Qualitative and quantitative analysis of human hair in wig products by near-infrared spectroscopy

Abstract

With the increasing demand for wigs in the fields of beauty, fashion, healthcare, and theater, the requirements for wig quality are also rising. Among them, human hair wigs are the most popular. In pursuit of profit, many merchants counterfeit human hair wigs with substandard products. The methods for human hair identification, such as microscopic observation, DNA analysis, and protein analysis, are almost exclusively targeted at forensic applications. These methods are not suitable for the detection of human hair in wigs, as the hair undergoes certain pre-treatments during the process of making wigs. To address the challenges in human hair identification, we developed a method based on Near-Infrared Spectroscopy (NIR) combined with chemometrics for the qualitative and quantitative analysis of human hair. Statistical methods such as similarity matching, qualification testing, Principal Component-Mahalanobis Distance Discriminant Analysis (PCA-DA), partial least squares-discriminant analysis (PLS-DA), and partial least squares regression (PLS) were employed to analyze the NIR data and establish multiple qualitative and quantitative models. The validation samples were substituted into these models and yielded accurate test results, demonstrating the feasibility of this method. The research results indicate that this method can be applied for the identification and quantitative analysis of human hair in wig products. Our research provides a rapid, convenient, and non-destructive method for the qualitative and quantitative analysis of human hair. This method is effective in combating counterfeit human hair wig products in the market.

Keywords

NIR human hair animal hair wig PCA PLS-DA

Introduction

The manufacturing materials for wigs typically fall into three categories: human hair, animal hair, and chemical fiber. The type of material used for wigs determines their quality, price, and wearability. Generally, human hair is the most expensive, with animal hair coming next, and chemical fiber being the cheapest. As the pursuit of beauty increasingly intensifies, wigs have evolved from a solution for hair loss to a fashionable accessor.¹ With the growth of the wig industry, there is a large amount of non-human hair products on the market that are disguised as human hair. Some low-quality wigs can even cause skin allergies.² However, due to the relatively small size of the wig industry in the past, there has been a lack of comprehensive methods for identifying wig materials. Wig products, similar to textiles, can be considered a special type of textile. A large amount of textile fibers is also used in wig products. In the past, the testing of wig materials typically refers to the testing standards for textile fibers, such as ISO 1833, AATCC 20, AATCC 20A, etc. However, the testing methods in these standards are only effective for the detection of chemical fibers and are not applicable for human hair or animal hair. There is no human hair in textiles, and there are no testing methods for human hair. Therefore, there is an urgent need to develop a comprehensive and effective testing method for wig materials.

Both human hair and animal hair are protein fibers, sharing similar chemical and physical properties. Therefore, some identification methods are applicable to both. These methods include microscopy observation, DNA analysis, RNA analysis, protein analysis, and spectroscopy analysis. However, the identification of human hair is mostly applied in the forensic field for evidence analysis and individual identification at crime scenes, while the identification of animal hair is primarily used in the textile industry for qualitative and quantitative analysis of fibers. Microscopic observation is a traditional method for fiber identification, which includes optical microscopy³ and scanning electron microscopy (SEM).⁴ These methods distinguish different types of hair fibers by comparing their morphological characteristics, such as diameter, tips, roots, and the scale layer. Based on these morphological and textural features (such as textural parameters extracted using the Gray-Level Co-occurrence Matrix, GLCM), feature fusion technology has been developed.⁵ This technology is typically combined with image recognition, machine learning, and chemometrics to achieve automated identification. During the wig-making process, acid treatment is often applied to achieve a better fitting effect. This process destroys the scale layer structure and alters the morphological features, rendering these detection methods ineffective.

DNA identification is a method based on the uniqueness and diversity of DNA molecules. DNA is extracted from the sample to be tested, and the target DNA fragments are amplified using the Polymerase Chain Reaction (PCR) technique for identification analysis.^6,7 Due to the instability of DNA and the low DNA content in many samples, traditional DNA analysis techniques are often not applicable. Therefore, RNA analysis and protein analysis have been developed as alternative approaches. RNA analysis, based on the polymorphism of RNA and combined with MALDI-TOF MS (Matrix-Assisted Laser Desorption/Ionization Time-of-Flight Mass Spectrometry), has been used for the identification of hair shafts.^8,9 Hair fiber in wig products mostly lacks hair follicles and contain minimal DNA and RNA, making DNA analysis and RNA analysis difficult to apply for identification.

