Ep 25 — Food Industry: Adulteration Detection (Milk, Honey, Olive Oil)

Series: Encyclopedia of Infrared Spectroscopy: From Principles to Practice
Chapter: Part 3 · Intermediate — Industrial Applications
Target Audience: Food inspection agency technicians, food enterprise QC, import/export inspection personnel
Prerequisites: Ep 14 (ATR), Ep 20 (Quantitative Basics), Ep 23 (Blend Identification Approach)
Reading Time: ~48 minutes

Introduction: The Cost of a "Fake Honey"

In 2019, a provincial market regulatory bureau received consumer complaints that a local brand of honey had "thin taste and abnormal crystallization." After testing using FTIR-ATR and chemometrics, it was found that the sample's infrared spectrum in the 1050–1150 cm⁻¹ region (C-O stretching of sugars) was significantly different from pure honey — showing typical characteristics of high fructose corn syrup (HFCS) [1].

Further PLS-DA model determination indicated that the sample contained approximately 35% high fructose corn syrup. Subsequent investigations revealed that the company had purchased HFCS from a chemical plant and directly blended it into low-cost honey, passing off inferior goods as quality products. Ultimately, the company's food production license was revoked, and its principal responsible person was held criminally liable [1].

"Honey adulteration with cheap sweeteners (sucrose, HFCS) is one of the most common food frauds worldwide, and FTIR spectroscopy combined with chemometrics is a powerful screening tool."
—— Downey G et al., Food Chemistry, 2014 [2]

Food adulteration is a global problem. According to the global food safety database — the "USP Food Fraud Database" — milk, honey, and olive oil are the three foods with the highest adulteration rates [3]. This episode will focus on these three food categories, systematically learning the application of FTIR in adulteration detection — from single-peak methods to chemometrics, from rapid screening to compliance determination.

1. Food Adulteration: An "Ancient" Problem

1.1 What is Food Adulteration?

Food Adulteration refers to the intentional addition of cheap or non-food substances to food for undue economic gain [3][4]:

1.2 Current Status of Milk, Honey, and Olive Oil Adulteration

1.2.1 Common Milk Adulteration Methods [4][5]

1.2.2 Honey Adulteration [2][6]

• Addition of sucrose syrup: most traditional
• Addition of HFCS: most common due to low cost
• Addition of beet syrup: similar sugar composition to honey, difficult to distinguish
• Direct artificial honey: corn syrup + flavoring

1.2.3 Olive Oil Adulteration [7]

• Adulteration with soybean oil, sunflower oil, corn oil: cost reduction
• Adulteration with hazelnut oil: composition similar to olive oil
• Adulteration with low-quality olive oil: passing off inferior goods
• Label fraud: origin and grade falsification

1.3 Role of FTIR in Food Testing

Advantages of FTIR in food adulteration detection [2][4][5]:

① Fast: ATR measurement in 1–2 minutes
② Multi-component: One scan covers all components
③ No reagents: Environmentally friendly, low consumption
④ Trace sample: One drop sufficient for measurement
⑤ Combined with chemometrics: Can identify very complex adulteration

Limitations:

• Aqueous solution interference: Strong water absorption can mask some signals
• Low concentration adulteration: Detection limit typically 1–5%
• Matrix effects: Vary greatly among different foods; each requires establishment of dedicated models

2. FTIR Detection of Milk Adulteration

2.1 Infrared "Fingerprint" of Milk

Milk is a complex system containing water, fat, protein, lactose, and minerals [4][5]. Its main infrared characteristics are:

🔗 Extension: The amide I and amide II double peaks (1650/1540 cm⁻¹) of milk come from milk proteins. On ftir.fun amide functional group page, it can be seen that 1650 and 1550 cm⁻¹ are common amide peak positions; the page shows the distribution of amide-related peaks (support count updates with library) [8]. These two peaks in the milk ATR spectrum are typical manifestations of the amide region.

2.2 Water Addition in Milk: Simple but Effective

Water dilution is the simplest form of milk adulteration. FTIR detection principle [4][5]:

• O-H stretching enhancement: More water, peak at 3300 cm⁻¹ increases
• Amide I/II weakening: Protein concentration decreases
• 1740 cm⁻¹ weakening: Fat concentration decreases

Simple Quantification:
$$\text{Water addition percentage} \approx 1 - \frac{A{1540, sample}}{A{1540, control}}$$

ATR Measurement Limitation: When water addition < 5%, sensitivity is limited because the O-H peak is already strong. Requires chemometrics or more precise methods such as refractometry.

