Juicy peach maturity discrimination method based on near infrared spectrum

Abstract

The invention relates to a juicy peach maturity discrimination method based on a near infrared spectrum, and belongs to the field of rapid nondestructive testing. The method comprises the following steps: firstly, setting parameters of a near-infrared analyzer for the maturity of juicy peaches through a near-infrared spectrum and an SVM (Support Vector Machine) algorithm; secondly, acquiring the spectrum of the juicy peaches, and preprocessing the spectrum; and finally, taking the preprocessed data as input of an SVM algorithm, determining an optimal parameter combination of a penalty coefficient cost and a kernel function coefficient gamma through a cross validation and grid search method, and establishing a juicy peach maturity identification model.

Description

Technical Field

[0001] The present application relates to the field of rapid non-destructive detection of peach maturity, and in particular to a method for identifying peach maturity based on near infrared spectroscopy. Background Art

[0002] Peaches are a drupe plant of the genus Prunus in the Rosaceae family. They are native to China and are cultivated all over the world. Peaches are popular among consumers for their rich juice, soft flesh, sweetness and fragrance. They generally mature in mid-to-late July. The maturity of the fruit during the harvest period directly determines the storage time and final edible quality of the fruit. Harvesting too early or too late will result in high fruit loss rate and poor quality. Accurately selecting the best time to harvest the fruit is the key to reducing fruit loss and improving fruit quality.

[0003] The internal quality of the fruit is an important part of the fruit quality evaluation, among which the soluble solids content (SSC) is the main factor affecting the taste of the fruit and is closely related to the degree of maturity of the fruit. However, the collection of SSC is mainly based on destructive testing, which is not very applicable in the field. At present, the judgment of peach harvest period is still mainly based on human appearance judgment, which is greatly affected by subjectivity and cannot adapt to the development direction of smart orchards. In addition, due to the high requirements of the application environment and high cost of acoustic property detection technology, electrical property detection technology and electronic nose technology, they are mostly used for post-harvest fruit grading under closed conditions, and cannot achieve rapid maturity detection of peaches in the field.

[0004] Near infrared spectroscopy technology has the characteristics of simultaneous detection of multiple components, fast collection speed, low cost, high accuracy, no pre-treatment, non-destructiveness and pollution. Its principle is that after near infrared light irradiates the fruit, the spectral characteristics are related to the structure, texture and composition of the fruit, and can be used for quantitative and qualitative analysis of the chemical and physical quality of the fruit. Among them, diffuse reflection can detect the sugar and acidity of peaches, pears, apples and other fruits with thin skin.

[0005] In recent years, many domestic experts have used near-infrared spectroscopy to analyze the sugar content of pears, watermelons, and peaches. For example, Fu Xiaping et al. used near-infrared spectroscopy to detect the firmness of "Xueqing" pears at different maturity stages, and the predicted root mean square error was 4.23N, which confirmed the feasibility of using near-infrared technology to detect fruit firmness; Han Donghai et al. used near-infrared to conduct rapid non-destructive testing of the quality of small Jingxiu watermelons during their growth process. From the PLS modeling results, it can be seen that both tissue moisture content and soluble solids content can be obtained at the top and equatorial parts of the melon. Pan Leiqing et al. used Fourier near-infrared spectroscopy technology, multivariate scattering correction, first-order derivative and convolution smoothing methods to pre-process the original spectra of the samples, and combined the partial least squares method to establish a mathematical model to predict the sugar content of peaches during the shelf life, and obtained relatively good prediction results.

[0006] However, most of the existing research focuses on the detection of internal components of the fruit, re-verifying the maturity through component detection, or predicting the shelf life of the fruit after harvest. There are relatively few studies on the discrimination of the maturity of different peaches. For example, Carlomagn et al. (2004) used the sugar content and hardness of peaches to judge the maturity. They used the near-infrared transmission spectrum in the 730-900nm region, pre-processed the spectrum using a noise reduction method based on packet wavelet transform, and then used the minimum distance classifier to estimate the classification of each experimental data. The recognition accuracy of this method for peaches only reached 82.5%, so it is difficult to meet market demand. Summary of the invention

[0007] In view of the above problems, the present invention discloses a method for determining the maturity of peaches based on near infrared spectroscopy, which uses near infrared spectroscopy technology to obtain a peach maturity degree at 4000-10000 cm -1 The wave number range is used to detect peaches at different harvesting stages, providing a basis for determining the harvesting stage of peach fruits.

