Rapid and Real-time Determination of Polyphenols in Gongju (Chrysanthemum morifolium Ramat.) at Different Storage Periods by Multispectral Imaging System

Abstract

Gongju, as one of the most famous Chrysanthemum morifolium Ramat. varieties, is considered to be an herbal tea in Asia. Polyphenols determine the quality of Gongju directly. However, polyphenols are instable during the storage process. The feasibility of using multispectral imaging (MSI) system to detect the content of chlorogenic acid, apigenin, and total flavonoids in Gongju at different storage periods was investigated. Spectral data were analyzed with three chemometrics models, including PLS, SVM, and BPNN. Compared with PLS and BPNN, SVM had the greatest prediction result of chlorogenic acid, apigenin, and total flavonoids with R²_p of 0.843, 0.872, 0.734, and RMSEP of 2.926, 0.059, 1.609 mg/g, respectively. These results suggested the MSI system under the SVM model provided a reliable prediction performance on changes of polyphenols in Gongju during storage, so that given a fast, and non-destructive quality identification method for the chrysanthemum industry.

Introduction

Chrysanthemum as the dried anthodium of Chrysanthemum morifolium Ramat., that has various pharmacological effects, such as anti-bacterial (Mezache et al., 2009; Shunying et al., 2005), anti-inflammatory (Cheng et al., 2005) and blood lipid and glucose reduction (Nepali et al., 2018; Yamamoto et al., 2015). Thus, it has been a popular traditional Chinese medicinal and edible crop which was widely used in various health care products. Modern research shows that Chrysanthemum contains plenty of active ingredients, including volatile oils (Zhang et al., 2010), polysaccharides (Du et al., 2016), and polyphenols (Lin and Harnly, 2010). Among these compounds, flavonoids and chlorogenic acid, as typical polyphenols, can be indicators to evaluate the quality of Chrysanthemum because they are related to the antioxidant activity (Li et al., 2019), flavors and taste (Fletcher, 2011; Peterson and Totlani, 2005).

Recent studies have shown that polyphenols are affected by temperature, pH, oxygen, light, and enzyme activity during production or storage (Bakowska et al., 2003; Li et al., 2013), which makes the content variation of polyphenols during storage process has been concerned. Therefore, dynamic detection of flavonoids and chlorogenic acid content could monitor the quality variation during Chrysanthemum storage and is of great significance for the effective utilization and healthy development of the Chrysanthemum industry.

To date, many traditional chromatography-based methods have been used for the detection of active ingredients in Chrysanthemum, including high-performance liquid chromatography (HPLC) (Wu et al., 2017; Zhang et al., 2007), liquid chromatography-mass spectrometry (LC-MS) (Wang et al., 2015) and gas chromatography-mass spectrometry (GC-MS) (Józefczyk et al., 1999). Nevertheless, all of these methods have some common disadvantages, such as destroying the structural integrity of Chrysanthemum, acquiring intricate preprocessing, and time-consuming et.al.

Recently, many spectral imaging techniques, including multispectral imaging, hyperspectral imaging, and near-infrared (NIR) spectroscopy, have been applied to the quality detection of agriculture in a non-destructive way (Cozzolino et al., 2003; ElMasry et al., 2012; Liu et al., 2014). Multispectral imaging (MSI) system, can acquire both spatial and spectral information from the target with less data than hyperspectral imaging, which makes it show potential application for foodstuff detection (Ma et al., 2014), for instance, polyphenols were analyzed in tomato fruit and Iron Buddha tea using MSI (Liu et al., 2015; Xiong et al., 2015), the possibility of classifying varieties and determining the luteolin concentration of Chrysanthemum (Shui et al., 2018). Nevertheless, there are few studies explored the quantitative analysis of flavonoids and chlorogenic acids simultaneously in Chrysanthemum by MSI system.

The objective of this work was to investigate the content variation of chlorogenic acid (CA), apigenin (API), and total flavonoids (TF) in Gongju at three different storage periods, and develop a prediction method for CA, API, and TF in Gongju via MSI system combined with different chemometrics models, including partial least squares (PLSR), support vector machine (SVM), and backpropagation neural network (BPNN).