2.3 Urea Addition to Milk

Urea CO(NH₂)₂ contains two primary amine groups, with clear infrared features [5][9]:

• 3450, 3340 cm⁻¹: N-H stretching (primary amine doublet)
• 1630 cm⁻¹: C=O stretching (amide I region)
• 1460 cm⁻¹: N-H bending
• 1150 cm⁻¹: C-N stretching

Key Diagnostic Peak: The appearance of a doublet at 3450/3340 cm⁻¹ is a clear signal of urea addition.

Quantification: Using the peak area at 1150 cm⁻¹ for a calibration curve, the urea content can be quantified.

📷 Figure 1: ATR-FTIR spectra comparison of pure milk and milk with added urea
Source: PÚBLICO E DE LIMA et al. [5]
https://doi.org/10.1016/j.lwt…

2.4 Melamine Addition to Milk

Melamine (C₃H₆N₆) contains a triazine ring + three amino groups, and was the cause of the 2008 Chinese melamine milk powder incident [10].

Infrared Features [10]:

• 3470, 3420, 3330 cm⁻¹: N-H stretching (multiple amino groups)
• 1650 cm⁻¹: Triazine ring C=N stretching
• 1550, 1460 cm⁻¹: N-H bending
• 1020 cm⁻¹: C-N stretching

Detection Challenge: The amount of melamine added is typically < 1% (to avoid abnormal Kjeldahl nitrogen readings), making it difficult to detect directly with conventional ATR [10].

Solutions:

• TFA extraction enrichment: Extract melamine using trifluoroacetic acid, concentrate, and measure
• Chemometrics: Use PLS models to identify melamine at < 1%
• μ-FTIR: Observe for melamine crystalline particles

⚠️ Regulatory: GB/T 22388—2008 specifies GC-MS and LC-MS as methods for melamine detection; FTIR is used for rapid screening and is not a legal confirmatory method [10].

2.5 Application of Chemometrics in Milk Testing

Milk infrared spectra are complex, making chemometrics crucial [4][5][9]:

① PCA (Principal Component Analysis): Identify abnormal samples

• PCA clustering of pure milk samples
• Adulterated samples deviate from the cluster
• Visualize "abnormal" samples

② PLS-DA (Partial Least Squares Discriminant Analysis): Classification

• Training set: pure milk vs. adulterated milk (various types)
• The model can discriminate "whether adulterated"
• Accuracy can exceed 95%

③ PLS (Quantitative): Quantify adulteration ratio

• Training set: samples with known urea content
• The model predicts urea content
• Detection limit: approximately 0.1–0.5%

3. FTIR Detection of Honey Adulteration

3.1 Infrared "Fingerprint" of Honey

Honey mainly contains fructose, glucose, sucrose and other sugars (approximately 80%) and water (approximately 17%) [2][6].

🔗 Extension: Honey adulteration screening should simultaneously examine the O–H stretching region and the sugar C–O (approximately 1050–1150 cm⁻¹). For O–H, refer to the hydroxyl page; for the C–O backbone, it is better to refer to the ether/C–O related page or carbohydrate page (if data is available on the page); do not force sugar C–O to be "the main content of the hydroxyl page".

3.2 Detection of Honey Adulterated with HFCS

High Fructose Corn Syrup (HFCS) is the most common adulterant in honey [2][6]. HFCS is produced by hydrolysis and isomerization of corn starch, and mainly contains fructose and glucose—very similar to the sugar composition of honey.

3.2.1 Single-Peak Method for Differentiation

However, HFCS differs from honey in certain trace components:

• Honey contains natural pollen protein: double peaks at 1650/1540 cm⁻¹
• HFCS has almost no protein: double peaks are weak or absent
• Honey contains organic acids: weak peak at 1730 cm⁻¹
• HFCS has almost none: 1730 cm⁻¹ absent

Judgment method [2][6]:

• Measure the $A{1650}/A{1050}$ ratio
• Pure honey > 0.05
• Honey adulterated with HFCS < 0.03

3.2.2 Chemometric Discrimination

A more reliable method is PLS-DA [2][6]:

• Training set: pure honey + honey adulterated with HFCS (0%, 10%, 20%, … 100%)
• Full spectrum (4000–400 cm⁻¹) for PLS-DA
• Accuracy can reach 98%

📷 Figure 2: FTIR spectral comparison of pure honey and honey adulterated with HFCS
Source: Downey G et al., Food Chemistry, 2014 [2]
https://doi.org/10.1016/j.foo…

