Rapid Identification of the Quality of Eucommiae Cortex During the Whole Process of Salt Processing Based on Chromaticity and Near‐Infrared Spectroscopy

ABSTRACT

Background

As one of the traditional Chinese medicines, Eucommiae Cortex (EC) was often prepared with salt in the clinic to enhance its efficacy. However, its processing degree had always been judged by “experience,” and the quality of EC decoction pieces after processing could not be accurately controlled.

Objectives

This study aims to establish a qualitative model and a quantitative model in the salting process of EC, which could quickly determine its processing degree and internal quality.

Materials and Methods

First, determine the chromaticity values and the contents of geniposidic acid, chlorogenic acid, geniposide, pinoresinol diglucoside, and hyperoside in 123 batches of differently processed EC, and analyze their correlations. The results showed that the chromaticity values had different degrees of correlation with the components. The orthogonal partial least squares discriminant analysis (OPLS‐DA) method was used to establish the chromaticity qualitative discriminant model and the near‐infrared (NIR) qualitative discriminant model. The two qualitative models well distinguished the samples of four categories: raw EC, processing less, salt eucommia, processing too much, and the model verification results were good. Finally, an NIR quantitative model was established by using the partial least squares (PLS) method, combined with chroma value, content, and NIR spectroscopy.

Results

There were different degrees of correlation between the chromaticity values and the components. The qualitative and quantitative discriminant models established were good enough to distinguish the four types of samples: raw eucommia, insufficient processing, salt eucommia, and excessive processing, and the model verification results were good.

Conclusions

The intrinsic characteristic component content and chroma value of EC in the salting process could be predicted simultaneously by the model, which provided a rapid determination method for the quality control of EC in the processing process.

Short abstract

Chromaticity values and six characteristic component contents of Eucommiae Cortex during salt processing were determined and correlated analyzed. Combined with near‐infrared spectroscopy, OPLS‐DA qualitative models and PLS quantitative model were established to rapidly distinguish processing degrees and simultaneously predict internal quality of salt‐processed Eucommiae Cortex.

1. Introduction

EC had a tonic effect on the liver and kidney, and it was a good tonic for them. At the same time, it also had the effects of strengthening people's bones and preventing abortion. Raw EC contained more hard rubber material EC gum, which inhibited the dissolution of active ingredients. After high‐temperature heating, hard rubber could be destroyed, which promoted the dissolution of active ingredients and improved the therapeutic effect [1]. Therefore, traditional Chinese medicine was often based on processed products. Salt‐processed EC was made from raw EC with salt water. It was warm and not dry and could reach the lower coke and be introduced into the kidney. In terms of efficacy, EC had an antihypertensive effect. According to relevant research reports, the antihypertensive effect of salt‐processed EC was about twice as strong as that of raw EC. At the same time, modern chemical methods were used to compare the content of water‐soluble extracts of salt‐processed EC and raw EC. The results confirmed that the content of water‐soluble extracts of salt‐processed EC was 8.07% higher than that of raw EC [2]; the content of lead in salt‐processed EC decreased by more than 30%, which reduced the toxic effect of EC, indicating that salt processing could reduce toxicity. The main chemical constituents of EC were lignans, iridoids, phenylpropanoids, etc. [3]. It had a regulatory effect on the cardiovascular system and the nervous system [4, 5]. It could also protect the liver and kidneys [6, 7], reduce blood pressure [8], anti‐aging [9], promote bone cell proliferation [10, 11], anti‐tumor, and so on [12, 13]. In the early stage, through chemical composition analysis and pharmacodynamic analysis, the research group screened out six components, including geniposidic acid, chlorogenic acid, geniposide, genipin, pinoresinol diglucoside, and hyperoside, as the potential material basis for the treatment of kidney‐yang deficiency before and after salt processing of EC [14].

At present, high‐performance liquid chromatography (HPLC) [15, 16] and fingerprint and more sensitive liquid chromatography–mass spectrometry (LC‐MS) are often used to evaluate the quality of EC decoction pieces [17, 18, 19] Although these methods have high accuracy, they are often complicated in sample processing, analysis process, and data processing. The evaluation results cannot be obtained quickly and will cause damage to the samples. In this study, near‐infrared (NIR) spectroscopy combined with machine vision technology was used to control the quality of the whole process of EC salt making. NIR spectroscopy analysis technology had the advantages of fast analysis speed, simple sample processing, convenient operation of the instrument, and high sensitivity of the instrument system, which could monitor and analyze the measured samples in real time. NIR technology was often used in judging the authenticity of Chinese medicinal materials, judging which origin it came from, and analyzing its internal components. Machine vision technology was to use the machine to simulate the human eye and to judge the degree of EC salt by color data, so as to avoid the subjective judgment of human eye observation. At the same time, the correlation between the color and component content of EC samples before and after salt processing was studied, and the change rules of color and internal components before and after salt processing were explored. NIR analysis can provide the chemical composition information of the sample, and machine vision analysis can provide the appearance feature information of the sample. By combining these two methods, it can not only make up for the shortcomings of a single method but also achieve multidimensional evaluation of samples and improve the accuracy and reliability of detection, which provided new ideas for grasping the quality of processing traditional Chinese medicine decoction pieces.