Materials and Methods

(1) Sample preparation    Gongju, as one of the most typical Chrysanthemum variety, was used as the experiment sample in this study. Dried Gongju materials were purchased from the Kangmei Pharmaceutical Co., Ltd in Bozhou, China. Gongju were harvested and processed in the year of 2016, 2017, and 2018, respectively. All samples without any damage or disease by visual detection were selected for further analysis and were stored at 4 °C in a desiccator. Each group of Gongju included 40 samples and was randomly divided into a training set and a predicting set at a ratio of 30:10. Therefore, the training set contained 90 samples was used to build the prediction model, whereas the predicting set contained 30 samples was used to validate the reliability of prediction model.

(2) Multispectral imaging system    Multispectral images of Gongju were gained by a VideometerLab instrument (Videometer A/S, Hørsholm, Denmark). The MSI system contains 19 wavelengths in 405, 435, 450, 470, 505, 525, 570, 630, 645, 660, 700, 780, 850, 870, 890, 910, 940 and 970 nm ranging from visible to NIR region. The test sample was set in a specially-made sphere, which was coated with matte titanium paint and guaranteed a uniform reflection of the projected light so that a uniform light is formed on the entire sphere. The digital camera above the sphere was a part of the receiving system, which gathered surface reflections and recorded the data with a standard monochrome charge-coupled device chip. Light-emitting diodes (LEDs) were set at the rim of the sphere, each wavelength of LEDs with narrowband spectral radiation distribution were evenly distributed across the entire edge (Fig. 1). These characteristics undertake the best dynamic range and minimize the gloss-related effects, shadow effects, and specular reflections. Before acquiring images, turn on the multispectral imaging system to warm up for at least 30 minutes, and then use diffuse white targets, diffuse dark targets, and geometric targets for calibration. Each LEDs strobe, generate a monochrome image with 32-bit floating point precision, spatial resolution is 2 056 × 2 056 pixels.

Fig. 1.

The primary step of the multispectral imaging system and different wavelengths image of Gongju.

(3) Preprocessing of multispectral images    A necessary and basic step in MSI system processing after finishing image acquisition is the image segmentation process, which is the extract meaningful feature parts in the image, including the edges and regions. Based on the spectral characteristic differences of different components, canonical discriminant analysis (CDA) (Cruz-Castillo et al., 1994) and regions of interest (ROI) segmentation were used. Spectral data was obtained by computer followed the image segmentation, which identified and extracted the spectral character from images. Through image segmentation, a sample was generated into a segmented image, where a single flat part of the Gongju is used as the main ROI to extract spectral data from the sample.

(4) Data analysis    Different chemometrics models, including PLSR, SVM, and BPNN, were applied for mensurable process of the spectral data. All data and analyses were carried out with the commercial software Matlab 2011a (The Mathworks Inc., Natick, MA, USA), Origin8.5 and GraphPad Prism5.

(a) Partial least squares regression (PLSR)

PLSR, as a kind of classic linear calibration method for multivariate data analysis, is wildly used for predicting levels of a substance in many products (Baranska et al. 2006; Lin et al. 2009). PLSR extracts major factors or latent variable (LV). The model is ground on the score of each major factor or the cumulative contribution rate of LV.

(b) Support vector machine (SVM)

The data was found a hyperplane and segmented by SVM, a binary classification model. The mechanism of segmentation is to maximize the interval, which is finally transformed into a convex quadratic programming problem to solve (Sun et al., 2012). SVM was a supervised machine learning approach, developed to solve non-linear problems (Liu et al., 2016). The optimization problem of SVM considers both empirical risk and structural risk minimization; therefore, it has excellent analytical stability (Devos et al., 2009).

(c) Backpropagation neural network (BPNN)

BPNN has been one of the most fashionable neural network topologies since it handles the complex relationship between data input and output. The output data verified the similarity between the result and the training model. The difference between the calculated network output and the expected value can be defined as the network output error. By adjusting the training process and weighting factor, the network output error is gradually reduced to reach the desired selection level (Barma et al., 2011).

(d) Evaluation of model prediction effect

To establish an accurate and reliable prediction model for various components in Gongju, three kinds of model, including PLSR, SVM, and BPNN, was built to compare the prediction effect. The performance of models was estimated by the root mean square error of cross-validation (RMSEC), the root means square error of prediction (RMSEP), and the correlation coefficients (R) for calibration (R_c) and prediction (R_p). Generally, a model with higher R_c, R_p, and lower RMSEP, RMSEC, can be regarded as a reliable model (Liu et al., 2017).