[0008] Specifically, this application is implemented through the following technical solutions:

[0009] A method for identifying peach maturity based on near infrared spectroscopy, the specific steps of which are as follows:

[0010] 1) Thermo Antaris II Fourier transform near-infrared analyzer was used to collect near-infrared spectra of peaches of different maturity using diffuse reflectance method;

[0011] 1.1) Set the parameters of the near infrared analyzer and use diffuse reflection method;

[0012] 1.2) Using the air inside the instrument as the background, the spectrum scan wave number is 10000~4000cm -1 , scanning interval 3.856cm -1 , each spectrum has 1557 points, 32 scans, and a gain of 4x;

[0013] 1.3) Collect near-infrared spectra of peach samples with the air inside the instrument as the background;

[0014] 2) Preprocess the obtained spectral data.

[0015] 2.1) Establish a near-infrared spectral library of peaches at different maturity levels;

[0016] 2.2) The SNV method was used to preprocess the near-infrared spectral data, and PCA was used to preprocess the sample data to eliminate the correlation and noise of the data;

[0017] 2.3) Perform principal component analysis on the obtained spectral data, and use the scores after principal component analysis as characteristic parameters;

[0018] 3) Using RESULT-Integration software and adopting radial basis function as the kernel function of support vector machine, a prediction model for peaches of different maturity was established:

[0019] 3.1) The specific expression of the radial basis function is:

[0020] K(X;,Xj)=exp(-γ||X;-Xj||)2; (1)

[0021] Among them, γ=1 / 2σ2, which is a free parameter;

[0022] 3.2) The optimal parameter combination of penalty coefficient cost and kernel function coefficient gamma was determined by cross-validation and grid search methods, and a peach maturity discrimination model based on near infrared spectroscopy was established;

[0023] 3.3) Use the model to make predictions under the optimal parameters, establish a peach maturity discrimination model and calculate the accuracy of the model.

[0024] Furthermore, in the above step 2.2), the near-infrared spectrum is transformed by standard normal variables (SNV), which can reduce the impact of errors. In all spectra, the absorbance values ​​at each wavelength point must have a certain coverage range. The spectrum is pre-processed, and the average value n of the spectrum is subtracted from the original spectrum, and then divided by the standard deviation of the spectrum data.

[0025] Difference γ, in order to obtain spectral data without error effect, the formula is:

[0026]

[0027] This application collects peach samples from different degrees of maturity, uses a near-infrared analyzer to collect the spectrum of peaches, performs SNV preprocessing on the spectrum, and uses PCA to preprocess the sample data to achieve the purpose of eliminating the correlation and noise of the data, and extracts the principal component containing the sample data information, reduces the dimension of the sample space, and then uses SVM to train the samples to obtain the optimal hyperplane to divide the samples. Then the preprocessed spectral data is used as the input of the SVM algorithm, and the optimal combination of the penalty coefficient cost and the kernel function coefficient gamma is determined to establish a peach maturity discrimination model, and finally the model is used for prediction and calculation of the model accuracy. This method can realize rapid non-destructive detection of peach maturity through near-infrared spectroscopy technology combined with the SVM algorithm, providing a basis for intelligent and precise picking of peaches. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 These are photos of peaches of different ripeness in Example 1;

[0029] Figure 2 This is the original near-infrared spectrum of peach;

[0030] Figure 3 Near infrared spectra of the samples after pretreatment: (a) SNV; (b) MSC; (c) D1st; (d) SGS;

[0031] Figure 4 The middle is the score diagram of the three maturity principal components;

[0032] Figure 5 This is a flow chart of the judgment method for this application. DETAILED DESCRIPTION

[0033] Example 1

[0034] (1) Collect peach samples

[0035] Xiahui No. 8 used in this embodiment is a medium-ripe honey peach 'Zhaohui' and 'Xiahui' developed by the Horticulture Institute of Jiangsu Academy of Agricultural Sciences in 2001.