In this study, a great number of samples were collected. Through the previous research group's chemical composition experiments before and after salt processing of EC combined with pharmacodynamic experiments, six characteristic components before and after processing were finally screened out [8]. The content and chromaticity values of the six characteristic components in the samples were determined, and the chromaticity qualitative discriminant model was established. It found a correlation between the content of characteristic components and color before and after salt processing of EC. Finally, the NIR qualitative and quantitative model of the characteristic components of EC salting process was established. It provided a scientific basis for the quality‐level judgment and quality assurance of EC salt preparation.

2. Materials and Methods

2.1. Samples and Reagents

Three batches of raw EC decoction pieces (used for sample preparation during processing), 10 batches of raw EC decoction pieces and their salt products (raw EC: YXS1–YXS10; raw EC: YXS1–YXS10; salt EC: YXY1–YXY10), 35 batches of commercially available EC (S1–S35) and 35 batches of salt EC (Y1–Y35) were purchased from domestic formal Chinese medicine decoction pieces enterprises. The quality met the pharmacopeia standards.

All ECs were carefully identified by Professor Ge Weihong from Zhejiang University of Traditional Chinese Medicine and identified as processed products of EC. (The specific information is shown in Table S1.) Three batches of EC raw pieces were purchased from Zhejiang University of Traditional Chinese Medicine Chinese Herbal Pieces Co. Ltd. and processed according to this preparation process standard from the company (including less processed and over‐processed samples), According to the processing method of salt Eucommia ulmoides in the 2020 edition of Chinese Pharmacopeia, “take EC piece or strip, add salt water (EC per 100 kg, mixed with salt 2 kg), stuffy, placed in a stir‐fried container, heated at 220°C for 7–9 min, until the silk was easy to break, the surface was burnt black, remove and cool.” During the processing, samples were taken at 0, 1.5, 3, 4.5, 6, 7, 8, 9, 10, 11, and 12 min, respectively, with a sampling amount of about 500 g each time. A total of three batches of EC were prepared for the processing of the above methods. The three batches were numbered A0–A10, B0–B10, and C0–C10, respectively. After drying, the samples were powdered, passed through the 355 ± 13 μm sieve, and set aside. The specific information is shown in Table S2. All E. ulmoides Oliv. were tested by the provisions of the Chinese Pharmacopeia, and the appearance and internal quality met the standards.

Geniposidic acid (Lot 111828‐201805, purity 98.1%), chlorogenic acid (Lot 110753‐201817, purity 96.8%), pinoresinol diglucoside (Lot 111537‐20106, purity 91.7%), hyperoside (Lot 111521‐201809, purity 94.9%), and geniposide (Lot 110749‐2019, purity 97.1%) were purchased from the National Institutes for Food and Drug Control (Beijing, China). Genipin (Lot 21041202, purity 99.40%) was purchased from Glip Biotechnology Co. Ltd. (Chengdu, China). Acetonitrile was purchased from Tedia Company (USA). Phosphoric acid was purchased from Zhejiang Hannuo Chemical Technology Co. Ltd. (Zhejiang, China), and methanol was purchased from Guanghua Technology Co. Ltd. (Guangdong, China). Pure water is purchased from Hangzhou Wahaha Group Co. Ltd. (Hangzhou, China).

2.2. Determination of Content

2.2.1. Preparation of Test Solution and Reference Solution

About 2 g of EC sample powder (passing through the 355 ± 13 μm sieve) was taken and accurately weighed. In the conical bottle with a stopper, 50 mL of 50% methanol was accurately added, and the weight was weighed. Ultrasonic treatment (frequency 40 kHz, power 500 W) was performed for 40 min. After cooling, the cooled conical flask and solution were weighed repeatedly, and the weight of volatilization needed to be supplemented with 50% methanol to the weight before ultrasound. Shake well, filter, accurately measure 25 mL of subsequent filtrate in an evaporator, evaporate dry, and constant volume in a 5‐mL volumetric flask with 50% methanol. After centrifugation, the supernatant was taken.

