IR spectroscopy studies the interaction between matter and infrared radiation. The main principle is that IR light’s energy could trigger the mechanical motion of specific molecular bonds when the IR light passes the sample, which is called IR absorption. A specific characteristic absorption presented in the IR spectra is employed for analysis according to the absorption frequency of chemical bonds and functional groups (Stark et al., 1986). The mechanical motion (vibration and rotation patterns) of atoms connected by covalent bonds includes symmetric and asymmetric stretching and scissoring, wagging, rocking, and twisting (Johnson and Naiker, 2020). IR spectroscopy contains richer group information with tremendous advantages in analysing and identifying organic substances, which has been widely used since the 1960s (Smith, 2011) and can be used for both qualitative and quantitative analysis (Stuart, 2005). The ingredients of concern in herbs, such as saponins, polysaccharides, flavonoids, triterpenoids, and polyphenols, consist of various organic functional groups that contribute to characteristic bands or peaks in IR spectra. The differences among spectra can be conducted to analyse the quality discrepancies of herbs. In the field of herbs, IR spectroscopy technique has been used since the early 1980s (Zou et al., 2005). Nowadays, IR spectroscopy is the most widely used technology in the quality detection of herbs, such as the identification of species, origins, grades, and the prediction of compound contents (Yin et al., 2019).

A typical IR spectrometer comprises a radiation source, a wavelength selection device, a sample holder, a photoelectric detector, and a computer system (Porep et al., 2015). The spectra acquisition modes include transmission, reflection, transflection, and interaction, which differ in how the detectors are placed with respect to the samples (Alander et al., 2013; Monnier, 2018). The IR region is conditionally divided into three subregions, including near-infrared (NIR, 12,820-4,000 cm^-1), mid-infrared (MIR, 4,000-400 cm^-1), and far-infrared (FIR, 400-33 cm^-1). The quality analysis of herbs mainly focuses on NIR and MIR spectra caused by molecular vibration. FIR spectra excite molecular rotation and have strong water absorption, which is more suitable for heavy metal analysis (Su and Sun, 2018).

Various chemical bonds related to fundamental vibrations of molecules could be detected in MIR spectroscopy. The number of scans, resolution, and scan regions are vital parameters that affect signal-to-noise ratio (SNR) and spectra quality. The MIR region can be divided into two distinct regions, 4,000-1,500 cm^-1 and 1,500-400 cm^-1, which are called functional group region and fingerprint region, respectively. In the fingerprint region, 4,000-2,500 cm^-1 is X-H (where X is C, N, O, or S) stretching vibration. 2,500-2,000 cm^-1 corresponds to triple bonds, such as C≡C, C≡N). Double-bonded functional groups, like C=C, C=O, C=N, mainly lie in 2,000-1,500 cm^-1. The peak near 3340, 1739, and 1670 cm^-1 were assigned to the stretching vibration of O-H, ester carbonyl groups, and C=C, respectively, in the IR spectrum of Dictamnus dasycarpus Turcz (Liu et al., 2020). The peaks at 1684, 1517, and 1031 cm^-1 were observed and compared to distinguish the different Rhodiola species (Tang et al., 2020). The region of 1200-950 cm^-1 was chiefly assigned to the vibration of C-O related to polysaccharides (Wu et al., 2019).