(5) Determination of polyphenols    After collecting spectral data, all Gongju samples were powdered with a mortar in liquid nitrogen immediately and passed through 100 µm pore mesh. Each sample was weighed, and extracted with 3 mL 70% methanol (v/v) per 0.1 g. The extraction process was performed in an ultrasonic cleaner bath (G100S, Shenzhen Chuangmei Cleaning Equipment Co., Ltd, Shenzhen, China) at 25 °C for 30 min. Then, the samples were centrifuged at 4 000 rpm for 10 min to collect the supernatant, and the supernatant was used for the detection of polyphenols. All reference substances were bought from Sigma (Sigma-Aldrich, Shanghai, China).

(a) Measurement of chlorogenic acid (CA) and apigenin (API)

HPLC system (Agilent 1260 Infinity II, Agilent Technologies, Palo Alto, CA, USA) with Agilent SB-C18 column (4.6 × 250 mm, 5 µm) at 25 °C was used to detect CA and API content. All sample supernatant was filtered through a 0.22 µm syringe filter before injecting it into HPLC. The mobile phase was a mixture of methanol and deionized water in the ratio of 65:35 (v/v), and the flow rate was 0.8 mL/min. The injection volume of each sample was 20 µL, and the Ultraviolet detection wavelength was 350 nm (Shui et al., 2018).

(b) Measurement of total flavonoids (TF)

This measurement used a colorimetry method for determining the total flavonoids through absorbance at a wavelength of 510 nm (Chlopicka et al., 2012). Compared with the standard rutin substance, the total flavonoids in the sample are quantitatively determined. Accurately measure rutin reference solution (0.1 mg/mL, 70% methanol) 0, 0.50, 1.00, 2.00, 3.00, 4.00, 5.00 mL, diluted to 5 mL. Then adding 0.3 mL of 5% sodium nitrite solution and shaking for 6 min, adding 0.3 mL of 10% aluminum nitrate solution and shaking for 6 min, adding 2 mL of 1.0 mol/L sodium hydroxide solution and shaking for 12 min, diluting system to 5 mL. Measure the absorbance at a wavelength of 510 nm, with the zero tubes as the blank, the rutin content (µg) as the abscissa, and the absorbance as the ordinate, draw a standard curve and calculate the correlation coefficient (r). According to the standard curve preparation procedure, taking specimen 0.5 mL determines the absorbance at 510 nm. Based on the standard curve, the calculated rutin content is equal to the absorbance of the sample.

Results and Discussion

(1) Reflectance spectral profiles of Gongju    The mean reflectance spectral data of three different storage periods Gongju (Gongju 2016, Gongju 2017, Gongju 2018) were extracted from the MSI in a wavelength range from 405 to 970 nm. As shown in Fig. 2, the shapes of the reflectance spectral curves were similar among Gongju at different storage periods and kept a tendency of escalating. The reflectance values of Gongju 2018 were the highest, and Gongju 2016 was the lowest, suggesting the reflectance values were decreased with increasing storage time. Moreover, the curves of Gongju 2016 and Gongju 2017 overlapped, since the reflectance values of Gongju 2018 approached the other two kinds of Gongju in the NIR region (780–970 nm).

Fig. 2.

Average reflection spectra from MSI for different storage periods of Gongju.

Polyphenols, carotenoids, and other substances determines the changed color of plants (Kishimoto and Ohmiya, 2006; Ohmiya, 2011), and the color variation of Gongju were clearly observed in Fig. 2. These results suggested that many ingredients may have changed during the storage process. The difference of reflectance spectral value in the visible region (405-700 nm) was mainly caused by the changed color of Gongju, whereas the physical form and ingredients contributed to the difference of reflectance spectral value in NIR region (Shui et al., 2018).

(2) Reference measurement of CA, API, and TF    To quantify the contents of CA, API, and TF in Gongju from different harvest years, HPLC or colorimetry methods were used, and the results were shown in Table 1. Forty samples from each group of Gongju were randomly divided into two groups, of which thirty were in the calibration set, and ten were in the prediction set. According to the results, the ingredients of Gongju in different harvest years varied greatly after storage. (Fig. 3). The bars in Fig. 3 showed the standard deviation (SD) of CA, API, and TF content in each group of Gongju.

