📄 agriculture

Spectral-Correlation and Random Matrix Theory- Analysis of Raman Spectra for Detection of Milk Adulterants

This study demonstrates that combining spectral-correlation diagnostics with Random Matrix Theory, specifically through the Brody parameter, provides a robust, statistically grounded framework for detecting and distinguishing milk adulterants (hydrogen peroxide, urea, and formalin) by quantifying the degree of spectral randomization in Raman spectra.

Original authors: Ravinder Tirupati, Manimaran P., Ghanashyam Krishna M., Samrat L. Sabat

Published 2026-09-30

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Ravinder Tirupati, Manimaran P., Ghanashyam Krishna M., Samrat L. Sabat

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Food safety relies heavily on the ability to spot when a natural product has been tampered with, and milk is a frequent target for such deception. Unscrupulous sellers sometimes add chemicals to milk to make it look thicker, last longer on the shelf, or pass basic quality tests, even though these additives can be harmful to human health. To catch these tricks, scientists often use a technique called Raman spectroscopy, which works by shining a laser on a sample and listening to the unique way the molecules vibrate in response. Every substance has a distinct vibrational signature, much like a fingerprint, allowing researchers to identify what is present in a glass of milk. However, milk is a complex mixture of proteins, fats, and sugars, and its natural spectrum is crowded with overlapping signals. When adulterants are added, they do not just add a new line to this spectrum; they can subtly scramble the relationships between all the existing signals, creating a pattern of change that is difficult for standard computer programs to spot, especially if the amount of added chemical is small or if the data is noisy.

A team of researchers at the University of Hyderabad has developed a new way to look at this problem by treating the entire spectrum not as a list of peaks, but as a web of connections. Instead of asking "what is the height of this specific peak?", they asked "how do all the different parts of the spectrum move together?" They collected laser readings from pure milk and from milk mixed with three common adulterants: hydrogen peroxide, urea, and formalin. For each group, they built a massive map showing how every single point in the spectrum correlated with every other point. They then applied a branch of mathematics known as Random Matrix Theory, which is typically used to study complex systems like atomic nuclei or financial markets, to see if the patterns in these maps were orderly or chaotic. This approach allowed them to measure the "disorder" introduced by each additive with a high degree of precision, revealing that different chemicals disrupt the milk's natural structure in fundamentally different ways.

The study began with a collection of 124 laser readings. Thirty-one of these came from pure milk, while the rest were samples where the researchers had carefully mixed in varying amounts of hydrogen peroxide, urea, or formalin. The team used a powerful laser to capture the vibrational signals from each sample, creating a detailed picture of the molecular activity. They then processed this data to remove background noise and built a correlation matrix for each group. In simple terms, this matrix is a giant grid that records how closely the behavior of one part of the spectrum matches the behavior of another part across all the samples. In pure milk, this grid showed strong, organized blocks of connection, reflecting the fact that the proteins, fats, and sugars in milk naturally move in sync. When the researchers introduced the adulterants, these organized blocks began to change, but each chemical did so in a unique manner.

Hydrogen peroxide caused a very specific, localized disturbance. The researchers found that it created a sharp break in the connections right around the frequency where peroxide molecules vibrate, but the rest of the milk's spectrum remained largely intact and orderly. It was as if a single thread in a woven fabric had been cut, leaving the surrounding weave undisturbed. Urea, on the other hand, produced a much stronger and more structured disruption. It created a pronounced shift in the connections near its own vibrational signature, and this effect rippled outward to influence neighboring parts of the spectrum. This suggested that urea was not just sitting in the milk but was actively interacting with the surrounding molecules, creating a new, rigid pattern of behavior that was different from the natural flow of pure milk.

Formalin presented the most dramatic change of all. Unlike the other two chemicals, which made specific, localized changes, formalin caused a widespread weakening of the connections across the entire spectrum. The organized blocks seen in pure milk dissolved into a more scattered, random pattern. This result aligned with the known chemistry of formalin, which is a cross-linking agent that binds to proteins in many different places at once, subtly scrambling the entire molecular structure rather than targeting a single spot. To quantify these observations, the researchers used a statistical measure called the Brody parameter, which acts as a single number to describe how random a system is. Pure milk scored low on this scale, confirming its highly ordered nature. As the researchers added adulterants, the score moved higher, indicating increasing disorder. Formalin pushed the score closest to the maximum possible randomness, while urea and hydrogen peroxide fell in between, creating a clear, numerical ranking of how much each chemical disturbed the milk's natural state.

The significance of this work lies in its ability to detect adulteration by looking at the structure of the data rather than just the intensity of specific signals. Traditional methods often struggle when the amount of added chemical is very small or when the background noise is high, because they rely on finding a distinct peak that stands out. This new approach, however, detects the subtle ways in which an additive changes the relationships between all the molecules in the sample. By treating the spectrum as a complex system and measuring its statistical randomness, the researchers demonstrated that they could distinguish between pure milk and milk containing harmful additives with high sensitivity. The study suggests that this method could eventually be used to build faster, more reliable screening tools for food safety, capable of spotting even low-level contamination that might otherwise go unnoticed. While the researchers note that further testing is needed to establish exactly how little of an additive can be detected, the results provide a powerful new mathematical lens for understanding the hidden complexity of food quality.