MIR spectra are collected mainly by Fourier transform infrared (FTIR) spectrometers, which are equipped with a Michelson interferometer instead of the traditional grating monochromator, significantly improving the scanning speed, SNR, and the wavelength resolution of MIR spectroscopy. The advantages of FTIR are as follows: (i) non-destructive or only slightly damages the sample; (ii) needed sample quantities are small for measuring; (iii) requires minimal sample preparation at most. Meanwhile, the shortcomings of spectrum complication, quantification, and sample constraint are needed to be considered. Spectrum complications and quantification are solved by digitalisation and chemometric methods. More advanced FTIR techniques are developed to overcome the sample constraint, by which samples do not undergo time-consuming preparation that lets samples be combined with KBr. ATR is a sampling technique that is used to obtain high-quality data on liquid and solid. ATR-FTIR relies on the total internal reflection of infrared light in an internal reflection element or crystal with a high reflection index in direct contact with the measured sample, simplifying sample preparation (Feng et al., 2013). Amazing consistent sampling and higher accuracy may be achieved due to the presence of multi-reflective crystals, in which light is reflected on the sample many times, thereby increasing the absorbance (Lohumi et al., 2015). The limitation of ATR-FTIR is that it is challenging to achieve an ideal optical fit between the sample and the ATR crystal (Monnier, 2018). Diffuse reflectance FTIR (DRIFT) spectrometers and FTIR photoacoustic spectroscopy (FTIR-PAS) are developed for the direct determination of powder samples. The diffuse reflection accessories can collect the diffuse reflected light with absorption-attenuation characteristics caused by an uneven or rough surface, which obtains spectral signals with a good SNR to the maximum extent (Huang et al., 2008). DRIFT is suitable for the surface structure analysis of opaque or irregular solid samples. The advantage is that almost no preparation is required for the sample, which can be in powder form or film. FTIR-PAS collects spectral data from the pressure fluctuations generated by thermal expansion, which is detected by a sensitive microphone. Photoacoustic techniques mainly include modulated excitation and generation of sound waves in gaseous samples, modulated excitation of liquid and solid samples with an indirect generation of sound waves in the adjacent gas phase, and pulsed excitation and generation of pressure pulses in liquid and solid samples (Schmid, 2006). Rather than focusing on what is transmitted or reflected, FTIR-PAS measurement relies on the energy absorbed by samples, making it suitable for high-scattered, opaque, weak-absorbed, and low-concentration samples (Du and Zhou, 2011).

Unlike almost all modern MIR spectrometers based on Fourier transform, monochromator/detector principles in scanning NIR spectroscopy are variable. NIR spectroscopy lies between visible and infrared light, comprising broad bands associated with molecular overtones and combinations of vibrations. According to different combinations, simple molecules with few basic vibration modes can present many overtones in the NIR spectroscopy. NIR is sensitive to hydrogen groups such as O–H, N–H, and C–H (Wang et al., 2016). Therefore, the moisture of samples is needed to be considered (Buening-Pfaue, 2003). Infrared signals are easier to detect, but the overlapping of NIR spectra will affect the interpretation. As a result, NIR spectral data is analysed with a combination of chemometric methods to extract valuable information.

The change in the frequency of light scattered by molecules as it travels through a medium is called Raman scattering, discovered by C.V. Raman in 1928, relying on the inelastic scattering of photons known as Raman scattering (Raman and Krishnan, 1928). Raman scattering is a combined light scattering phenomenon produced by the interaction of light and matter molecules. The principle of Raman spectroscopy is analysing the scattering spectra with different frequencies from the incident light, which is applied to the study of the molecular structure of matter in specific wavenumber (Qin et al., 2019). Raman spectra cover a range of 4,000-50 cm^-1. The advantages of Raman include sensitivity to chemical structure within the fingerprint regions and easy analysis without pre-treatment. Besides, due to the weak Raman scattering of water, Raman can be applied in an aqueous environment. The vibrations of various functional groups in herbs produce peaks at different positions due to unique spectroscopic fingerprints. The information of a class of chemical compounds with similar molecular structures can be deduced. For example, the peak at 1626 cm^-1 is assigned to the stretching vibration of C=C bonds (Wong et al., 2015). Adulterations can also be recognized by comparing the peaks in Raman spectra. Therefore, the detection of species, adulteration, and ingredients, as well as processing monitoring using Raman spectra, is a feasible application in the field of herbs.

