Food authentication development is necessary for food control because this is the process that will validate whether a particular food complies with its label description, origin of the food (species, gene, geographical), production method (conventional, organic, free-range, etc.), and processing method (frozen, etc.) [29–31]. The spectroscopic methods are fast becoming popular [32,33]. These approaches yield chemical fingerprints that are unique to each food and may be used to differentiate between them or to validate them. The idea behind vibrational spectroscopy is that atom-to-atom links inside molecules vibrate at frequencies that can be represented by physical laws and that can be calculated [34]. Using the assumption that bond energy comes from the vibrations of a diatomic harmonic oscillator and that they satisfy Hooke’s Law, it is possible to estimate the lowest fundamental frequencies of any two atoms joined by a chemical bond, as shown in Eq. (1).

(1)v=12πkμ

The vibrational frequency is v, the classical force constant is k, and the decreased mass of the two atoms is μ. Although this is effective for the fundamental frequency of simple diatomic molecules, the electron-withdrawing or electron-donating features of the surrounding atoms have a significant impact on the bond strength and length, as well as the frequency of molecular bonds. These “k” values can vary substantially, resulting in energy variations that can be computed and utilized to interpret spectral data, as shown in the graph below. In contrast to the usual spring model for molecular vibration, there is no continuous range of energy levels. Quantum theory, on the other hand, describes the existence of different levels of energy. For a diatomic molecule, the vibrational Hamiltonian is employed to solve the time-independent Schrödinger equation, which is not time-dependent.

(2)−ℏ2∂2Ψ(x)2m∂(x)+V(x)Ψ(x)=EΨ(x)where ℏ is the reduced Planck’s constant, m is the mass of the molecule, ∂ is the derivative operator, Ψ is the wave function, V is potential energy, E is energy eigenvalue, x is denting, all these quantities are in one direction. For the energy levels of diatomic molecules, a simplified form of these levels may be put out as follows:

(3)Ev=(v+12)ℏ2πkμ(v=0,1,2,….)

Alternatively, the equation may be simplified by employing the quantum term v.

(4)Ev=(v+12)ℏ(v=0,1,2,….)

When it comes to polyatomic compounds, there are a lot of different energy levels to choose from. To a first approximation, such a molecule may be thought of as a collection of diatomic, independent, harmonic oscillators. This case's equation may be generalized as

(5)E(v1,v2,v3,…)=∑i=13N−60(vi+12)ℏ

When an energy level transitions from 0 to 1 in any of the vibrational states v-1, v-2, v-3,..., the transition is deemed basic and is allowed under selection criteria. The overtone is the transition from the ground state to v-i. = 1, 2, 3..., while all the others are zero. Transitions from the ground state to a state for which combinations such as vi=1,vj=1,vk=1, or vi=2, etc.., are also possible. The use of overtones and combinations is not permitted in the purest sense, although they do appear as weak bands as a result of Fermi resonance or anharmonicity [34]. As a result of the electron clouds of the two connected atoms, as well as charges on their nuclei, an energy barrier is created, preventing the cores of the atoms from coming into contact with one another during the compression process. As a result, when the vibrational energy exceeds the dissociation energy at the stretch’s extension, the bond will finally dissolve. The stretching frequencies are listed in Table 1.

When a quick and objective study is required, absorption and fluorescence spectroscopies in the visible and infrared range are acceptable possibilities.

One of the most widely used infrared spectroscopy technology is FTIR spectroscopy. Using Michelson interferometer Fig. 1a, interferograms of both the sample and the background can be collected. The interferograms collected can be passed through the fast Fourier transform algorithm to produce single beam spectra as shown in Fig. 1b, which then passes the spectrum after transforming it as transmittance spectrum for analysis.

Principal component analysis (PCA) is a traditional statistical method for changing the characteristics of a dataset into an uncorrelated collection of points known as principal components. Chemometrics approaches are founded on statistical transformation techniques such as PCA, which are used to change the characteristics of a dataset into an uncorrelated collection known as principal components. The concept is that by removing some principal components (PCs), the data may be reduced to a lower number of dimensions with minimal information loss. Each PC is a linear combination of the original inputs, and each PC is orthogonal, therefore collinearity isn’t an issue. The most extensively used chemometric classification approach is Soft Independent Modeling of Class Analogy (SIMCA). SIMCA builds a distinct PCA model for each set of data in both classes in binary classification; in prediction, the test sample’s distance to either model is determined. The test sample is then put through statistical testing to see if it belongs to one of the two classes [36]. Aside from that, the chemometric method known as Partial Least Squares (PLS) is commonly employed for quantitative analysis. When doing multivariate regression, PLS reduces data via PCA (using concentration information to produce PC scores) before using linear regression to the PC scores. The examination of method linearity was carried out in PLS calibration models to demonstrate a proportionate connection between absorbance and adulterated concentrations of the samples. Some characteristics, such as the regression coefficient (R²), root mean standard error of estimation (RMSEE), and root mean standard error of prediction (RMSEP), were analyzed in order to select the optimum calibration model (RMSEP). The PLS techniques were utilized to calculate the FTIR spectra filters, which included normal and other spectra filters such as its derivatives, multiplicative signal correction (MSC), and standard normal variate (SNV), among others.