Technical Summary: Spectral-Correlation and Random Matrix Theory Analysis of Raman Spectra for Detection of Milk Adulterants

Problem Statement
The detection of adulterants in milk is a critical challenge for food safety. While optical spectroscopic techniques like Raman spectroscopy offer unique molecular fingerprints for identifying additives such as hydrogen peroxide ( $H_2O_2$ ), urea ( $CH_4N_2O$ ), and formalin ( $CH_2O$ ), conventional analysis methods face limitations. Standard chemometric approaches (e.g., PCA, PLS, SVM, CNN) often rely on empirical learning from peak intensities or specific spectral patterns. These methods can be sensitive to noise, fluorescence backgrounds, and require large, labeled training datasets. Furthermore, they may fail to capture subtle, additive-induced perturbations in the covariance structure of high-dimensional spectral data that do not manifest as distinct, resolvable peak shifts. The authors posit that milk adulteration is fundamentally a problem of correlation structure rather than just peak intensity, necessitating a framework that distinguishes between genuine chemical correlations and statistical randomness.

Methodology
The study employs a framework combining Spectral-Correlation Analysis and Random Matrix Theory (RMT) to analyze Raman spectra of milk.

Data Acquisition: A total of 124 Raman spectra were acquired: 31 from pure milk (4.5% fat) and 93 from milk adulterated with varying concentrations of $H_2O_2$ , urea, and formalin. Spectra were recorded using a 532 nm confocal Raman microscope.
Preprocessing: Raw spectra underwent asymmetric least squares (ALS) baseline correction, Savitzky–Golay smoothing, and normalization to the [0, 1] range.
Correlation Matrix Construction: For each class (pure and each adulterant), a $700 \times 700$ spectral correlation matrix ( $C$ ) was computed from the mean-centered, standardized intensity data of the 31 samples in that class.
Eigenvalue Decomposition: The matrices were decomposed to extract eigenvalues and eigenvectors. The top three eigenvectors were analyzed to identify dominant spectral correlation patterns. Difference eigenvectors ( $\Delta v_k$ ) were calculated by subtracting pure milk eigenvectors from adulterated ones to isolate additive-specific perturbations.
RMT Analysis:
- Unfolding: Eigenvalues were unfolded using spline-based smoothing to ensure a uniform mean level density.
- Spacing Distribution: The nearest-neighbor spacing distribution ( $P(S)$ ) of the unfolded eigenvalues was compared against the Wigner Surmise (Gaussian Orthogonal Ensemble) and the Poisson distribution.
- Brody Parameter: To quantify the degree of spectral order versus randomness, the empirical spacing distributions were fitted to the Brody distribution, yielding a single parameter $\beta$ (where $\beta=0$ represents Poisson/random and $\beta=1$ represents Wigner-Dyson/fully correlated).

Key Results

Correlation Structure: Pure milk exhibited broad, high-intensity block domains in its correlation matrix, indicating strong, coherent biochemical covariation.
- $H_2O_2$ : Induced a localized reduction in correlation near 878 cm $^{-1}$ (O–O stretching), creating a narrow, chemically isolated perturbation without disrupting global coherence.
- Urea: Produced a strong, structured deviation centered at 1006 cm $^{-1}$ (C–N stretching), affecting multiple eigenmodes and extending into neighboring spectral regions.
- Formalin: Caused a global weakening of spectral correlations across the entire 500–3200 cm $^{-1}$ range, consistent with its role as a crosslinking agent that diffusely perturbs protein vibrational modes.
Eigenvalue and Eigenvector Analysis: Pure milk showed a sharp decay in eigenvalues, reflecting chemical uniformity. Adulterated samples showed altered distributions: urea maintained elevated eigenvalues due to strong deterministic contributions, while formalin produced a dispersed spectrum with slower decay, indicating stochastic, widespread perturbations. Difference eigenvectors successfully isolated additive-specific Raman shifts (e.g., the sharp peak at 1006 cm $^{-1}$ for urea).
Quantitative Randomness (Brody Parameter): The Brody parameter $\beta$ $β$ provided a continuous measure of spectral disorder:
- Pure Milk: $\beta = 0.749$ (Most structured/deterministic).
- Urea-adulterated: $\beta = 0.801$ (Modest shift toward randomness).
- $H_2O_2$ -adulterated: $\beta = 0.841$ (Intermediate randomness).
- Formalin-adulterated: $\beta = 0.970$ (Near-fully randomized, closest to GOE statistics).

Significance and Claims
The paper claims to be the first comprehensive application of correlation-matrix and RMT diagnostics to Raman spectra of milk. The primary contribution is the establishment of RMT-based spectral correlation analysis as a robust, adulterant-specific diagnostic framework that operates on the statistical structure of the spectrum rather than individual peak intensities.

Key claims include:

Quantitative Metric: The Brody parameter serves as a single, statistically grounded metric to rank adulterants by the severity of the correlation-structure disruption they induce.
Unsupervised Capability: The approach is largely unsupervised and does not require large labeled training datasets, addressing a key limitation of machine learning classifiers in food safety.
Interpretability: Unlike "black-box" models, the method yields interpretable results where difference eigenvectors localize perturbations to specific, chemically identifiable Raman shifts.
Sensitivity to Hidden Disorder: The method is sensitive to changes in covariance structure, potentially offering improved detection of weak or low-concentration adulteration that does not produce clearly resolvable peak shifts.

The authors conclude that this framework complements existing Raman-based detection methods by providing a physics-based, quantitative measure of spectral disorder, offering a pathway toward real-time, multi-adulterant food authentication systems. They note that while the hypothesis of improved sensitivity to low-concentration adulteration is motivated by the method's sensitivity to covariance changes, establishing specific concentration-dependent detection limits requires future dedicated studies.

More like this