Geniposidic acid, chlorogenic acid, geniposide, genipin, pinoresinol diglucoside, and hyperoside were accurately weighed and placed in an appropriate capacity bottle. 50% methanol was added to prepare a mixed reference stock solution with a concentration of 3207.87, 477.22, 1029.00, 299.19, 1061.89, and 112.93 μg/mL, respectively.

The U3000 high‐performance liquid chromatograph (Thermo Fisher, USA) with diode array detector (DAD) was used for analysis. Agilent ZORBAX‐Extend‐C18 (4.6 × 250 mm, 5 μm) column was used. The mobile phase was 0.1% phosphoric acid solution (A)–methanol (B), gradient elution (0–23 min, 5%–16.5% B; 23–35 min, 16.5%–20% B; 35–40 min, 20%–21.5% B; 40–55 min, 21.5%–24% B; 55–70 min, 24%–27.5% B; 70–85 min, 27.5%–32.5% B; 85–130 min, 32.5%–47% B); the column temperature was 35°C. The flow rate was 1.0 mL/min. The detection wavelength was 230 nm.

The accuracy, repeatability, and stability of the HPLC‐DAD method were verified. A certain amount of reference stock solution was accurately absorbed and diluted with 50% methanol solution. The concentrations of each reference substance were as follows: geniposidic acid: 3207.87, 801.97, 534.65, 400.98, 200.49, and 100.25 μg/mL; chlorogenic acid: 477.22, 119.31, 79.54, 59.65, 29.83, and 14.91 μg/mL; geniposide: 1029.00, 257.25, 171.50, 128.63, 64.31, and 32.16 μg/mL; genipin: 299.19, 74.80, 49.87, 37.40, 18.70, and 9.35 μg/mL; pinoresinol diglucoside:1061.89, 265.47, 176.98, 132.74, 66.37, and 33.18 μg/mL; hyperoside: 112.93, 28.23, 18.82, 14.12, 7.06, and 3.53 μg/mL. Under the same HPLC chromatographic conditions, each concentration of reference solution was injected with 10 μL. The regression equation was obtained by linear regression of peak area (y) to concentration (x, μg/mL). The accuracy, repeatability, and stability of the HPLC‐DAD method were verified.

2.2.2. Sample Determination

The reference solution was accurately absorbed 10 μL, and the test solution was accurately absorbed 10 μL, which was injected into the high‐performance liquid chromatograph, respectively. The peak area data of the chromatogram of each concentration reference were obtained, and then the peak area data of the sample chromatogram were extracted. The standard curve method was used to calculate the content of geniposidic acid, chlorogenic acid, geniposide, genipin, pinoresinol diglucoside, and hyperoside in the sample.

2.2.3. Dynamic Analysis of the Content of Characteristic Components

The study selected samples from the entire salt‐processing procedure of EC, as well as continuous samples from the same batch to analyze the content changes of characteristic components in the processing process and before and after processing.

2.3. Acquisition of Chromaticity Value and Establishment of Chromaticity Qualitative Models

The chromaticity value of the EC samples was collected by CM‐5 spectrophotometer (Konica Minolta, Japan). Detection light source was pulse xenon lamp. The observation angle was 10°. The measurement aperture was 4 mm. The measurement wavelength range was 360–740 nm. The repeatability standard deviation △E was within 0.07, and the measurement mode was SCI reflected light mode. The EC was crushed, passed through the 355 ± 13 μm sieve, and an appropriate amount of powder was tiled at the bottom of the quartz dish. After reaching uniform and seamless, the measurement could be started. A batch of samples needed to be measured three times, and the final data were the mean of three times. It was concluded that the colors represented by L*, a*, and b* were light and shade, red and green, and yellow and blue, respectively. Finally, the total chromaticity value E*ab needed to be calculated, and the formula was $E^{*}\mathit{ab}=\sqrt{L^{*2}+a^{*2}+b^{*2}}$.

Then, the determination method of chromaticity value was tested, and its precision, repeatability, and stability were tested to ensure that the method was stable and feasible. The same analysis method was used to perform six repeated measurements on the same sample powder to evaluate its precision. The same batch of samples was made into six batches of powder in parallel for measurement to evaluate its repeatability. In addition, the same determination method was used to determine the same sample powder at different time points (0, 2, 4, 6, 8, and 10 h) to evaluate its stability.

After obtaining the chromaticity data of all samples, SIMCA14.1 software was used to process the data. OPLS‐DA was selected to establish the model in order to quickly judge the salting degree of EC by color, and the model was verified by substitution test.