Table 1. Comparison of CA, API and TF content (mg/g) in Gongju at different storage periods. (n= 40)

	Sample	Set	Amount	Mean	SD	Range
CA	Gongju	Calibration	30.000	35.270	8.593	19.45–54.28
	2018	Prediction	10.000	37.610	3.863	30.88–43.90
	Gongju	Calibration	30.000	53.930	9.974	31.65–70.68
	2017	Prediction	10.000	45.910	5.252	39.04–54.13
	Gongju	Calibration	30.000	31.660	5.027	20.83–47.07
	2016	Prediction	10.000	32.390	5.019	23.78–39.19
API	Gongju	Calibration	30.000	0.171	0.031	0.130–0.235
	2018	Prediction	10.000	0.174	0.030	0.141–0.224
	Gongju	Calibration	30.000	0.218	0.044	0.151–0.338
	2017	Prediction	10.000	0.262	0.027	0.224–0.306
	Gongju	Calibration	30.000	0.395	0.072	0.209–0.533
	2016	Prediction	10.000	0.313	0.100	0.211–0.480
TF	Gongju	Calibration	30.000	16.710	3.840	10.88–29.80
	2018	Prediction	10.000	17.450	2.333	14.56–23.00
	Gongju	Calibration	30.000	25.710	8.889	12.05–50.36
	2017	Prediction	10.000	21.660	2.325	18.56–26.20
	Gongju	Calibration	30.000	18.290	7.545	7.185–34.71
	2016	Prediction	10.000	18.580	2.673	14.58–22.31

SD: standard deviation

Fig. 3.

Content of (A) CA, (B) API, (C) TF in Gongju.

According to previous studies, in other foodstuffs or herb, polyphenols varied with storage conditions such as time and temperature (Chen et al., 2013; Fu et al., 2017; Piasecka et al., 2013). Congruously, through analyzing polyphenols compounds in Gongju at different storage periods, we found that the storage of Gongju caused API to change on a rising trend. Furthermore, the tendencies change of CA and TF implied the degradation and synthesis of polyphenols were simultaneous, but the rate might be inequable at different storage periods. Meanwhile, the difference of polyphenols content caused by different harvest years cannot be ignored.

(3) Prediction of polyphenols in Gongju by different chemometrics models    Based on the distinction in the reflectivity curves of Gongju at different storage periods, three different chemometrics method was tested to process the spectral data for detection of polyphenols. As seen in Table 2, compared with PLSR and BPNN model, the SVM model exhibited the best prediction performance in CA, TF, and API content. And Fig. 4 showed the SVM model for prediction of CA, API and TF had the highest R²_c of 0.936, 0.992, 0.631, R²_p of 0.843, 0.872, 0.734, RMSEC of 3.187, 0.009, 4.882 mg/g, and RMSEP of 2.926, 0.059, 1.609 mg/g, respectively. The prediction results processed by the SVM model improved in the order of TF, CA, and API. Since API content was in a lower variability, the prediction result was better than the other two polyphenols. Besides, TF as a complex mass of flavonoids was determined by the traditional colorimetry method, and it was not as precise as HPLC (Liu et al., 2015). Therefore, it is rational that the TF prediction result was less accurate.

Table 2. The prediction result of CA, API and TF content in Gongju by PLS/SVM/BPNN models

Quality	Model	R2c	RMSEC (mg/g)	R2p	RMSEP(mg/g)
CA	PLS	0.687	7.049	−0.261	8.044
	SVM	0.936	3.187	0.843	2.926
	BPNN	0.572	2.451	−1.195	2.313
API	PLS	0.703	0.058	0.412	0.101
	SVM	0.992	0.009	0.872	0.059
	BPNN	0.633	0.010	0.374	0.003
TF	PLS	−0.052	6.183	−24.189	14.362
	SVM	0.631	4.882	0.734	1.609
	BPNN	0.231	2.847	−2.424	1.375

Fig. 4.

Measured vs. predicted content of (A) CA, (B) API, and (C) TF by SVM model.