[0036] The medium-ripe nectarine 'Ruiguang No. 18' was artificially hybridized with the female parent and the male parent, and was introduced by the Horticulture Institute of Sichuan Academy of Agricultural Sciences in 2009. It was identified and named as a late-ripening peach variety by the Jiangsu Provincial Crop Variety Review and Evaluation Committee in November 2012 (Yu Mingliang et al., 2014). The sample size of this embodiment is 282 peach fruits, which are transported back to the laboratory within 24 hours after picking. After confirmation by professional and technical personnel, samples with mechanical damage on the surface and abnormal appearance are removed. During the subsequent physical and chemical testing, fruits with internal lesions are also removed. Subsequently, according to manual experience and combined with the color, size and hardness of the samples, they are manually divided into three degrees of maturity, namely 75% mature, 85% mature and 85% mature, and then numbered as follows Figure 1 shown.

[0037] (2) Set instrument parameters

[0038] Thermo Antaris II Fourier transform near infrared analyzer was used to collect the near infrared spectrum of peaches by diffuse reflectance. The air inside the instrument was used as the background, and the spectrum was scanned at a wave number of 10000-4000cm -1 , scanning interval 3.856cm -1 , each spectrum has 1557 points, 32 scans, and a gain of 4x;

[0039] (3) Collecting near-infrared spectra

[0040] Six positions 60° apart were selected on the equator of the peach fruit to measure the diffuse reflectance spectrum, and the average value of the six positions was taken as the spectrum measurement value of the fruit. Figure 2 This is the original spectrum of the peach sample;

[0041] (4) Spectral preprocessing

[0042] Due to the differences in the internal chemical composition and tissue structure of peaches picked at different times, the collected spectra have certain similarities. In addition, factors such as the physical properties of the peach fruit surface, changes in the internal structure, and the detection environment conditions often cause spectral baseline shifts and high-frequency noise, affecting the test results and model performance. Therefore, appropriate spectral preprocessing methods are needed to eliminate interference, noise, and useless information (Chu Xiaoli, 2016) in order to improve the accuracy and robustness of the model.

[0043] This embodiment mainly adopts the following four preprocessing methods for comparison and selection of the optimal processing method. The results show that the SNV result is best when converted to a standard normal variable.

[0044] A: Transformation of standard normal variate (SNV)

[0045] The inhomogeneity of the sample to be tested will cause scattering of light when it passes through, penetrates or is reflected from the sample. The scattering of light will cause errors in the diffuse reflectance spectrum. SNV can reduce the impact of errors. In all spectra, the absorbance value of each wavelength point must have a certain coverage range. By pre-processing the spectrum, the average value n of the spectrum is subtracted from the original spectrum, and then divided by the standard deviation γ of the spectrum data, in order to obtain spectral data without error effects, that is:

[0046]

[0047] Wherein, i=1, 2, ..., n; n is the number of samples.

[0048] B: Multiplicative scatter correction (MSC)

[0049] MSC and SNV have similar effects, both of which can eliminate the scattering effect (Li Zhe, 2016). The reflectance absorbance value at the sample wavelength point shows an approximate linear relationship with its average absorbance spectrum. The linear regression of the spectrum set can obtain the intercept and slope of the straight line, which can correct each spectrum collected. The reflectance effect and uniformity of the sample are reflected by the intercept and slope respectively. The calculation formula is as follows:

[0050]

[0051]

[0052] Where (3) represents the average spectrum, (4) represents the correction effect of MSC, n is the number of samples, C represents the spectral matrix of the sample set, and C i is the spectrum of the i-th sample, x i and i is the ith spectrum C i With the average spectrum The slope and intercept of the linear regression.

[0053] C:Derivative(Der)

[0054] The first-order derivative represents the slope of a certain point on the spectral curve of the sample, which has the effect of clarifying the absorption peak in the original spectrum; the second-order derivative represents the change in the slope of the corresponding point, which can accurately determine the center position of the absorption peak in the original spectrum (Weng Shifu, 2010; Zheng Yongmei et al., 2004). The derivative has the function of eliminating the interference of baseline and other background and amplifying or separating overlapping information, but it will generate some noise information. Therefore, when using this method, an appropriate differential width should be selected. If the width is too small or too large, it will lead to large noise and loss of a lot of detail information. Some researchers will combine the derivative with the smoothing denoising method (Li Ming, 2018). For a scattered spectrum x k , the formula for calculating the first-order derivative and second-order derivative spectrum at wavelength k and difference width g is as follows:

[0055]

[0056]

[0057] D: Convolution smoothing (Savitky-golay smoot, SGS)

[0058] Smoothing is a simple algorithm that can reduce random noise and improve the signal-to-noise ratio. Among them, the Savitzky-Golay convolution smoothing method is currently the most widely used method. It achieves the purpose of smoothing the original spectrum by taking several points before and after the smoothing point for polynomial least squares fitting (Savitzky, 1964). When using this method, it is necessary to select an appropriate moving window width. Too large or too small a width will affect the spectral resolution effect, resulting in a decrease in the prediction ability of the model.