A typical Raman spectrometer consists of five components: laser light source, filter, sample cell, monochromator, and detector (Eliasson et al., 2008). There are many types of lasers, ultraviolet laser, visible laser, and NIR laser, available to be applied. The selection of laser depends on samples and detection purposes, which can be considered in three aspects. (i) The intensity of the Raman signal. According to the acknowledged relationship, I_Raman∝1/λ⁴, the shorter wavelength of the laser, the stronger the Raman signal. (ii) Avoid fluorescent interference to prevent the annihilation of the Raman signal by fluorescent signal. Choosing an excitation laser outside the fluorescent region, like an ultraviolet laser or NIR laser, can avoid the fluorescence effect. (iii) The need to analyse samples at different depths. The longer the wavelength of the laser, the deeper the penetration (Lee et al., 2013). Basic Raman measurement techniques contain backscattering, transmission, and spatially offset Raman spectroscopy (SORS) (Qin et al., 2019). The backscattering collection mode is mainly used for sample surface inspection. The transmission collection mode is more suitable for the bulk composition of samples that are non- or weak-absorbing inside (Eliasson et al., 2008). SORS has the capability to obtain layered information on samples by setting a series of lateral offsets (Nicolson et al., 2021). Generally, the selection of lasers and measurement modes is based on the characteristics of the samples. Shorter excitation wavelengths could excite stronger Raman signals, but higher energy damages the sample more. Meantime, the cost and volume of the instrument increase.

The main disadvantages of Raman spectroscopy are the thermal effects of the sample, fluorescence interference, and weak Raman signals (Wang et al., 2018). Such obstacles could be overcome by the advancements in devices and materials (Chen et al., 2017), which lead to a greater variety of analytical techniques (Gala and Chauhan, 2015). Fourier transform Raman spectroscopy (Liao et al., 2004), resonance Raman spectroscopy (RRS) (Robert, 2009), confocal Raman spectroscopy (Barbillat et al., 1994), and surface-enhanced Raman spectroscopy (SERS) (Sharma et al., 2012; Pérezjiménez et al., 2020), is feasible to enhance Raman signals by 10³-10⁶ times, which evolves the instruments and samples processing. FT-Raman adopts Fourier transform technique and is equipped with a NIR laser (1064 nm) as an excitation light source that avoids fluorescence interference. But its baseline drift and poor reproducibility affect its Raman signal. RRS depends on the resonance effect that the frequency of the laser matches an electronic transition of the irradiated molecule. Its instrument requires an adjusted light source. Confocal Raman spectroscopy which is the coupling of Raman to microspectroscopic instruments, can provide a high-resolution image rich in information. The development of techniques and instruments creates more practical applications in Raman spectroscopy.

In the field of herbs, SERS has become a hotspot for analysis. Fleischmann et al. (1974) discovered SERS during measurements of Raman scattering of pyridine on rough silver electrodes. SERS amplifies conventional Raman signals by combining nanostructures of noble metals with the sample, whose theoretical mechanisms involve electromagnetic enhancement and chemical enhancement. Nanomaterials improvements and chemical modifications offer more possibilities for SERS applications, advancing towards selectivity, in situ, and non-destructive sampling detection (Lin and He, 2019; Langer et al., 2020), which has achieved feasible applications in adulteration detection (Dao et al., 2019), compound identification (Gu et al., 2018), and on-site qualitative screening (Zhu et al., 2014). The complex matrix effect and limited multi-analyte capability are needed to be considered. Nevertheless, the great compatibility with other techniques, such as separation techniques and other innovations and variants of Raman spectroscopy, makes SERS promising.

In the field of herbs, clinical efficacy is due to multiple compositions working in concert. IR and Raman spectra that reflect the comprehensive chemical profiles related to composition are feasible to be applied in the qualitative analysis of herbs, including identifying the species, grades, origins, and quantitative prediction. In Figure 2 , Panax notoginseng is selected as a typical case, and IR and Raman spectroscopy techniques are adopted for quality and safety inspection throughout the process. The comparison of IR and Raman spectroscopy is summarized in Table 1 . Each technique has its own advantages and disadvantages. The detection method selection should be based on sample characteristics and detection purpose. Herb materials’ compounds are complex, and IR and Raman spectroscopy techniques complement each other. IR spectroscopy detects the molecule with IR absorption when its dipole moment changes. The molecular bond without dipole moment but with polarizability change can be detected in Raman spectroscopy. The characteristic peak information of MIR spectroscopy is pointed to specific databases that are relatively complete. Meanwhile, as mentioned in section 2.2, the poor spectral reproducibility and SNR in Raman spectra were caused by the fluorescence and sample matrix effect, which leads to the status that Raman spectroscopy is less widely used than IR spectroscopy in herbs (Woo et al., 1999c).