Although the PLSR model is widely used in chemometrics, quite a few studies have demonstrated that its prediction performance for some substances is not as good as non-linear models, such as SVM and BPNN models (Gowen et al., 2007). One possible explanation is the relationship between spectral information and the content of some compounds may be complex and not strictly linear. BPNN model always be impacted by the choice of initial training parameters. Owing to three groups of Gongju had multiple initial training parameters, BPNN model was possibly only suitable for local optimal solutions but not global optimal solutions (Kuo et al., 2008). On account of the self-learning and self-adjustment characteristics of the SVM model, it could achieve a better prediction result (Cortes and Vapnik, 1995).

Conclusion

In this study, the application of the MSI system for rapid and non-destructive detection of polyphenols, including chlorogenic acid, apigenin, and total flavonoids in Gongju at different storage periods, was explored. Firstly, we found the content of polyphenols in Gongju may vary greatly due to harvest and storage years. By comparison with PLSR and BPNN models, the SVM model had the best prediction performance on the all three ingredients. The above results indicated that multispectral imaging combined with the SVM model could be used as a reliable method to detect the content of various polyphenols in Gongju, and it is also applicable to chrysanthemums at different storage periods.

Acknowledgements    This study was supported by the Fundamental Research Funds for the Central Universities (JZ2018QTXM0553, JZ2020HGTB0044).