Protein analysis has a broader range of applications than DNA analysis, due to the greater chemical stability of proteins compared to DNA. The molecular weight of proteins is very large, making direct analysis impossible. They must first be decomposed into smaller peptides before mass spectrometry can be used to identify differences and distinguish between different species. A large number of chemical reagents are used in this decomposition process. Liquid Chromatography/Electrospray Ionization Mass Spectrometry (LC/ESI-MS), MALDI-TOF MS, and Liquid Chromatography-Mass Spectrometry (LC-MS) are commonly used for protein analysis to perform qualitative and quantitative analysis of test samples.^10–12 Fei¹³ developed a MALDI-TOF MS-based protein analysis method for the identification of human hair in wig products.

ATR-FTIR (Attenuated Total Reflectance Fourier Transform Infrared Spectroscopy) and NIR are both used spectroscopic techniques. Although they are based on different spectral regions—with ATR-FTIR covering the mid-infrared region of 500–4000 cm⁻¹ and NIR covering the near-infrared region of 3500–12,500 cm⁻¹—they share similarities in principle. Both methods acquire chemical information by detecting the absorption or reflection of specific wavelengths of light by the sample. They typically employ multivariate statistical methods (such as Principal Component Analysis, PCA and Partial Least Squares, PLS) to analyze the spectral data, extract key information from the samples, and establish a correlation model between the spectral data and the sample types.¹⁴ Ultimately, these models are used to classify and identify unknown test samples. ATR-FTIR is still at its initial stages and is commonly used as a preliminary screening method.^15–17 NIR has been widely used in chemical industry,¹⁸ agriculture¹⁹ and the food industry,^20,21 such as analyzing differences in origin, and has developed a set of rapid and effective testing methods. There has been substantial research on the use of NIR for the identification of animal hair.^22–24 Yang Qiao²⁵ employed NIR to distinguish between human hair and animal hair in wigs, confirming the feasibility of using NIR for the identification of human hair. Our work differs from previous studies in two main aspects. First, we have developed a more extensive set of models for distinguishing between human hair and animal hair, with a larger sample size, resulting in more accurate and robust models. Second, we have introduced qualitative models for differentiating hair from chemical fibers, as well as quantitative models for human and animal hair. This led to the creation of a comprehensive method for human hair identification, allowing us to independently determine the content of human hair in wig products using our approach. Yang Qiao’s research can only authenticate products that have been proven to be made solely from human hair or animal hair. Moreover, it is unable to conduct quantitative analysis of human hair.

For products like wigs, which are expensive, come in a wide variety of styles, and are produced in relatively small quantities for each style, the NIR method, which is rapid and non-destructive, is particularly well-suited for identification. Compared with the microscopic observation method, NIR is not affected by the microscopic physical form of fibers during the detection process, resulting in more accurate analytical results. Compared with DNA analysis and protein analysis, the instruments used in NIR are more affordable, and there is no need for sample pretreatment or the use of chemical reagents. These factors greatly reduce the cost of detection.With the advent of portable near-infrared spectrometers, sample detection can be completed on-site, eliminating the need to bring samples back to the laboratory. NIR is well-suited for regulatory authorities to standardize the wig market and protect consumer rights. This method can also be explored for application in human hair identification within the forensic field.

In our work, focusing on human hair in wigs as the research target, multiple near-infrared models were established using the NIR method to achieve the qualitative and quantitative analysis of human hair. In wigs, animal hairs commonly used include camel, yak, and sheep hairs. Since there’s no need to specifically identify the animal species in testing, these hairs were grouped into a single category for distinction from human hair. Both human hair and animal hair were fibers made of proteins and share similar chemical and physical properties. In our research, they were sometimes collectively referred to as “hair”. Using NIR, a qualitative hair model was first established for the identification of hair, which excludes chemical fibers. Then, a qualitative model was developed to differentiate between human hair and animal hair. Finally, a quantitative model for human hair was established to determine the concentration of human hair. Collect the near-infrared spectrum of the sample to be tested, and by inputting the spectrum into these established models, it was possible to identify whether it is human hair and calculate the content of human hair.

