, calculate the average absorbance of the fruits at different wavelengths, establish a coordinate system, and draw a graph of the average absorbance change at different time after picking the fruits; S4: Perform an outlier test on the fruit spectral data obtained by the measurement. After removing the singular points, use the sample algorithm to divide the measured data of fruit hardness, pH value, SSC, and DM quality into a calibration set and a prediction set. Use the calibration set data to train the model, use the prediction set data to test the model performance, and display the distribution characteristics of the sample set data through the comprehensive set. S5: The original spectrum is preprocessed by using multivariate scattering correction, standard normal transformation, SG convolution derivative, SG convolution smoothing, vector normalization, maximum-minimum normalization, and a combination of SG convolution smoothing and SG convolution derivative, SG convolution smoothing and multivariate scattering correction and SG convolution derivative, multivariate scattering correction and maximum-minimum normalization and SG convolution derivative preprocessing algorithms to remove noise from the spectrum; S6: Analyze the distribution and contribution rate of the principal component eigenvalues ​​of the indicators of different varieties of fruits, calculate the score coefficient matrix of the components, obtain the principal component expression, and construct the comprehensive evaluation index F of the maturity of different varieties of fruits: F=a*Zx1+b*Zx2+c*Zx3+d*Zx4 In the formula, Zx1, Zx2, Zx3, and Zx4 are the hardness value, pH value, SSC value, and DM value after standardization, respectively; a, b, c, and d represent different fruit varieties; S7: The constructed formula for comprehensive evaluation of fruit maturity was used to calculate the F value of each variety, and a box plot was used to conduct a comprehensive evaluation of fruit maturity using the F values ​​of samples with different maturity levels.

2. A fruit maturity comprehensive evaluation method based on near infrared spectroscopy technology according to claim 1, characterized in that: The preprocessing algorithm includes multivariate scatter correction, which includes the following calculation formula: Where: X represents the spectral data matrix of the modeling set samples, Xi is the spectral data of the i-th sample, ai and bi are the slope and intercept of the linear regression of the i-th spectrum Xi and the average spectrum X.

3. A fruit maturity comprehensive evaluation method based on near infrared spectroscopy technology according to claim 1, characterized in that: The preprocessing algorithm includes a standard normal transformation, which includes the following calculation formula: Where: X represents the spectral data matrix of the modeling set samples, Xi is the spectral data of the i-th sample, and X is the average spectrum.

4. A fruit maturity comprehensive evaluation method based on near infrared spectroscopy technology according to claim 1, characterized in that: The preprocessing algorithm includes SG convolution derivative, which includes the following calculation formula: In the formula: dm / dxm represents the operator of the m-th order derivative; y(x) represents the value of the original spectral signal at the position; ∑ represents summation, which is used to perform weighted summation on the data in the window; k represents the index in the window; N represents the window size, that is, the length of the filter; m represents the order of the derivative; Ck(m) represents the convolution coefficient, which represents the weighted coefficient calculated after the polynomial fitting, and these coefficients are determined according to the window size N and the derivative order m; y(xk) represents the value of the original signal at point xk, where xk is a neighborhood point centered on the current point x.

5. A fruit maturity comprehensive evaluation method based on near infrared spectroscopy technology according to claim 1, characterized in that: The preprocessing algorithm includes convolution smoothing, and convolution smoothing includes the following calculation formula: Where: After the window width is determined, the convolution weight ωk is obtained by looking up the table to obtain the corresponding ωk.

6. A fruit maturity comprehensive evaluation method based on near infrared spectroscopy technology according to claim 1, characterized in that: The preprocessing algorithm includes normalization, and the normalization includes the following calculation formula: Where: , n is the number of wavelength points, k=1, 2, ..., m.

7. A fruit maturity comprehensive evaluation method based on near infrared spectroscopy technology according to claim 1, characterized in that: The preprocessing algorithm includes maximum-minimum normalization, which includes the following calculation formula: Where: X is the original data value, which represents a single data point that needs to be normalized; Xnorm is the normalized data value, that is, the value after maximum-minimum normalization processing; Xmin is the minimum value in the data set, that is, the minimum value in the original data; Xmax is the maximum value in the data set, that is, the maximum value in the original data.

8. A fruit maturity comprehensive evaluation method based on near infrared spectroscopy technology according to claim 6, characterized in that: The normalization includes partial least squares (PLS), which is used to establish a regression model, decompose the independent variable and dependent variable data matrix, introduce the dependent variable information into the decomposition process of the independent variable data, so that the main component of the independent variable is directly associated with the content of the measured component of the sample. The steps of partial least squares include: The content matrix of the measured components of n samples Y = (yi)n×1 and the spectral reflectance matrix of n samples at p wavelength points X = (Xij)n×p are decomposed into principal components: Where: U and Q are the score matrix and load matrix of the measured component content matrix Y, respectively; T and P are the score matrix and load matrix of the spectral matrix X, respectively; F and E are the measurement errors of the spectral matrix and the measured component content matrix, respectively. A regression model between U and T is established: U = TB + E B=TTU(TTT)-1 Where: E is the random error matrix, B is the regression coefficient matrix. When making predictions, the content of the unknown sample's measured component can be obtained from the unknown sample's spectral matrix according to the formula Ynew = Tnew BQT; Ynew is the new content matrix of the measured component, Tnew is the new spectral matrix. In the modeling process, the number of principal components is determined, and the interactive verification method is used to determine the number of principal components. The optimal number of principal components is determined with the lowest predicted root mean square error value.

9. A fruit maturity comprehensive evaluation method based on near infrared spectroscopy technology according to claim 6, characterized in that: The regression model of partial least squares method is as follows: Y = bX + e Where b is the vector of regression coefficients, e is the model residual, and X is the model variable. The quality of all models is quantitatively compared using the corrected correlation coefficient, corrected root mean square error, and average relative error to determine the detection model.

10. A shelf life prediction method based on near infrared spectroscopy technology, characterized in that: The method for comprehensive evaluation of fruit maturity based on near infrared spectroscopy technology comprises any one of claims 1 to 9, further comprising: After removing abnormal fruits from the samples, the samples are graded and assigned values ​​according to the number of days required for the fruits to reach full maturity; The samples were divided into a calibration set and a prediction set, and the partial least squares method was used to establish a fruit storage potential detection model. During the detection process, if non-integer situations occurred, the threshold was used for discrimination.

Images