IR and Raman spectroscopy techniques are available to detect samples that are in the original state without sample pre-processing. However, simple sample preparation is employed before collecting spectral data to gain high-quality spectra and better analysis results. Dried samples after grinding and tableting, or extracts, are two commonly used herbal raw materials for analysis, which decreases the matrix effect. Digital technology offers a data processing method to remove irrelated variables, which makes sample pre-processing unnecessary. SPA-LDA algorithm extracted seven effective variables to achieve the three origins discrimination of Ginseng in piece form using NIRS (Chen et al., 2020b).

Spectral analysis can realize the quality identification and evaluation of herbs with different qualities. Currently, the accuracy and precision of herb detection using IR and Raman spectroscopy techniques are not up to the standard analysis methods. The outstanding advantages of rapid, accurate, and online analysis endow the application prospect of the digital detection and automation industry.

Influenced by samples’ physical factors (compactness, smoothness, particle size, etc.), instrument error, and the experimental environment, IR and Raman spectroscopy inevitably present some irrelevant information to the target samples, which results in baseline drift, spectral overlap, and background noise. Therefore, spectral pre-treatment is used to remove defects observed in the spectra and amplify the differences in the raw spectra of samples (Baker et al., 2014).

After many attempts, baseline and noise correction (Liu and Yu, 2016), derivation techniques (Rinnan, 2014), and light scattering correction (Buddenbaum and Steffens, 2014) are considered effective pre-treatment methods, which are common strategies in the spectral detection of herbs. Baseline correction (BC) is widely applied to correct the spectra by particle size variation and instrumental factors (Cadet and Offmann, 1996). It needs to be noted that variable baseline signals and cosmic rays in Raman spectra contribute to spectral contamination causing less sensitivity. Baseline correction and cosmic ray removal are the primary purposes of Raman spectra pre-treatment (Li et al., 2015). Noise correction helps to improve the signal-to-noise ratio (Liu et al., 2019). Smoothing and filtering (SF), wavelet transform (WT), and normalization (Norm) are applied to correct the baseline or reduce noise. Common derivative pre-treatments include the first derivative (FD), second derivative (SD), and third derivative (TD), which are used to amplify spectral differences. Multiplicative scatter correction (MSC) and standard normalized variate (SNV) are applied to correct the spectral errors caused by scattering.

Additionally, multiple pre-processing methods were combined to attain the best performance in studies (Rinnan et al., 2009). The effective and efficient improvements in models were compared to obtain the most optimal one for further analysis (Xu et al., 2019).

The data collected by vibrational spectroscopy techniques is generally too much and may result in redundant information interfering with the correlation, which is unfavourable to the establishment of the model. Extracting feature wavelengths is a practical approach to improve the robustness of the model. Though not an indispensable step when processing the data, variable selection is usually employed to remove or eliminate useless variables with noise and irrelevance, even interference. By using these methods, the bias caused by chemical, physical, and instruments decreases, which results in better predictions and simpler models (Zou et al., 2010). Simple identification tasks can even be realized directly by comparing the change rules of characteristic peaks.

Chemometrics combined with vibrational spectroscopy techniques is an effective tool that can be understood as applying mathematics and statistics to chemical data processing (Rohman, 2019). Conventional statistics focus on a particular situation based on predefined distributions and model assumptions, not properly applied to every kind of data (Breiman, 2001). Digitized modeling can be improved and perfected continuously to expand the scope of application, building more robust models with higher accuracy.

Authenticity is the primary importance, which is the first step in the whole process of herb production. Herbs that belong to the same genus, even the same family, have a similar appearance. However, the value of herbs from various species is definitely different. The species’ characteristics make it impossible to be used interchangeably. The difficulty of species discrimination causes the phenomenon of counterfeit products and the misuse of raw materials. IR and Raman techniques, the feature bands presented in the spectra can be analysed to identify species.