Experimentals

Materials

In our study, 247 human hair samples included hairpiece products, hair patch products, and raw materials. The 247 animal hair samples consisted of 80 yak hairs used in wig production, 80 camel hairs, and 80 sheep wools. Chemical fibers included polyester (PET), nylon (PA), polyacrylonitrile (PAN), polypropylene (PP), polyvinyl chloride (PVC), with 50 samples of each. There were 176 mixtures of human hair and animal fur in various proportions, with mixing ratios evenly distributed from 0% to 100%.

NIR spectra acquisition

An FT-NIR spectrometer (Bruker, MPA, Germany) equipped with an integrating sphere (IdentiCheck Reflectance Accessory) for reflectance was used. For each spectrum, 32 scans were recorded in the spectral region between 12,500 and 3500 cm⁻¹. The Unscrambler X 10.4 (CAMO Software AS, Oslo, Norway) was used for preprocessing of spectral data and the development of analytical models.

Modeling

Hair qualitative model

In NIR modeling, the representativeness of samples is a key factor in ensuring the accuracy and robustness of the model. Representative samples can reflect all the characteristics and variability of the subject under study, thereby enhancing the model’s predictive and generalization capabilities. Representativeness is mainly reflected in the diversity of samples, which can be demonstrated through differences in origin, processing methods, batches, and manufacturers. The sources of human hair raw materials include China, India, and Vietnam. The raw materials for animal hair are sourced from different provinces in China. The processing methods for human and animal hair can be categorized into untreated, acid-treated, and dyed. Chemical fibers are sourced from multiple manufacturers and production batches.

Classify hair (human hair and animal hair) and all chemical fibers into two distinct categories, and construct a qualitative model to differentiate hair types by analyzing the near-infrared spectral differences between these categories. 494 hair samples (247 human hair samples and 247 animal hair samples) were randomly divided into two parts at a ratio of 3:1. Among them, 371 samples were used as calibration samples to establish a calibration model, while the remaining 123 samples and 150 chemical fiber samples were used as validation samples to verify the accuracy of the model and ensure its usability. There are two methods used to establish a qualitative model for the hair, which are the similarity matching method and the conformity testing method.

In the similarity matching method, the spectral data were preprocessed using Standard Normal Variate (SNV) transformation and first derivative method in the range of 6500 to 4000 cm⁻¹. SNV is a preprocessing method that involves subtracting the mean value of a spectrum from the original spectral data and then dividing by the standard deviation of that spectrum to obtain standardized spectral data. It is used to reduce spectral errors caused by scattering effects among calibration samples. The first derivative is calculated by performing a first-derivative computation on the spectral data, which enhances spectral features and reduces baseline drift. The similarity matching threshold was set at 95. If the similarity matching value (SMV) was greater than 95, it was determined to be human hair. If the similarity matching value was less than 95, it was determined to be synthetic fiber.

In the conformity testing method, the spectral data were preprocessed by the second derivative method in the range of 6500 to 4000 cm⁻¹. The second-order derivative is obtained by performing a second-derivative calculation on the spectral data, which enhances the apparent resolution of the spectrum and helps to resolve overlapping peaks. The conformity index limit value was set at 8; if the Conformity Index (CI) was less than or equal to 8, it was determined to be a hair fiber, and if CI was greater than 8, it was determined to be synthetic fiber.

Human hair qualitative model

Human hair and animal hair were categorized as two distinct groups. A qualitative model for hair identification was established by calculating the differences in their near-infrared spectra. A total of 494 hair fibers (247 human hair samples and 247 animal fur samples) were randomly divided into calibration samples and validation samples in a 3:1 ratio. Two methods were used to establish a qualitative model for human hair to distinguish between human hair and animal hair, which are Principal Component-Mahalanobis Distance Discriminant Analysis (PCA-DA) and Partial Least Squares-Discriminant Analysis (PLS-DA).