References

•  Bakowska,  A.,  Kucharska,  A.Z., and  Oszmiański,  J. (2003). The effects of heating, UV irradiation, and storage on stability of the anthocyanin–polyphenol copigment complex. Food Chem., 81, 349-355.
•  Baranska,  M.,  Schütze,  W., and  Schulz,  H. (2006). Determination of lycopene and β-carotene content in tomato fruits and related products: comparison of FT-Raman, ATR-IR, and NIR spectroscopy. Anal. Chem., 78, 8456-8461.
•  Barma,  S.D.,  Das,  B.,  Giri,  A.,  Majumder,  S., and  Bose,  P. (2011). Back propagation artificial neural network (BPANN) based performance analysis of diesel engine using biodiesel. J. Renewable Sustainable Energy, 3, 013101.
•  Chen,  Y.J.,  Kuo,  P.C.,  Yang,  M.L.,  Li,  F.Y., and  Tzen,  J.T. (2013). Effects of baking and aging on the changes of phenolic and volatile compounds in the preparation of old Tieguanyin oolong teas. Food Res. Int., 53, 732-743.
•  Cheng,  W.,  Li,  J.,  You,  T., and  Hu,  C. (2005). Anti-inflammatory and immunomodulatory activities of the extracts from the inflorescence of Chrysanthemum indicum Linne. J. Ethnopharmacol., 101, 334-337.
•  Chlopicka,  J.,  Pasko,  P.,  Gorinstein,  S.,  Jedryas,  A., and  Zagrodzki,  P. (2012). Total phenolic and total flavonoid content, antioxidant activity and sensory evaluation of pseudocereal breads. Lwt Food Science & Technology, 46, 550-555.
•  Cortes,  C. and  Vapnik,  V. (1995). Support-vector networks. Machine Learning, 20, 273-297.
•  Cozzolino,  D.,  Smyth,  H.E., and  Gishen,  M. (2003). Feasibility study on the use of visible and near-infrared spectroscopy together with chemometrics to discriminate between commercial white wines of different varietal origins. J. Agric. Food Chem., 51, 7703-7708.
•  Cruz-Castillo,  J.,  Ganeshanandam,  S.,  MacKay,  B.,  Lawes,  G.,  Lawoko,  C., and  Woolley,  D. (1994). Applications of canonical discriminant analysis in horticultural research. HortScience, 29, 1115-1119.
•  Devos,  O.,  Ruckebusch,  C.,  Durand,  A.,  Duponchel,  L., and  Huvenne,  J.P. (2009). Support vector machines (SVM) in near infrared (NIR) spectroscopy: Focus on parameters optimization and model interpretation. Chemom. Intell. Lab. Syst., 96, 27-33.
•  Du,  N.,  Tian,  W.,  Zheng,  D.,  Zhang,  X., and  Qin,  P. (2016). Extraction, purification and elicitor activities of polysaccharides from chrysanthemum indicum. Int. J. Biol. Macromol., 82, 347-354.
•  ElMasry,  G.,  Barbin,  D.F.,  Sun,  D.W., and  Allen,  P. (2012). Meat quality evaluation by hyperspectral imaging technique: an overview. Crit. Rev. Food Sci. Nutr., 52, 689-711.
•  Fletcher,  J.N. (2011). Isolation, Identification, and biological evaluation of potential flavor modulatory flavonoids from Eriodictyon californicum. The Ohio State University
•  Fu,  M.,  Xu,  Y.,  Chen,  Y.,  Wu,  J.,  Yu,  Y.,  Zou,  B.,  An,  K., and  Xiao,  G. (2017). Evaluation of bioactive flavonoids and antioxidant activity in Pericarpium Citri Reticulatae (Citrus reticulata ‘Chachi’) during storage. Food Chem., 230, 649-656.
•  Gowen,  A.,  O'Donnell,  C.,  Cullen,  P.,  Downey,  G., and  Frias,  J. (2007). Hyperspectral imaging-an emerging process analytical tool for food quality and safety control. Trends Food Sci. Technol., 18, 590-598.
•  Józefczyk,  A.,  Markowski,  W.,  Mardarowicz,  M., and  Glowniak,  K. (1999) Preliminary GC/MS study of the essential oil isolated from Chrysanthemum maximum. Pharmaceutical biology, 37, 8-12.
•  Kishimoto,  S. and  Ohmiya,  A. (2006). Regulation of carotenoid biosynthesis in petals and leaves of chrysanthemum (Chrysanthemum morifolium). Physiol. Plant., 128, 436-447.
•  Kuo,  C.F.,  Chang,  C.D.,  Su,  T.L., and  Fu,  C.T. (2008). Optimization of the dyeing process and prediction of quality characteristics on elastic fiber blending fabrics. Polym.-Plast. Technol. Eng., 47, 678-687.
•  Li,  S.,  Lo,  C.Y.,  Pan,  M.H.,  Lai,  C.S., and  Ho,  C.T. (2013). Black tea: chemical analysis and stability. Food Funct., 4, 10-18.
•  Li,  Y.,  Yang,  P.,  Luo,  Y.,  Gao,  B.,  Sun,  J.,  Lu,  W.,  Liu,  J.,  Chen,  P.,  Zhang,  Y., and  Yu,  L.L. (2019). Chemical compositions of chrysanthemum teas and their anti-inflammatory and antioxidant properties. Food Chem., 286, 8-16.
•  Lin,  H.,  Chen,  Q.,  Zhao,  J., and  Zhou,  P. (2009). Determination of free amino acid content in Radix Pseudostellariae using near infrared (NIR) spectroscopy and different multivariate calibrations. J. Pharm. Biomed. Anal., 50, 803-808.