According to Che Man et al., lard and other animal body fats from meat, such as chicken fat, beef fat, and mutton fat, were extracted from the flesh by rendering the fatty tissues in the meat processing process with slight variation. All of the substances used in this experiment were of the highest analytical quality. Spectroscopy was then used to examine the pure lipids. The purified samples are made from the pure fats obtained after the extraction operation, as shown in Fig. 2.

Lamb, beef, and chicken body fats were blended with lard to produce a set of 10 pure samples and 30 samples with 10–50 percent weight-weight ratio of lard in the lamb, beef, and chicken samples. This method follows Rohman et al.

Two programs were used to post-process the data: Spectrograph 1.1 and MATLAB R2017b. Spectrograph 1.1 was used to extract information from the spectrum, and the data was pre-processed as needed. The findings of the pre-processing were further analyzed using MATLAB R2017b. The quality of lard adulteration was assessed using Principal Components Analysis (PCA), whereas the amount of lard adulteration was assessed using Partial Least Square (PLS). The instrument used in this study is from Frontier FT-IR by PerkinElmer. The adulterated samples and their percentage composition is shown in Table 2.

The FTIR Spectra of pure fats are shown in Fig. 3. Each of these spectra is separated into four regions: the first covers the range from 4000 to 2500 cm⁻¹, the second covers the range from 2500 to 2000 cm⁻¹, the third covers the range from 2000 to 1500 cm⁻¹, and the fourth covers the range between 1500 and 800 cm⁻¹. Initially, we have values of transmittance of each sample. Then, by taking the reciprocal transmittance, we get the absorbance of the material at different wavelengths. The table shows ten samples of lamb, chicken, and beef with a predetermined proportion of lard combination.

With the wavelength spectrum starting from 4000–650 cm⁻¹, we get each sample’s absorbance value. Our primary dataset consists of all the beef, chicken, lamb, and pork values, as shown in Fig. 3. Three regions are identified along with fingerprint region in the spectrum. 2^nd region (2500–2000 cm⁻¹) is omitted out of analyses due to its unchanging absorbance value for all the species. For classification purposes, we have detached the dataset into four different data sets, as we have only one meat in the category of haram (Lard). In contrast, all three meats lie in the variety of halal classification (Chicken, lamb and beef). The FTIR Spectra from all samples are represented by the absorbance value; in the fingerprint area, all models have different patterns of peaks and shoulders. In both spectra, the absorption of individual peaks is somewhat different, hence the chosen fingerprint region differs slightly. In contrast to pork, the difference in absorbance is indicated individually for beef, chicken, and lamb.

The chemometric of principal component analysis was used to classify the pure pork meat as well as other beef, chicken, and lamb samples. PCA’s wavenumber regions were also improved. Finally, because of their capacity to offer adequate separation among the analyzed samples, the same wavenumbers employed for quantitative analysis, ranges between 4000–650 cm⁻¹, were chosen for PCA modeling. The classification was performed using SIMCA with PLS and SIMCA with PCA. Due to the presence of zeros in our data set, we performed baseline correction to make zeros attain positive values. Cross-validation is done before PCA classification; along with this, we have done grouping based on scores plot.

PCA method divided the problem into four principal components. Component PC-1 holds an 83% score while 11% score is kept by PC-2, as shown in Fig. 8. The distribution of loading values using the PCA method shown in Fig. 4 separates samples of Lamb (L1-L10), Chicken (C1-C10), and Pork (P1-P10); however, the score plots of Beef (B1-B4) and Beef (B5-B10) scatters along the line, while chicken sample score is far away from pork and below the zero scores of PC-2.

Meanwhile, the lamb and chicken samples distinguish by separating the total pieces below and above the score line. In contrast, only pure pork samples are concentrated on the zero scores’ line axis. Moreover, the variable contribution plot shown in Fig. 5 indicates the beef and chicken samples contribution along with pork variables, whereas only lamb is a highly correlated feature which stipulates the similarity of pork and lamb variables as presented in Fig. 5.

PCA results at 1^st region indicate that some of the chicken and pork samples are clustered together, even though there is still overlap between specimens of different species as presented in Fig. 6.

PCA results at the 3^rd region show that chicken and pork samples clustered together without overlap, lamb samples have one outlier. However, the beef samples proved challenging since the results are still scattered. The three-dimension PCA shows better results, though with all-beef samples lined around the first PCA component as shown in Fig. 7.

As the fingerprint region, PCA results show that chicken and pork samples clustered together without overlap. Lamb samples have one outlier, but the beef samples are scattered as shown in Fig. 8. This region is the most significant region as all the samples are separated without overlapping the PCA mapping.