Principal Component Analysis (PCA) is one of the most widely used tools for data dimensionality reduction. It involves decomposing the spectral data matrix of the samples to generate new variables, known as principal components, thereby reducing the dimensionality of the spectral data.²⁶ By projecting onto the first two or three principal components (those with the largest eigenvalues), a large amount of interfering information can be effectively filtered out. Replacing the original spectral data with the principal component score matrix obtained through calculation significantly reduces the number of computational variables. Subsequently, discriminant analysis can be performed using these principal components.^27,28 The most commonly used principal component-based discriminant method is PCA-DA. This method sets a specific threshold based on the centroid of the principal components and the deviation of the samples. It assigns different categories to fixed spaces. When a test sample falls into a corresponding space, it is classified as that category. If it does not fall into any space, it is determined that the sample does not belong to any of the defined categories. The ability to distinguish between samples can be clearly observed from the principal component score plot. Similar samples will cluster together, while samples with greater differences will have distinct principal components. If the principal component regions corresponding to different classes do not overlap, this indicates that the samples can be distinguished. The greater the distance between them, the better the separation effect is. In PCA-DA, after the spectra were preprocessed with the second derivative method in the range of 6500 to 5500 cm⁻¹and 5000 to 4500 cm⁻¹, principal component regression analysis was conducted to create a multidimensional space composed of principal components. The principal component values for each category will cluster together in this multidimensional space, while values from different categories will be distributed further apart. The Mahalanobis distance of the samples in the model was calculated to determine their classification into respective categories.

PLS-DA is a well-known linear classification technique that integrates the feature extraction capability of PLSR (Partial Least Squares Regression) with the discriminative power of classification methods. It can be used for descriptive modeling, predictive modeling, and the selection of discriminative variables, thereby generating multiple outcomes. In PLS-DA, the reference value for human hair samples was set to 1, and the reference value for animal fur samples was set to 0; they were the categorical variables. After the spectral data was preprocessed with the first derivative in the range of 6500 to 5500 cm⁻¹ and 5000 to 4500 cm⁻¹, it was subjected to partial least squares regression analysis with the categorical variables. The specific criteria for classification and discrimination are as follows: if the estimate value Y > 0.5 and the deviation was less than 0.5, the sample was determined to be human hair; if Y < 0.5 and the deviation was less than 0.5, the sample was determined to be animal hair; if the deviation was greater than 0.5 the result was inconclusive and the sample was not assigned to either category. The decision coefficient (R²) of the PLS-DA model should not be lower than 0.9, indicating a high degree of variance explained by the model. The R² is an important metric used to measure the goodness of fit of a model to the data. It reflects the proportion of the variance in the dependent variable that can be explained by the independent variable. The closer R² is to 1, the better the model fits the data. The root mean square error of cross-validation (RMSECV) or the root mean square error of prediction (RMSEP) should not exceed 0.5, which suggests that the model has a good predictive accuracy with small prediction errors. RMSECV is the standard error calculated from the reference values of the calibration samples and the predicted values obtained using cross-validation. RMSEP is the standard error calculated from the reference values of the validation samples and their predicted values. The accuracy rate of the discriminatory analysis for the validation samples should not be lower than 95%.

Human hair quantitative mode

The human hair quantitative model was used for analyzing the content of human hair in mixtures with animal hair. The prepared 176 mixtures of human hair and animal hair were randomly divided into calibration samples and validation samples in a ratio of 3:1. A composite method combining first derivative and multiplicative scatter correction (MSC) was used to preprocess the near-infrared spectral data of the calibration samples. MSC is a method that corrects measured spectra to a reference spectrum, thereby reducing these irrelevant variations. It is used to eliminate baseline drift and amplitude changes in spectra caused by scattering effects and differences in particle size. The data within the wavelength range of 7550 to 5440 cm⁻¹ was selected and used for PLS regression analysis with the reference values of the calibration samples to establish a quantitative model. The R² of the model should be no less than 0.9, and RMSECV or RMSEP should not exceed 5.0.

Results and discussion

NIR spectra analysis

NIR spectra of hair samples and chemical fibers were characterized. Human hair and animal hair spectra are very similar (see Figure 1(a)), which allows us to categorize both human hair and animal fur as one group, namely “hair”. This similarity can be utilized in spectroscopic analysis to develop models that can identify and quantify hair materials without distinguishing between human and animal origins. The secondary harmonic absorption peak of C-H is located at 7400–8800 cm⁻¹, with relatively weak absorption. The first harmonic peak of O-H stretching vibration is at 7000 cm⁻¹. The combination of O-H stretching and H-O-H bending vibrations of water molecules, which are characteristic absorption peaks of water, is found at 5200 cm⁻¹. The overtone vibration of C-H in protein side chains is at 5800 cm⁻¹. The amino acid composition of keratin and the spatial conformation information of macromolecules are mainly concentrated in the range of 5000–4000 cm⁻¹. These spectral data are consistent with previous related research.²⁹ The different near-infrared absorption bands of various functional groups help us select the appropriate spectral range for modeling. The spectra of hair differ significantly from those of chemical fibers (see Figure 1(b)). Hence by employing methods such as similarity matching or conformity testing, one can analyze the similarity between the spectral characteristics of test samples and human hair spectra to establish a qualitative model for human hair. The similarity matching method and the conformity testing method are commonly used for the control and authenticity assessment of pharmaceutical quality.^30,31 If the drug is contaminated with other impurities, it can be rapidly identified that the drug quality does not meet standards. Similarly, in the qualitative analysis of hair, if a sample is made of chemical fibers or contains chemical fibers, it can be quickly determined that the tested sample is not pure hair. PCA-DA and PLS-DA are more effective for distinguishing categories with high similarity.