2.4. Correlation Analysis Between the Chromaticity Value and Content

The data of chroma value and the content of characteristic components of EC were imported into SPSS25.0 software for bivariate correlation analysis, and the correlation between chroma value and characteristic components of EC was compared.

2.5. Acquisition of NIR Spectra and Establishment of NIR Qualitative Models

2.5.1. Acquisition of NIR Spectra

An Antaris II Fourier transform NIR spectrometer (Thermo Fisher, USA) was used to crush the EC sample (passing through the 355 ± 13 μm sieve), took out the sample cup of the quartz material supporting the instrument, took about 10 g of the sample powder, mixed, and tiled in the sample cup. The built‐in background was used as a reference. At this time, the reference needed to be deducted and the spectrum of the sample were collected. The sampling of EC was carried out by integrating sphere diffuse reflection. The wavenumber range of 4000–10,000 cm⁻¹ was set as the wavenumber range of the spectrum, and 8.0 cm⁻¹ was set as the resolution of the scanning spectrum to ensure that the spectrum of EC could clearly fall into the range. The scanning signal was accumulated for 64 times to ensure that the temperature during the experiment was generally 25°C ± 2°C, and the relative humidity also needed to be maintained at 45%–50%. Each EC needed to be scanned three times, and the final data are the average of three times.

2.5.2. Establishment of NIR Qualitative Models

Using TQ Analyst (V9.8; Thermo Fisher Scientific, Waltham, MA, USA). The spectral data of 123 batches of samples collected in the processing process of EC were quantified and exported by the software, and the data files were integrated and imported into SIMCA14.1 (Umetrics, Sweden) software for processing. OPLS‐DA was selected to establish the model in order to quickly judge the salt degree of EC with NIR spectral data, and finally, the model was verified by permutation test.

2.6. Establishment of NIR Quantitative Models

The required index data were imported into the 8.3 version of the TQ analyzer to establish a model. The model could quantify the index data of EC, so as to quickly judge the changes of the internal components of EC during the salting process, so as to select the best processed products. PLS was used as a modeling method to predict the content of compounds in the sample. Firstly, the pretreatment method was selected, and the correlation coefficient (R), root mean square error of calibration (RMSEC), and root mean square error of prediction (RMSEP) were used as indicators to evaluate the effect of the pretreatment method on model performance. Secondly, the optimal number of factors was selected by RMSEC and RMSEP. Spectral bands need to be intercepted before spectral preprocessing to reduce redundant information. In addition, the number of principal factors determines the degree of simplification of the data and the amount of information retained by the model, and it is also necessary to select the appropriate number of principal factors. Finally, the model was established. After selecting the optimal pretreatment method and the number of factors, the NIR pretreatment spectra of 123 batches of E. ulmoides samples were divided into calibration set and prediction set according to the ratio of 9:1 to construct the model. Ten batches of EC samples were randomly selected as the prediction set, and the remaining batches of EC were all used to establish the model as the calibration set. In order to ensure the accuracy and reliability of the model, the chromaticity data and intrinsic component content of the EC samples in the validation set should be within the range of the calibration set.

3. Results and Discussion

3.1. Results of Content Determination

As demonstrated in Table S3, the recovery test results for geniposidic acid, chlorogenic acid, geniposide, genipin, pinoresinol diglucoside, and hyperoside confirmed the accuracy of the HPLC method, whereas the precision, stability, and repeatability were also satisfactory (Table 1). The HPLC chromatograms of standard substance, EC raw material, and salt EC are shown in Figure 1. The dynamic changes of the content of characteristic components in three batches of EC during the whole process of salt making are shown in the figure. It can be seen from Figure 2 that after the three batches of EC are salted, the content of the six characteristic components generally increased first and then decreased, and the content of each component reached the maximum during the processing period of 7‑9 min [20].

TABLE 1.

Methodological investigation of the results (n = 6).

Components	Standard curve	R 2	Linear range (μg/mL)	Inter‐day precisions (RSD%)	Intra‐day precision (RSD%)	Stability (RSD%)	Repeatability (RSD%)	Recovery
								Mean%	RSD%
Geniposidic acid	y = 0.112x − 8.8025	0.999 2	100.25–3207.78	1.84	1.51	0.98	1.13	99.08	1.22
Chlorogenic acid	y = 0.1638x + 0.328	0.999 8	14.91–477.22	1.45	1.37	0.78	1.24	102.83	1.33
Geniposide	y = 0.123x − 1.9429	0.999 3	32.16–1029.00	0.87	1.68	0.46	0.79	100.48	1.29
Genipin	y = 0.1976x + 0.0583	0.999 7	9.35–299.19	1.75	0.89	1.23	1.62	97.39	2.61
Pinoresinol diglucoside	y = 0.1402x − 3.6076	0.999 4	33.18–1061.88	1.31	0.66	1.51	0.65	97.27	2.10
Hyperoside	y = 0.2333x − 0.0848	0.999 2	3.53–112.93	1.58	1.24	0.81	0.72	103.44	1.01

FIGURE 1.