[0059] Table 1 Results of different spectral preprocessing methods

[0060]

[0061] Note: SNV: Transformation of standard normal variate; MSC: Mutiplicative scatter correction; SG1: Savitzky-Golay first derivative; SG2: Savitzky-Golay second derivative; SGS: Savitzky-Golay smoot.

[0062] The spectra obtained by the above four methods are as follows Figure 3 As shown in Table 1, Figure 3 In (a)-(d), SNV, MSC, D1 st , the spectra after SGS processing, it can be seen that (a) the result shows that the SNV result is best when converted to a standard normal variable.

[0063] (5) Establishing an origin identification model

[0064] The SVM classification method based on PCA (Principal Component Analysis) first uses PCA to preprocess the sample data to eliminate the correlation and noise of the data, extract the principal component containing the sample data information, reduce the dimension of the sample space, and then use SVM to train the samples to obtain the optimal hyperplane to divide the samples.

[0065] Figure 4 The middle is the score diagram of the three maturity principal components. There is a clear distinction between peaches of different maturity, but there is also a small overlap. The cumulative contribution rate of the first twelve principal components is 99.56%.

[0066] The spectral data after PCA preprocessing was used as the input of the SVM algorithm to establish a peach maturity recognition model. The radial basis kernel function was selected as the kernel function of SVM. The optimal parameter combination of the penalty coefficient cost and the kernel function coefficient gamma was determined through cross-validation and grid search methods, and the accuracy of the model was calculated. Figure 4 As shown in Figure 1, the optimal parameters are cost = 1692 and gamma = 1.2207 × 10-4. After training, 282 support vectors are obtained, with 97, 75, and 110 support vectors for 75%, 80%, and 85% maturity, respectively.

[0067] 6) Calculate accuracy

[0068] The accuracy of SVM algorithm in distinguishing the maturity of peaches was 90.2%.

[0069] Figure 5 The present invention provides a flow chart of a method for determining the maturity of peaches based on near infrared spectroscopy.

Claims

1. A method for identifying peach maturity based on near infrared spectroscopy, characterized in that: The specific steps are as follows: 1) Collect near-infrared spectra of peaches at different maturity levels; 2) preprocessing the spectral data obtained in step 1); 2.1) Establish a near-infrared spectral library of peaches at different maturity levels; 2.2) The SNV method was used to preprocess the near-infrared spectral data, and PCA was used to preprocess the sample data; 2.3) Perform principal component analysis on the obtained spectral data, and use the scores after principal component analysis as characteristic parameters; 3) Using radial basis function as the kernel function of support vector machine, a prediction model for peaches of different maturity is established: 3.1) The specific expression of the radial basis function is: K(X;,Xj)=exp(-γ||X;-Xj||)2; (1) Among them, γ=1 / 2σ2, which is a free parameter; 3.2) The optimal parameter combination of penalty coefficient cost and kernel function coefficient gamma was determined by cross-validation and grid search methods, and a peach maturity discrimination model based on near infrared spectroscopy was established; 3.3) Use the model to make predictions under the optimal parameters, establish a peach maturity discrimination model and calculate the accuracy of the model.

2. The peach maturity identification method based on near infrared spectroscopy according to claim 1, characterized in that: In step 2.2), the SNV method is used to preprocess the near-infrared spectral data, which means that in all spectra, the average value n of the spectrum is subtracted from the original spectrum, and then divided by the standard deviation γ of the spectral data to obtain spectral data without error effect, and the formula is:

3. The peach maturity identification method based on near infrared spectroscopy according to claim 1, characterized in that: Step 1) collecting near infrared spectra of peaches of different maturity levels refers to: 1.1) Set the parameters of the near infrared analyzer and use diffuse reflection method; 1.2) Using the air inside the instrument as the background, the spectrum scan wave number is 10000~4000cm -1 , scanning interval 3.856cm -1 , each spectrum has 1557 points, 32 scans, and a gain of 4x; 1.3) Using the air inside the instrument as the background, collect the near-infrared spectrum of the peach sample.

Images