Figure 1.

NIR spectra. (a) Hair (b) chemical fibers.

Hair qualitative model

A qualitative model for hair identification was established by similarity matching method. A total of 123 hair samples and 150 chemical fiber samples were used to verify the effectiveness of this model. The similarity matching method is a technique based on the Gram-Schmidt vector orthogonalization process for an overall evaluation of similarity.³² Through this method, the SMV between the test spectrum of an unknown sample and the reference spectrum can be calculated; the higher the value, the greater the similarity. The SMV threshold is typically set to the minimum SMV value of the calibration samples. However, to ensure that all validation samples are correctly identified, the SMV threshold can be set slightly lower. In this study, the SMV threshold was set at 95. Figure 2 shows the validation results of the model established by the similarity matching method. The SMV of each hair verification samples was greater than 95, while the SMV of each chemical fiber verification samples was much less than 95. The significant gap in SMV indicates that there is a substantial difference in the chemical composition between chemical fibers and hairs, making them easy to distinguish. The verification sample identification accuracy rate was 100%. The qualitative model for hair established using the similarity matching method was effective.

Figure 2.

The validation results of the model established by the similarity matching method.

A qualitative model for hair identification was established by conformity test method. The principle of the conformity test method involves calculating the average value and standard deviation for each wavelength point in the reference spectrum set, using the average value plus or minus a certain multiple of the standard deviation as the control range for that wavelength point. If the absorbance of each wavelength point in the test spectrum does not exceed the predetermined control range, it indicates that the test sample was similar to the reference sample. The CI was obtained by dividing the difference between the absorbance of the test sample and the average value by the average deviation. The setting method of the CI limit value is similar to the SMV threshold. The CI limit value is usually set to the maximum CI value of the calibration samples. Similarly, to ensure that all validation samples are correctly identified, the CI limit value can be set slightly higher. Figure 3 displays the validation results of the model established by the conformity testing method. The CI of each hair verification sample was less than 8, and the CI of each chemical fiber verification sample was significantly greater than 8. The verification sample identification accuracy rate was 100%. The qualitative model for hair established using the conformity testing method was also effective.

Figure 3.

The validation results of the model established by the conformity testing method.

During the acquisition of near-infrared spectra, it is inevitable that some interference information unrelated to the effective information of the sample to be measured is included. The presence of these interferences, to a greater or lesser extent, affects the analysis of the data and the establishment of the model. To mitigate or even eliminate various interferences on the spectrum, appropriate methods are typically employed to conduct mathematical preprocessing of the spectral signals. The aim is to eliminate as much as possible the noise and background fluctuations in the spectrum that are unrelated to the attributes of the variables to be measured, with the expectation of establishing more robust models and enhancing predictive accuracy. In this study, the SNV, MSC, and first derivative used are all common preprocessing methods. The choice of these preprocessing methods in the process of model building is not fixed and can be varied, as long as the established model is valid.