Species discrimination using NIR reflectance spectroscopy was dated back to 1999 in Ginseng radix et rhizome (Woo et al., 1999b). The identification of F. thunbergia Miq from the genus Fritillaria (Meng et al., 2015), peach and apricot kernels (Kajino et al., 2021), the extracts of Ganoderma lucidum and Vesicolor (Shao et al., 2015), and Eleutherococcus senticosus from other eight herbs (Lucio-Gutiérrez et al., 2011) using NIR spectra was achieved. High discrimination accuracy in Ginseng (Yap et al., 2007) and Lingzhi species (Wang et al., 2019) was obtained based on FTIR combined with ML. SVM correctly discriminated against two species by the MIR and NIR model (Chen et al., 2020a). Gao et al. (2005) illustrated the feasibility of identifying different species of Fritillariae bulbus by convolution transform visualization fingerprint. Detection techniques, applications, and specific data processing methods of different herb species are concluded and listed in Table 3 .

The preliminary conclusion is drawn that the classification accuracy decreases with increasing categories, and the close geographical relationship makes it more challenging to discriminate. There are two main concerns to expanding the acceptance of IR and Raman spectroscopy techniques. One is model transferring capability (Li et al., 2013), and another is the miniaturization of devices (Lin et al., 2020). Chen et al. (2017) compared the benchtop and hand-held FT-IR equipment, pointing out that hand-held spectrometers had a promising prospect.

Geographical origins are the second factor needed to be considered after species. Herbs are directly influenced by growing conditions like climate, soil, and altitude. The quality disparity of herbal raw materials from different origins exists. Where herbs are widely recognized with the highest value is called a geo-authentic area, and consumers appreciate herbs from the geo-authentic area. In this sense, geo-herbalism becomes the comprehensive evaluation criterion of excellent-quality herbs. Identifying the original source is beneficial to ensure quality consistency and avoid counterfeiting. IR and Raman spectroscopy techniques present the complete chemical profiles of herbs, paying more attention to the overall internal components.

Woo et al. (1999b) determined the geographical origins of Astragali radix, Ganoderma, and Smilacis rhizome with NIR reflectance spectroscopy in 1999. In the same year, Woo et al., (1999c) classified the cultivation areas of Ginseng radix et rhizoma using NIR and Raman techniques. From then on, scholars tried to achieve origins discrimination based on IR and Raman. Studies about the origin discrimination of herbs, like Salviae miltiorrhizae radix et rhizoma, Paridis rhizoma, and Notoginseng, have been carried out combined with various pattern recognition methods, including classic algorithms and innovative algorithms (Woo et al., 2005; Li and Qu, 2014; Liu et al., 2015; Han et al., 2016; Hui et al., 2018; Lai et al., 2018; Yang et al., 2018; Zhu et al., 2018; Fu et al., 2019; Yang et al., 2019). Detailed information containing techniques and data processing is listed in Table 3 , and the qualitative analyses of herbal raw materials are summarized. Spectral correlation coefficient and technique for order preference by similarity to ideal solution (TOPSIS) method were used to evaluate the quality from different producing areas of Cordyceps to find the most suitable growing region (Sun et al., 2019). Non-medicinal parts can also be used for origin identification. Different botanical parts of Gentianae radix et rhizoma were compared, and researchers found that leaves were the optimal material for geographical characterization (Wang et al., 2018; Liu et al., 2020). The findings illustrate the differences between medicinal and non-medicinal parts at the spectrum level.

The evolution of equipment produces more possibilities in classification. From the articles we searched, the prediction accuracy of IR or Raman spectroscopy techniques reached an acceptable level compared with chromatography methods (Chen et al., 2008; Wang et al., 2020). With unique advantages in heterogeneous sample detection due to the depth-profiling function, FTIR-PAS was first employed in Cordyceps sinensis, coupled with PNN (Du et al., 2017). Portable spectrometers are developing. FT-NIR and MicroNIR spectrometer succeeded in identifying four origins of Salvia miltiorrhiza. MicroNIR spectrometers had worse performance due to limited spectral information (Sun et al., 2020).

The grade of herbs is further subdivided according to their quality discrepancies, which depend on their growing conditions (wild-grown or cultivated, cultivation age, etc.) or parts. The difficulty of grade classification lies in the current form of herbs. The herbal raw materials are processed into powder before they enter the markets. Gad and Bouzabata (2017) failed to discriminate Turmeric powder bought from different commercial stores with various grades using FTIR. The possible reason is FTIR spectra aren’t sensitive to the same species whose phytochemical constituents are the same but in different concentrations.