HPLC chromatograms of mixed standard (A), raw EC (B), and salt‐processed EC (C) (1, geniposidic acid; 2, chlorogenic acid; 3, geniposide; 4, genipin; 5, pinoresinol diglucoside; 6, hyperoside).

FIGURE 2.

Changes in the content of six processing characteristic components in the whole process of salt processing of EC in three batches. Geniposidic acid (A), chlorogenic acid (B), geniposide (C), genipin (D), pinoresinol diglucoside (E), and hyperoside (F).

3.2. Chromaticity Value and Chromaticity Qualitative Models

The chromaticity values of EC samples are shown in Table S4. The colorimetric value detection method of EC samples was verified that the RSD was less than 2%, which indicated that the precision, repeatability, and stability of the color value detection method were good. The longer the salting time of EC, the lower the chromaticity index. This change is consistent with the color change of EC decoction pieces from yellowish brown to dark brown and finally to black. This shows that with the prolongation of salting time, the color of EC pieces gradually became darker, which is consistent with the changing trend of chroma parameters. The OPLS‐DA method is used to divide the whole process of EC samples into four categories. The medicinal materials and powders are shown in Figure 3. The measured EC color data were analyzed and imported into SIMCA14.1 software, and OPLS‐DA was selected to establish a chroma qualitative discriminant model. The results are shown in Figure 4A,B. EC is divided into four categories in the salting process: raw EC (0 min), less processed EC (0–4.5 min), finished processed EC (6–9 min), and over‐processed EC (10–11 min). The trend of clustering is consistent with the trend of color from light to deep in the EC salting process. The explained variances of Figure 3B: R²X = 0.992, which shows that the clustering ability of the model is good. The number of principal factors on the Y‐axis and X‐axis is 1. Through the permutation test diagram of Figure 4C, it can be seen that R ² is (0, −0.0303) and Q ² is (0, −0.101). The R ² and Q ² on the left side are smaller than the original value on the right side, which indicates that the established OPLS‐DA model is objective and effective and has good predictive ability.

FIGURE 3.

Medicinal materials and powder pictures of raw EC (A), less processed EC (B), finished processed EC (C), and over‐processed EC (D).

FIGURE 4.

The average chromaticity value of samples in the whole process of EC processing (A). The classification diagram of the color qualitative discriminant model in the whole process of EC processing (B). The whole process of EC processing chromaticity qualitative discriminant model replacement test diagram (C).

3.3. Correlation Analysis of Characteristic Component Content and Chromaticity in Salt Processing of EC

The samples of EC processing were divided into two parts: raw EC‐less processed EC and finished processed EC‐over‐processed EC.

The content data of the characteristic components were correlated with the chromaticity value indexes L*, a*, b*, and E*ab, respectively. The correlation coefficient and significance of the content data and the chromaticity value were obtained by using SPSS25.0 for bivariate correlation analysis. The results are shown in Tables 2 and 3 and Figure 5. It could be seen from the positive and negative correlation coefficients whether the content is positively and negatively correlated with the color value index. If it is positively correlated, the greater the chromaticity value, the higher the content; on the contrary, if it is negatively correlated, the greater the chromaticity value, the lower the content. From the size of the significance, it could be seen whether the color index value is significantly correlated with the content. The correlation results of Table 2 show that in the process of raw EC‐less processed EC–finished processed EC, the chroma value is negatively correlated with the content, the chroma values L*, a*, b*, and E*ab reduced, and the content of six components increased. The contents of chlorogenic acid, genipin, and hyperoside are significantly negatively correlated with chroma. The correlation results in Table 3 show that the chromaticity value is positively correlated with the content, the chromaticity values L*, a*, b*, and E*ab decreased, and the content of the six components reduced in the process of finished processed EC‐over‐processed EC. Among them, the contents of five components—geniposidic acid, chlorogenic acid, genipin, pinoresinol diglucoside, and hyperoside—are significantly positively correlated with chromaticity.

TABLE 2.