Human hair qualitative model

The human hair qualitative model was established by PC-DA, which can be used to identify whether a sample is human hair or animal hair. Figure 4 is the principal component score plot of the PCA-DA model, where samples of the same category cluster together, and those of different categories were separated from each other. There is a clear boundary between human hair and animal hair, with no overlapping regions, indicating that these two categories can be distinguished from each other. After PCA, the contribution of each variable to the principal components was calculated (see Figure 5); the greater the deviation from 0, the greater the contribution. Therefore, the loadings plot was used to select an appropriate spectral range for model building. The contributions to PC1, PC2, and PC3 were mainly concentrated in the range of 4000–6000 cm⁻¹, with the characteristic absorption peaks of water molecules focused between 5000 and 5500 cm⁻¹. The moisture regain rate of human hair and animal hair is very high and is easily affected by environmental changes, so this range was avoided as much as possible during the modeling process. The range of 6500–5500 cm⁻¹ and 5000–4000 cm⁻¹ was selected. The selection of principal components for discriminant analysis is of great importance. If too few principal components are selected, the model may fail to fully utilize the available information, leading to underfitting and a decrease in its discriminant ability. Conversely, if too many principal components are chosen, it may introduce excessive noise, causing overfitting of the model. The optimal number of principal components is typically determined by observing the changes in the sum of squared prediction residuals as the number of principal components increases. When further increases in the number of principal components no longer result in significant prediction residuals, the current number of principal components is considered the best choice. As shown in Figure 6, when the number of principal components reaches 7, the prediction residuals hardly change anymore, indicating that the optimal number of principal components is 7. Figure 7 shows the results of all calibration and validation samples calculated by the established PCA-DA model. Human hair samples and animal hair samples were distributed in two distinct areas, with calibration and validation samples of each category located within the same area, indicating that the discrimination results for all validation samples are correct.

Figure 4.

The principal component score plot of the PCA-DA model.

Figure 5.

PCA loadings plot.

Figure 6.

Changes of prediction residuals.

Figure 7.

The results of all calibration and validation samples calculated by the PCA-DA model.

PLS-DA is a statistical analysis method that integrates principal component analysis and multiple linear regression analysis.³³ It utilizes partial least squares to extract the main components of the samples, decomposing both the spectral matrix and the category matrix to maximize the differences between the spectra of different classes. Therefore, PLS-DA typically yields better classification and discrimination results than PCA-DA,^34,35 and the same spectral range as PCA can be selected for use in PLS-DA. In the PLS-DA model, the PLS regression analysis of the calibration samples in relation to the categorical variable reveals a very good correlation between them (as seen in Figure 8), with a R² of 0.9761, a RMSEC of 0.1094, and a RMSEP of 0.1192, indicating that the model has a very good fit. All estimate values of the human hair verification samples were close to 1, and all the estimate values of the animal hair verification samples were close to 0 (see Figure 9). All verification results for the samples were correct.

Figure 8.

Estimate values versus reference values in PLS-DA.

Figure 9.

Estimate values of verification samples.

Human hair quantitative mode

All samples used to establish the quantitative model were self-prepared. To ensure thathuman hair and animal fur were mixed as evenly as possible, both human hair and animal hair were cut into pieces smaller than 2 cm. The schematic diagram of the relationship between the estimate values and reference values of both calibration samples and validation samples was shown in Figure 10. The R² has reached 0.9898, indicating that the model fits well. The RMSECV is 3.06, and the RMSEP is 3.49. The closeness of these two values indicates that the model has very good stability. The established quantitative model can accurately test the content of human hair in the mixture. For wig samples containing some chemical fibers, the quantitative method for textile fibers can be referenced. Chemical fibers can be removed and their content calculated, then the remaining hair can be subjected to quantitative analysis to test for the content of human hair, and finally, the content of human hair in the test sample can be calculated.

Figure 10.

Estimate values versus reference values in PLS.

Conclusion

This study used NIR to establish multiple models, achieving the qualitative and quantitative analysis of human hair in wig samples. Using similarity matching method or conformity testing method to establish a hair qualitative model allows for the testing of whether the sample is hair and the exclusion of interference from chemical fibers. Using PCA-DA or PLS-DA to establish a human hair qualitative model enables the differentiation between human hair and animal hair, and testing to determine whether the hair comes from a human or an animal. All the test results for the validation samples of the above quantitative models were correct, indicating that all the modeling methods are effective. The content of human hair in a mixture with animal hair can be detected using a human hair quantitative model established by PLS method. The spectral ranges and preprocessing methods selected for all models are not unique, nor are they necessarily the best. As long as the established model can accomplish the qualitative and quantitative analysis of human hair, other spectral ranges and preprocessing methods can also be chosen. In conclusion, NIR is an effective method for both quantitative and qualitative analysis of human hair in wigs.