Parts discrimination against Ginseng is meaningful from both academic and commercial points of view. The determination of powdered products is still a problem needed more improvements. DR-NIR spectra were employed to classify different parts of Ginseng powder, considering the granularity of the powder. ATR-FTIR spectra were analysed to reveal the difference from molecular functional groups, whose score plots of PCA disclosed a regular and gradual difference in each part (Wu et al., 2011). After suitable normalization methods, ATR-FTIR spectra showed potential in this aspect (Lee et al., 2017). To distinguish wild-grown and cultivated herbs, penalized discriminant methods (Zhu and Tan, 2016) and Adaboost M1 algorithm (Chen et al., 2021) were proposed, both of which had higher computational efficiency and classification accuracy after data pre-processing and variables selection.

Low-quality herbs can be pretended to be high-quality when the limited supply cannot meet the increasing demand. Driven by benefit, illegal traders disobey laws to make counterfeits. The adulteration of herbs involves intentionally adding other low-cost or non-pharmaceutical raw materials with a similar appearance to replace or remove certain ingredients without the buyers’ knowledge. The abuse of adulteration leads to severe problems, such as unfair trade competition, public health risks, and social issues. Promising analytic methods are needed to identify the counterfeits, which avoids commercial fraud and guarantees medicine safety.

The application of vibrational spectroscopy techniques in adulteration detection is summarized in Table 3 . The functional group regions (4000-1300 cm^-1) have better capability to detect authentic Astragali radix (Zhang and Nie, 2010). The inferiority of unsupervised methods as indicated in either MIR or NIR spectroscopy (Liu et al., 2019; Yang et al., 2020) due to poor capability to extract effective information. Data pre-processing, variables selection, IR spectroscopic tri-step identification approach (Chen et al., 2015; Chen et al., 2018), and more supervised algorithms (Shao et al., 2016; Chen et al., 2019; Zhao et al., 2019), which aimed at reducing uncorrelated spectral information, were studied to reach satisfactory results, from adulterated binary samples to adulterated quaternary samples (Nie et al., 2013; Liu et al., 2019). We notice that it is difficult to transfer the models based on small samples directly to other samples because of the limitations of the representativeness of small samples and analytical techniques, as well as the various presence of adulterated chemical components (Li et al., 2020).

A silver nanoparticle wiper as a SERS substrate based on filter paper was made to distinguish nine kinds of dyes adulterated in herbs (Dan et al., 2015). The gold nanorods SERS-based approach functionalized with mono-6-thio-cyclodextrin (HS-β-CD) enhanced the detection capability by strengthening the chemistry interactions (Wu et al., 2018). The fabricated substrate and chemometrics methods (Lin et al., 2018) can improve detection sensitivity. Furthermore, Applying the portable substrate and Raman spectrometer is anticipated to achieve in situ detection (Liu et al., 2018).

As we refer to above, adaptive models’ establishment faces challenges in the complex composition of adulteration. In practical application, we need to judge whether the herbal medicine is adulterated and what the impure substance is the next step. Hence, we recommend developing untargeted identification that is advantageous for solving authentication problems (Li et al., 2017; Chen et al., 2020a). Expansion of the samples’ scales and optimization parameters of models are effective means to learn more features from the target class.

The influence of the factors mentioned above on the quality of herbs can be basically reflected in critical quality attributes. Critical quality attribute detection, linked to efficacies, has been subjected to more application prospects. The significance of detection is to ensure that the quality of herbs meets the standards for entering the market or has a uniform content consistent in the manufacturing process. We divide critical quality attributes into three categories, active medicinal ingredients, bioactive components, and other regulated indices (moisture, ash, etc.).

The ingredients in herbs, as one of the qualitative evaluation indexes, are listed in the pharmacopeia. Conventional analysis methods require strict extraction and purification. Other regulated indices that illustrate quality, as well as purity, need complicated and laborious operations. In many cases, IR and Raman spectroscopy techniques have been applied successfully, regarded as green and rapid technologies without reagent contamination, which is practical for achieving digital detection.