Correlation between effective component content and chromaticity of raw EC‐processed less‐salt‐processed EC samples.

Index component	Color index
	L*	a*	b*	E*ab
Geniposidic acid	−0.182	−0.091	−0.193#	−0.182
Chlorogenic acid	−0.474##	−0.568##	−0.545##	−0.489##
Geniposide	−0.025	−0.026	−0.042	−0.025
Genipin	−0.260##	−0.138	−0.315##	−0.266##
Pinoresinol diglucoside	−0.007	−0.018	−0.054	−0.013
Hyperoside	−0.242##	−0.430##	−0.274##	−0.253##

Note: # means significant correlation at the 0.05 level (bilateral), and ## means significant correlation at the 0.01 level (bilateral).

TABLE 3.

Correlation between active ingredient content and chromaticity of salt‐processed EC‐over‐processed samples.

Index component	Color index
	L*	a*	b*	E*ab
Geniposidic acid	0.345##	0.393##	0.331##	0.345##
Chlorogenic acid	0.486##	0.384##	0.412##	0.474##
Geniposide	0.087	0.129	0.074	0.086
Genipin	0.681##	0.655##	0.616##	0.675##
Pinoresinol diglucoside	0.756##	0.751##	0.721##	0.755##
Hyperoside	0.663##	0.597##	0.658##	0.662##

Note: # means significant correlation at the 0.05 level (bilateral), and ## means significant correlation at the 0.01 level (bilateral).

FIGURE 5.

Correlation between characteristic component content and chromaticity of raw EC‐less processed EC–finished processed EC (A). Correlation between characteristic component content and chromaticity of finished processed EC‐over‐processed EC (B).

3.4. The Results of NIR Qualitative Model

3.4.1. Acquisition of NIR Spectroscopy

Firstly, the spectra of EC samples are collected, and the spectra in Figure 6A are obtained. The chemical composition and characteristics of the samples could be identified by analyzing the spectra. Secondly, the spectral data are quantified and imported into SIMCA14.1 software. OPLS‐DA was selected to establish an NIR qualitative discriminant model. The purpose was to quickly determine the salt degree of EC by NIR spectroscopy. The results are shown in Figure 6B. The whole processing process of EC could be clustered into four categories, consistent with the results of the chromaticity qualitative model. Finally, the explained variances of Figure 6B: R²X = 0.991. The number of principal factors on the Y‐axis and X‐axis are 2 and 1, respectively. Through the permutation test diagram of Figure 6C, it can be seen that R ² is (0, 0.0059) and Q ² is (0, −0.0636). The slopes of the two regression lines were large, and the R ² and Q ² obtained by the random arrangement of the left side were smaller than the original value of the right side, indicating that the prediction ability of the original model was greater than that of the random arrangement Y variable, which proved that the model was effective and reliable, and there was no overfitting phenomenon.

FIGURE 6.

NIR average spectra of 123 batches of EC powder (A). NIR qualitative discriminant model classification map of the whole process of EC processing (B). NIR qualitative discriminant model replacement test map of the whole process of EC processing (C).

3.5. Results of NIR Quantitative Model

3.5.1. Selection of Pretreatment Methods, Wavenumber Range, and Factors

PLS was selected to establish a NIR quantitative model in order to quickly grasp the internal composition changes of EC during salt processing. In order to improve the accuracy of the quantitative model, it was necessary to eliminate all kinds of noise and interference by preprocessing the original spectrum before establishing the correction analysis model. The commonly used spectral preprocessing methods include non‐smoothing (NS), multivariate signal correction (MSC), and standard canonical transformation (SNV), Convolution smoothing filter (SG), Norris derivative smoothing filter (ND), the original spectrum (Sp), first derivative (1stD), and second derivative (2ndD). In this experiment, correlation coefficient (R) and RMSEP were used as indicators to investigate and compare them. Precision is usually expressed by RMSEP, which reflects the model's ability to predict new samples. Accuracy is usually represented by RMSEC, which reflects the model's ability to fit the calibration set samples. (The results of the pretreatment method are shown in Table S5, and the optimal treatment method of each component has been thickened.) When the correlation coefficient (R) is closer to 1, the stronger the linear correlation between the predicted value and the actual value of the model, the better the fitting effect of the model. When RMSEC is closer to 0, it means that the prediction error of the model on the calibration set is smaller, and the fitting accuracy of the model is higher. When RMSEP is closer to 0, it means that the prediction error of the model on the prediction set is smaller, and the generalization ability of the model is stronger. RPD is an indicator to measure the predictive ability of the model, which is usually used to evaluate the robustness and prediction accuracy of the model. Generally speaking, RPD > 2.0, indicating that the model has good predictive ability and can be used in practical applications. According to the results, when the pretreatment method was “SNV + 2ndD + ND,” the related parameters of geniposidic acid, chlorogenic acid, and geniposide were optimal. When the pretreatment method was “SNV + 1stD + ND,” the related parameters of genipin and hyperoside were the best. When the pretreatment method was “SNV + 1stD + NS,” the related parameters of pinoresinol diglucoside were the best. When the pretreatment method is “SNV + 2ndD + SG,” the L* and b* related parameters are optimal; when the pretreatment method is “SNV + 2ndD + NS,” the a* related parameters are optimal.