NIR is a rapid, non-destructive, simple, and environmentally friendly method. These characteristics imply that the cost of detection using this method is lower. The entire wig industry can spend less on maintaining the quality of wig materials, which helps to reduce market chaos. Compared with other testing methods, testing institutions are more willing to use NIR, which is conducive to making this method a standard across the industry. When consumers have doubts about the material of the product they purchased, the testing personnel can use this method to quickly identify the material of the wig product and provide test results within minutes, without causing any damage to the product. The market supervision department can use this method to conduct on-site random inspections and tests on wig samples and provide test results on the spot, which greatly improves efficiency. The market supervision department can use this method to conduct on-site random inspections and tests on wig samples and provide test results on the spot, which greatly improves efficiency. While this study provides a rapid method for identifying human hair, the method still has certain limitations. It is not suitable for the identification of an individual fiber, as current near-infrared spectrometers require a sufficient amount of sample to cover the instrument’s sampling window in order to effectively capture the near-infrared spectrum. When micro near-infrared spectrometers, similar to existing micro infrared spectrometers, are developed, this problem could be resolved. Future research could apply new statistical methods, such as Partial Least Squares Support Vector Machine (PLS-SVM)³⁶ and Orthogonal Least Partial Squares Discriminant Analysis (OPLS-DA),³⁷ which have already been applied in other fields, to further improve the accuracy and robustness of the models.

ORCID iD

Kaikai Jia https://orcid.org/0009-0006-4053-5079

Statements and Declarations

Footnotes

Conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Correction (May 2025):

Since the original online publication, the numbering of affiliations has been corrected.

References

Wang

JJJoSB

Entrepreneurship . Circulating fashion: a global production networks analysis of Chinese wig business in Africa. J Small Bus Enterpren 2022; 34(4): 397–418.

Weffort

Sales Martins

Plata

, et al. Do you know how to recommend a wig to your patient? Journal of cosmetic dermatology 2021; 20(3): 724–728.

Atav

Ergünay

Gürkan Ünal

PJTJoTTI

. Distinguishing yak and cashmere fibers from each other to prohibit mislabeling through microscopic and spectroscopic analysis. J Text Inst 2024; 115(3): 366–379.

Madkour

Abdelsabour-Khalaf

. Performance scanning electron microscopic investigations and elemental analysis of hair of the different animal species for forensic identification. Microsc Res Tech 2022; 85(6): 2152–2161.

Zhu

Huang

, et al. An identification method of cashmere and wool by the two features fusion. Int J Cloth Sci Technol 2022; 34(1): 13–20.

Zhang

Yang

, et al. Identification of cashmere and wool by DNA barcode. J Nat Fibers 2023; 20(1): 2175100.

Keerti

Ninave

. DNA fingerprinting: use of autosomal short tandem repeats in forensic DNA typing. Cureus 2022; 14(10): e30210.

Liu

Fan

Zhang

, et al. A human identification system for hair shaft using RNA polymorphism. Forensic Sci Int: Genetics 2023; 67: 102929.

Fan

Yang

, et al. Human identification of single hair shaft using a mass spectrometry mRNA typing system. Forensic Sci Int: Genetics 2025; 74: 103158.

10.

Paolella

Bencivenni

Lambertini

, et al. Identification and quantification of different species in animal fibres by LC/ESI-MS analysis of keratin‐derived proteolytic peptides. J Mass Spectrom 2013; 48(8): 919–926.

11.

Kuckova

Smirnova

Straka

, et al. Proteotypic peptides of hairs for the identification of common European domestic and wild animal species revealed by in-sample protein digestion and mass spectrometry analysis. J Sep Sci 2023; 46(13): 2300064.

12.

Jahanbanifard

Price

González

, et al. A novel method to analyse DART TOFMS spectra based on Convolutional Neural Networks: a case study on methanol extracts of wool fibres from endangered camelids. Int J Mass Spectrom 2023; 489: 117050.

13.

Fei

Han

Chen

, et al. A protein analysis-based method for authentication of human hair wigs. Text Res J 2023; 93(19–20): 4648–4655.

14.

El Maouardi

De Braekeleer

Bouklouze

, et al. Comparison of Near-Infrared and Mid-Infrared spectroscopy for the identification and quantification of argan oil adulteration through PCA, PLS-DA and PLS. Food Control 2024; 165: 110671.

15.

Sharma

Rawat

, et al. Rapid and non-destructive differentiation of Shahtoosh from Pashmina/Cashmere wool using ATR FT-IR spectroscopy. Sci Justice 2022; 62(3): 349–357.

16.