More than 80% of the studies adopted NIRS, demonstrating the advantages of NIRS in multi-component quantitative detection. PLS is the most popular in multi-component quantitative detection because it can reveal information for the dependent variable as well as reduce the dimensions of the spectral matrix. NIRS-PLS model was applied successfully in the prediction of the total ash and acid-insoluble ash of Prunellae spica (Rao and Xiang, 2009), glycyrrhizic acid of Puerariae lobatae radix (Mohri et al., 2009), active medicinal ingredients of Astragali radix (Zhan et al., 2017), Paeoniae radix alba (Luo et al., 2008), Amomum villosum (Guo et al., 2021), Morindae officinalis radix (Hao et al., 2020), Dipsaci radix (Du et al., 2017), and Notoginseng (Chen and Sørensen, 2000). These studies also further explain the factors that affect the content of the detection indexes. But sometimes NIRS-PLS caused over-fitting and low precision using full-wavelength spectra (Yang et al., 2003; Yan et al., 2020). The possible reasons could be (1) a limited number of samples. The range of component content distribution is narrow, which results in those differences among various quality are not significant enough to train the robust linear models. (2) low concentration of target components. NIRS proved unsuitable for content lower than 0.1%. Mid-infrared spectroscopy (MIRS) was regarded as a better predictor for analysing low concentration and NIRS utilized the complementarity (Liu et al., 2020).

Due to the fingerprint regions of MIRS and Raman spectroscopy, detection can also be achieved by analysing characteristic peaks. Pei et al. (2008) used the correlation analysis in Epimedii folium based on the peak at 1259 ± 1 cm^-1. The curcumin weight ratio formula was put forward based on band intensity ratios of Raman spectroscopy, analysing different layers of turmeric roots (Peng et al., 2015). The peak intensity at 727 cm^-1 of berberine was observed in Coptis chinensis and Phellodendron amurence using SERS (Zhao et al., 2014). TLC-SERS captured the detectable signals, Raman intensity (I₇₀₈/I₇₂₈), which served as the evaluation index in Coptidis rhizoma to discriminate and determine berberine and coptisine (Gu et al., 2018).

Table 4 displays a series of studies regarding quantitative detection that has been carried out by the IR and Raman spectroscopy techniques. The information about techniques, target ingredients, and data processing of concrete herbal plants is available in Table 4 . Bioactive components, like polysaccharides (Chen et al., 2012; Bu et al., 2013; Ma et al., 2018), flavonoids (Lau et al., 2009; Arslan et al., 2018), alkaloids (Chan et al., 2007; Qi et al., 2018), and antioxidant activity (Wong et al., 2015; Yi et al., 2020), extracted from herbs contribute to sensory quality and efficacy, evaluated by IR and Raman spectroscopy techniques. Pre-treatment, feature selection, non-linear regression methods, and data fusion were tried to improve prediction results. The SVM model presented good generalization performance for Epimedii folium, with its R² of more than 0.9 after extracting the feature wavelengths by GA (Yang et al., 2017a). The models built by PLSR and ANN were compared to predict the medicinal ingredients in rhubarb samples (Xue et al., 2018) and Lonicerae japonicae flos (Jintao et al., 2021), concluding that models preferred for different components were not the same. There is no doubt that the use of portable spectrometers promotes the application of IR (Wang et al., 2017) and Raman spectroscopy (Wong et al., 2015).

In short, different strategies should be selected according to purposes and requirements in production practice (Yan et al., 2021). Spectroscopic techniques have the outstanding advantage of simultaneous multi-component analysis. The rapid determination of herbal raw materials using IR and Raman spectroscopy techniques has high practical value in the pharmaceutical industry. Meantime, these techniques provide reliable technical support for the evaluation of the critical quality attributes and on-line measurements during the production process.

Knowing the index components of PHPs can evaluate the quality and provide a reference for dosage. Spectroscopic techniques are applied as rapid, non-invasive methods, requiring minimal sample pre-treatment. The successful distinguishing of thirty-six commercial brands of Ganoderma lucidum (Sun et al., 2001) and consistent characteristic peaks in PHPs compared with individual herbs (Bansal and Reddy, 2018) based on FTIR illustrated the evaluation can be done. Chen et al. (2016) adopted FTIR microspectroscopic imaging to collect pixel spectra. The direct and simultaneous recognition of multiple organic and inorganic ingredients in PHPs was achieved by comparing the reference spectrum and calibration set.