When establishing the calibration model, it is necessary to intercept the spectral band before spectral preprocessing to reduce redundant information. Therefore, the most suitable spectral range was optimized by TQ Analyst 8.3 software with the performance index (PI) as the index.

In the quantitative model established by selection of the number of principal factors PLS, the number of main factors determines the degree of simplification of the model to the data and the amount of information retained. Selecting too many main factors may lead to over‐fitting the model to the noise in the training data, whereas selecting too few main factors may lead to the model failing to fully capture the correlation and structure in the data, affecting the model's predictive ability. The root mean square error of cross‐validation (RMSECV) and the sum of squares of prediction residuals (PRESS) were two key indicators to evaluate the performance of PLS models, both of which could reflect the predictive ability of the model. The main factors of geniposidic acid, chlorogenic acid, geniposide, genipin, pinoresinol diglucoside, hyperoside, L*, a*, and b* were 10, 10, 10, 8, 7, 9, 10, 10, and 10.

3.5.2. Establishment and Validation of the Quantitative Model

The quantitative model was established by the optimal pretreatment method after screening. The specific model parameters and the correlation scatter plots between the actual values of each variable and the predicted values are shown in Table 4 and Figure 7. The spectra of the prediction set samples (n = 10) were input into the quantitative model to obtain the predicted values. The relative deviation was selected as the index to investigate the accuracy of the model. The results are shown in Table 5. The absolute values of the relative deviations of the characteristic component indexes and color indexes in 10 batches of samples are less than 3%, indicating that the accuracy of the model is high. In practical applications, it is used to analyze samples containing unknown content. By importing the NIR spectra of unknown samples into the established model, the content of related compounds can be predicted.

TABLE 4.

Effects of different spectral pretreatment methods on the quantitative model.

Index	Pretreatment	Spectral range (cm−1)	Number of principal factors	Rc	RMSEC	RMSEP	RPD
Geniposidic acid	SNV + 2ndD + ND	7829.58–4049.78	10	0.967 0	0.037 5	0.026 0	5.65
Chlorogenic acid	SNV + 2ndD + ND	6425.65–4207.91	10	0.984 3	0.002 3	0.002 8	4.64
Geniposide	SNV + 2ndD + ND	9977.89–3999.64	10	0.964 6	0.011 8	0.012 1	3.68
Genipin	SNV + 1stD + ND	9831.32–4115.35	8	0.969 6	0.004 8	0.002 7	7.26
Pinoresinol diglucoside	SNV + 1stD + NS	9831.32–4115.35	7	0.969 6	0.016 3	0.012 5	5.32
Hyperoside	SNV + 1stD + ND	6425.65–4207.91	9	0.973 5	0.001 1	0.001 9	2.53
L*	SNV + 2ndD + SG	9993.31–4157.77	10	0.987 8	0.350 0	0.770 0	2.93
a*	SNV + 2ndD + NS	6425.65–4207.91	10	0.993 9	0.063 8	0.127 0	4.61
b*	SNV + 2ndD + SG	6402.51–4034.35	10	0.990 3	0.206 0	0.412 0	3.62

FIGURE 7.

The correlation coefficient scatter plot of the actual measured values and NIR prediction of geniposidic acid (A), chlorogenic acid (B), geniposide (C), genipin (D), pinoresinol diglucoside (E), hyperoside (F), L* (G), a* (H), and b* (I).

TABLE 5.

Absolute value of relative deviation between the predicted value and the actual value (n = 10).