Mueller

Hecker

Goodpaster

JVJMJ

. Differentiation and classification of human head hair via spectroscopic analysis, data fusion, and neural networks. Microchem J 2024; 204: 111024.

17.

Bhatia

Sharma

, et al. Species discrimination from hair using ATR-FTIR spectroscopy: application in wildlife forensics. Sci Justice 2024; 64(3): 314–321.

18.

de Oliveira

Lourenço

FRJC

. Risk of false conformity assessment applied to automotive fuel analysis: a multiparameter approach. Chemosphere 2021; 263: 128265.

19.

Ning

Wang

Jiang

, et al. Quantitative detection of zearalenone in wheat grains based on near-infrared spectroscopy. Spectrochim Acta Mol Biomol Spectrosc 2022; 280: 121545.

20.

Vieira

Assis

de Queiroz

MELR

, et al. Building robust models for identification of adulteration in olive oil using FT-NIR, PLS-DA and variable selection. Food Chem 2021; 345: 128866.

21.

Jiménez-Carvelo

Martín-Torres

Ortega-Gavilán

, et al. PLS-DA vs sparse PLS-DA in food traceability. A case study: authentication of avocado samples. Talanta 2021; 224: 121904.

22.

Chen

Lan

Zhu

. Non-destructive identification of wool and cashmere fibers based on improved LDA using NIR spectroscopy. Autex Res J 2024; 24(1): 20230017.

23.

Chen

Men

, et al. Quantitative analysis of wool and cashmere fiber mixtures using NIR spectroscopy. AUTEX Research Journal 2024; 24(1): 20240010.

24.

Amorena

Fernández-Ahumada

Álvarez

DME

. Preventing adulteration: authentication and identification of camelids and other textile fibers with NIR spectroscopy. J Text Inst 2024; 1–9.

25.

Qiao

Chen

Sun

, et al. Identification of human hair wigs and animal hair wigs by the method of near infrared spectroscopy modeling. Text Res J 2022; 92(19–20): 3838–3843.

26.

Mohammed

Akawee

Saleh

, et al. The effectiveness of big data classification control based on principal component analysis. Bull Electr Eng Inf 2023; 12(1): 427–434.

27.

Hasan

BMS

Abdulazeez

Mining

. A review of principal component analysis algorithm for dimensionality reduction. Journal of Soft Computing and Data Mining 2021; 2(1): 20–30.

28.

Greenacre

Groenen

Hastie

, et al. Principal component analysis. Nat Rev Methods Primers 2022; 2(1): 100.

29.

Zoccola

Mossotti

, et al. Identification of wool, cashmere, yak, and angora rabbit fibers and quantitative determination of wool and cashmere in blend: a near infrared spectroscopy study. Fibers Polym 2013; 14: 1283–1289.

30.

Xia

C-Y

Drugs

. Study on fast prediction of film coating terminal point of Tianshu tablets by near-infrared spectroscopy. Zhongcaoyao 2019; 50(21): 5223–5230.

31.

Si-jun

Long

Hong-gen

, et al. A study on consistency evaluation of preparation of sinomenine hydrochloride based on multi-source data fusion technique. Journal of Instrumental Analysis 2023; 41(1): 121–127.

32.

Björck

ÅJ

. Numerics of gram-schmidt orthogonalization. Linear Algebra Its Appl 1994; 197: 297–316.

33.

Ruiz-Perez

Guan

Madhivanan

, et al. So you think you can PLS-DA? BMC bioinformatics 2020; 21: 1–10.

34.

Lasalvia

Capozzi

Perna

. A comparison of PCA-LDA and PLS-DA techniques for classification of vibrational spectra. Appl Sci 2022; 12(11): 5345.

35.

Lee

Liong

C-Y

Jemain

. Partial least squares-discriminant analysis (PLS-DA) for classification of high-dimensional (HD) data: a review of contemporary practice strategies and knowledge gaps. The Analyst 2018; 143(15): 3526–3539.

36.

Mohamad

Hussein

. Diagnosing anemia disease using partial least square (PLS) models and Support vector machines (SVM). Journal of Kurdistani for Strategic Studies 2024; 2(9): 157–173.

37.

How

Hamid

Liu

, et al. Using OPLS-DA to fingerprint key free amino and fatty acids in understanding the influence of high pressure processing in New Zealand clams. Foods 2023; 12(6): 1162.