•  Lin,  L.Z. and  Harnly  J.M. (2010). Identification of the phenolic components of chrysanthemum flower (Chrysanthemum morifolium Ramat). Food Chem., 120, 319-326.
•  Liu,  C.,  Liu,  W.,  Chen,  W.,  Yang,  J., and  Zheng,  L. (2015). Feasibility in multispectral imaging for predicting the content of bioactive compounds in intact tomato fruit. Food Chem., 173, 482-488.
•  Liu,  C.,  Liu,  W.,  Lu,  X.,  Chen,  W.,  Yang,  J., and  Zheng,  L. (2016). Potential of multispectral imaging for real-time determination of colour change and moisture distribution in carrot slices during hot air dehydration. Food Chem., 195, 110-116.
•  Liu,  C.,  Liu,  W.,  Lu,  X.,  Ma,  F.,  Chen,  W.,  Yang,  J., and  Zheng,  L. (2014). Application of multispectral imaging to determine quality attributes and ripeness stage in strawberry fruit. PloS one., 9, e87818.
•  Liu,  C.,  Liu,  W.,  Yang,  J.,  Chen,  Y., and  Zheng,  L. (2017). Non-destructive detection of dicyandiamide in infant formula powder using multi-spectral imaging coupled with chemometrics. J. Sci. Food Agric., 97, 2094-2099.
•  Ma,  F.,  Yao,  J.,  Xie,  T.,  Liu,  C.,  Chen,  W.,  Chen,  C., and  Zheng,  L. (2014). Multispectral imaging for rapid and non-destructive determination of aerobic plate count (APC) in cooked pork sausages. Food Res. Int., 62, 902-908.
•  Mezache,  N.,  Akkal,  S.,  Laouar,  H., and  Seguin,  E. (2009). Flavonoids from Chrysanthemum myconis and their antibacterial activity. Chem. Nat. Compd., 45, 715.
•  Nepali,  S.,  Cha,  J.Y.,  Ki,  H.H.,  Lee,  H.Y.,  Kim,  Y.H.,  Kim,  D.K.,  Song,  B.J., and  Lee,  Y.M. (2018). Chrysanthemum indicum inhibits adipogenesis and activates the AMPK pathway in high-fat-diet-induced obese mice. The American journal of Chinese medicine, 46, 119-136.
•  Ohmiya,  A. (2011). Diversity of carotenoid composition in flower petals. JARQ, 45, 163-171.
•  Peterson,  D.G. and  Totlani,  V.M. (2005). Influence of flavonoids on the thermal generation of aroma compounds. ACS Publications, pp. 143-160.
•  Piasecka,  E.,  Uczciwek,  M.,  Klewicki,  R.,  Konopacka,  D.,  Mieszczakowsk,  M.,  Szulc,  M., and  Bonazzi,  C. (2013). Effect of long-time storage on the content of polyphenols and ascorbic acid in osmo-convectively dried and osmo-freeze-dried fruits. J. Food Process. Preserv., 37, 198-209.
•  Shui,  S.,  Liu,  W.,  Liu,  C.,  Yan,  L.,  Hao,  G.,  Zhang,  Y., and  Zheng,  L. (2018). Discrimination of cultivars and determination of luteolin content of Chrysanthemum morifolium Ramat. using multispectral imaging system. Anal. Methods., 10, 1640-1646.
•  Shunying,  Z.,  Yang,  Y.,  Huaidong,  Y.,  Yue,  Y., and  Guolin,  Z. (2005). Chemical composition and antimicrobial activity of the essential oils of Chrysanthemum indicum. J. Ethnopharmacol., 96, 151-158.
•  Sun,  X.,  Chen,  K.,  Maddock-Carlin,  K.,  Anderson,  V.,  Lepper,  A.,  Schwartz,  C.,  Keller,  W.,  Ilse,  B.,  Magolski,  J., and  Berg,  E. (2012). Predicting beef tenderness using color and multispectral image texture features. Meat Sci., 92, 386-393.
•  Wang,  S.,  Hao,  L.J.,  Zhu,  J.J.,  Wang,  Z.M.,  Zhang,  X., and  Song,  X.M. (2015). Comparative Evaluation of Chrysanthemum Flos from Different Origins by HPLC-DAD-MS n and Relative Response Factors. Food analytical methods, 8, 40-51.
•  Wu,  Y.,  Wang,  X.,  Xue,  J., and  Fan,  E. (2017). Plant phenolics extraction from Flos chrysanthemi: Response surface methodology based optimization and the correlation between extracts and free radical scavenging activity. J. Food Sci., 82, 2726-2733.
•  Xiong,  C.,  Liu,  C.,  Pan,  W.,  Ma,  F.,  Xiong,  C.,  Qi,  L.,  Chen,  F.,  Lu,  X.,  Yang,  J., and  Zheng,  L. (2015). Non-destructive determination of total polyphenols content and classification of storage periods of Iron Buddha tea using multispectral imaging system. Food Chem., 176, 130-136.
•  Yamamoto,  J.,  Tadaishi,  M.,  Yamane,  T.,  Oishi,  Y.,  Shimizu,  M., and  Kobayashi-Hattori,  K. (2015). Hot water extracts of edible Chrysanthemum morifolium Ramat. exert antidiabetic effects in obese diabetic KK-Ay mice. Biosci. Biotechnol. Biochem., 79, 1147-1154.
•  Zhang,  C.,  Qin,  M.J.,  Shu,  P.,  Hong,  J.L.,  Lu,  L., and  He,  D.X. (2010). Chemical variations of the essential oils in flower heads of Chrysanthemum indicum L. from China. Chem. Biodiversity, 7, 2951-2962.
•  Zhang,  Q.,  Li,  J.,  Wang,  C.,  Sun,  W.,  Zhang,  Z., and  Cheng,  W. (2007). A gradient HPLC method for the quality control of chlorogenic acid, linarin and luteolin in Flos Chrysanthemi Indici suppository. J. Pharm. Biomed. Anal., 43, 753-757.