No.	Geniposidic acid (%)	Chlorogenic acid (%)	Geniposide (%)	Genipin (%)	Pinoresinol diglucoside (%)	Hyperoside (%)	L* (%)	a* (%)	b* (%)
1	0.57	1.20	2.26	2.26	2.52	1.52	0.14	0.40	0.97
2	1.99	0.12	2.95	2.35	2.12	1.67	0.14	0.00	0.09
3	1.79	0.50	0.69	2.00	0.79	1.52	0.00	0.00	0.49
4	2.02	1.80	2.72	0.76	2.54	0.00	0.22	1.14	1.05
5	2.06	1.53	1.32	0.66	2.74	1.23	0.08	1.95	1.15
6	0.79	2.27	2.56	2.73	1.56	2.67	0.10	0.00	0.00
7	0.91	0.12	0.12	2.58	2.17	0.90	1.25	1.57	2.72
8	2.42	1.69	2.70	1.77	2.42	1.17	0.62	0.97	0.97
9	2.65	0.88	2.41	2.43	0.70	1.51	0.25	0.68	0.65
10	0.46	2.84	1.06	3.01	0.03	0.00	0.58	0.18	1.87

3.6. Discussion

After being processed by salt, EC could be introduced into the kidney, which could enhance its effect of tonifying the liver and kidneys. From ancient times to the present, the processing methods of EC included honey, ginger juice, bran fried yellow, salt wine fried, salt water fried, and vinegar fried [20]. The processing variety of EC recorded in the 2020 edition of Pharmacopeia was salt‐processed EC, but the processing time and the degree of “broken silk” were not clear. In this study, the contents of six characteristic components in the whole processing process of EC were determined. The results showed that the contents of geniposidic acid, chlorogenic acid, geniposide, genipin, pinoresinol diglucoside, and hyperoside increased first and then reduced, and the contents of each component reached the maximum value in the processing period of 7–9 min. The appearance shape met the requirements of the pharmacopeia.

The traditional processing process often depended on the experience of the operator and the judgment of the naked eye, which may lead to inconsistency and non‐repeatability of the processing results. Using machine vision technology, the color change process was displayed in the form of data, which was more conducive to accurately controlling the quality of processed decoction pieces. By establishing a chromaticity qualitative discriminant model, rapid and objective discrimination of the processing process of EC pieces could be achieved.

During the processing of EC, L*, a*, b* and E*ab all showed a downward trend, which was consistent with the color change of the decoction pieces from yellowish brown to dark brown and finally to black. At the same time, the quantitative data were verified by the OPLS‐DA model, which could be obviously, separated into four groups: unprocessed EC, less processed EC, finished processed EC, and over‐processed EC and the correlation between chroma value data and the content change of processed characteristic components was significant, With the increase of salting time, the content of six components such as geniposidic acid and chlorogenic acid showed a significant positive correlation with L*, a*, and b* values. It can be understood that the longer the salting time of EC, the darker its color and the lower the content of the six active ingredients. Among them, chlorogenic acid, genipin, and hyperoside changed most significantly, indicating that these three components were closely related to the salting process of E. ulmoides , indicating that the salt processing degree of EC decoction pieces could be judged by the change of chroma value. Subsequently, the replacement test confirmed the model's predictability. The outcomes demonstrated that both models were capable of accurately identifying the EC processing degree, which facilitated a speedy assessment of the consistency of the processing quality of the EC pieces.

Finally, the salting process of EC was collected by NIR spectroscopy, and the qualitative discriminant model was established by using the quantitative data of the spectrum, and the whole process of salt processing was successfully clustered into four categories. After the combination of NIR spectral data and chromaticity and characteristic component content data, TQ Analyst was used to optimize the NIR spectra of each sample through pretreatment methods, selection of spectral bands, selection of main factors, and elimination of outliers. The NIR quantitative model of the EC salting process was established. After verification, it was proved that the model had good accuracy and successful modeling. It could be applied in actual production. After scanning the NIR spectrum of the sample, the established quantitative prediction model could be introduced to predict the content and chromaticity value of the processed characteristic components of the sample, so as to realize the real‐time monitoring and quality control of the processing process.

However, this study did not explore the relationship between color and efficacy level, and further experimental verification was needed to provide more comprehensive data support for the quality evaluation of EC.

Funding

This study was financially supported by Hangzhou Agricultural and Social Development Scientific Research Key Project (202204A06), Basic Public Welfare Research Program of Zhejiang Province (LTGN23H280002), Zhejiang Province's “Three Rural Nine Party” Science and Technology Cooperation Plan (2023SNJF031), and Science and Technology Innovation 2025 Major Project of Ningbo (20201ZDYF020069 and 2020Z089).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: Sample information of EC decoction pieces.

Table S2: The specific processing time of three batches of samples.

Table S3: Sampling recovery test.

Table S4: The measured L*, a*, b*, and E*ab values of the samples were measured.

Table S5: Effects of different spectral pretreatment methods on the quantitative model.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.