NIRS-PLS model was employed to determine two different sample presentations originating from a turmeric capsule and powder, obtaining ideal results in powder samples (Kasemsumran et al., 2014). The concentration of Coptis chinensis in suppositories was also predicted (Teraoka et al., 2012). Three presentations, capsule shells, contents, and intact capsules, of Yaobitong capsule, were analysed using NIRS-LSSVM model (Si et al., 2021). MIRS-PLS model was used to predict feeding levels of PHPs by detecting the excipient content (Guo et al., 2016) as well as the antioxidant activity of mixed herbal infusions (Venetsanou et al., 2017). Laser-induced breakdown spectroscopy (LIBS) with element information and MIRS with molecular information was fused to classify the compound Salvia miltiorrhiza by RF discrimination models (Liang et al., 2020).

Because the predictive performance of the model will be affected by samples (formula, batch, manufacturer, etc.) and algorithms (pre-treatment, feature extraction, modeling), the scope of the calibration set should cover the test set. Therefore, the robustness of models requires sufficient representative samples, still long-term research work.

PHPs have milder effects than western medicine in a slow curative effect with fewer side effects. Demand is increasing due to the growing popularity of herbal dietary supplements and clinical medicines. Synthetic drugs and regulated or toxic substances are added undeclared to PHPs for illegal profits, which results in consumers being vulnerable to counterfeiting and adulteration.

NIRS coupled with PLS-DA distinguished PHPs adulterated with sibutramine with a correct classification of 100%. Four variables selected by MLR-SPA were executed to build a quantitative model (Da Silva et al., 2015). Feng et al. (2014) improved the reverse correlation coefficient (RCCM) for threshold settings to test antidiabetic PHM illegally added with synthetic drugs. MIRS presents a ‘fingerprint’ with high sensitivity and selectivity in terms of sample peaks and peak intensities. MIRS showed the best performance among MIRS, NIRS, and Raman spectroscopy techniques in detecting PHM adulterated with sibutramine and phenolphthalein (Rooney et al., 2015). The performance of fused data using MIRS and NIRS was poorer than MIRS data alone (Deconinck et al., 2017). NIRS data failed to add valuable information according to the loading analysis.

Observing and comparing characteristic peaks in MIRS and Raman spectroscopy is also effective means of identification due to the significant differences between PHPs and adulterated products (Mateescu et al., 2017). Slimming herbal products, adulterated with illegal additives, were discriminated against by constructing synchronous and asynchronous maps (Miao et al., 2017). Univariate calibration (De La Asunción-Nadal et al., 2017) and mathematically fortified spectra (Walkowiak et al., 2018) based on ATR-FTIR offer a fast, eco-friendly, and cheap alternative for adulterations identification with good analytical features. The local straight-line screening (LSLS) algorithm, newly proposed in 2007 and modified in 2009, has proved the feasibility of detecting the illegal incorporation of synthetic drugs in PHPs after careful observation of the shape of the spectral line (Lu et al., 2007; Zhu et al., 2009). TLC-SERS was established and used to detect adulterated PHPs for curing diabetes (Zhu et al., 2014), cholesterol (Zhu et al., 2017), and sexual performance (Dao et al., 2019).

Current microscopic and chemical identification above showed the feasibility and application prospect of IR and Raman spectroscopy techniques in the quality and safety of PHPs. Various PHPs can be detected without complex procedure extraction of marker compounds. Vibrational spectroscopy techniques can be applied as a preliminary evaluation of suspicious PHPs, even though their current LODs are inferior to traditional chromatographic methods. The identified PHPs are confirmed using specific chromatographic methods afterward. Spectroscopy analysis methods are expected to be widely applied if the entire experimental procedure can be optimized, standardized, and automated. The development trend of small-type and portable spectrometers makes mobile laboratories feasible, conducted in the open market and throughout the herb distribution channel.