3.3 Detection of Honey Adulterated with Sucrose Syrup

Sucrose syrup (e.g., cane sugar syrup) is more difficult to detect than HFCS because [6]:

• Sucrose itself naturally occurs in honey
• Infrared spectra are highly similar to honey

Solutions:

• Isotope analysis (IRMS): The ¹³C of C4 plants (corn) differs from that of C3 plants (honey floral source)
• Characteristic peak ratios: Subtle differences in ratios such as 1045/1020 cm⁻¹
• Combined with chromatography: HPLC to determine oligosaccharide distribution

4. FTIR Detection of Olive Oil Adulteration

4.1 Infrared "Fingerprint" of Olive Oil

Olive oil is a mixture of triglycerides, mainly containing oleic acid (C18:1), linoleic acid (C18:2), palmitic acid (C16:0) and other fatty acids [7][12].

🔗 Extension: The C=O stretch of olive oil at 1746 cm⁻¹ belongs to the typical ester group. On the ftir.fun ester functional group page, 1740, 1730, and 1735 cm⁻¹ are the three more common ester C=O frequencies [13]. The 1746 cm⁻¹ of olive oil is slightly higher than typical esters due to the inductive effect of the ester group in triglycerides.

4.2 Detection of Olive Oil Adulterated with Soybean/Sunflower Oil

Different vegetable oils have slightly different fatty acid compositions [7][12]:

Key difference: Soybean, sunflower, and corn oils contain more unsaturated fatty acids → stronger 3007 cm⁻¹ (=C-H stretch) peak [12].

4.3 Single-Peak Method vs. Chemometrics

Single-Peak Method [12]:

• Measure the $A{3007}/A{1746}$ ratio
• Pure olive oil: approximately 0.07–0.10
• With 20% soybean oil: approximately 0.10–0.12
• With 50% soybean oil: approximately 0.13–0.15

Advantages: Simple and fast
Limitations: Detection limit approximately 10–15%, difficult to identify low-level adulteration

Chemometrics (PLS-DA) [7]:

• Full spectrum training
• Detection limit can be as low as 5%
• Accuracy > 95%

4.4 Detection of Olive Oil Oxidation

Olive oil itself can undergo oxidative rancidity, affecting quality. FTIR can monitor [12]:

• New peak at 1710 cm⁻¹: Free fatty acid C=O (from hydrolysis)
• O-H at 3470 cm⁻¹: Hydroperoxides
• Enhanced 967 cm⁻¹: Trans fats (oxidative isomerization)

"FTIR has become a routine tool for olive oil quality assessment, with the Codex Alimentarius and IOC (International Olive Council) recognizing several IR-based methods."
—— International Olive Council, Trade Standard, 2022 [7]

4.5 Regulatory Basis

• IOC (International Olive Council) COI/T.20/Doc. No 29: FTIR method for determining olive oil quality parameters

• AOCS Cd 1e-01: FTIR determination of trans fats

• GB 5009.168—2016: National food safety standard—Determination of fatty acids in foods (primarily chromatography; if IR appears in methods/appendices, it must be checked against the standard text; do not loosely write 'includes IR quantification')

V. Practical Advantages and Limitations of ATR-FTIR for Rapid Food Testing

5.1 Advantages of ATR for Rapid Food Testing

5.2 Limitations of ATR for Rapid Food Testing

① Aqueous solution interference

Water has strong IR absorption [4][5]:

• 3300 cm⁻¹ (O-H stretch): broad and strong
• 1640 cm⁻¹ (H-O-H bend): broad
• These two regions are almost "blind" for sample signals

Solutions:

• Use crystals with shallower penetration depth (e.g., Ge, penetration 0.2 μm)
• Use chemometrics to subtract water contribution
• Use drying (but lose volatile information)

② Detection limit limitations

ATR penetration depth is only 0.5–5 μm, making trace adulteration difficult to detect directly [5][9]:

• Melamine < 0.1%: difficult with conventional ATR
• Adulteration < 5%: requires chemometrics

③ Matrix effects

Different food matrices vary greatly; models cannot be extrapolated [9]:

• Milk PLS model cannot be used for juice
• Honeys from different floral sources require independent models
• Olive oils from different origins show significant differences

5.3 Sample Preparation Tips

Milk: Direct ATR, 1–2 drops sufficient, no pretreatment [4]

Honey: If crystallized, melt in a 40°C water bath [6]

Olive oil: Direct ATR, avoid bubbles [7]

Solid foods: Grind and press KBr pellet, or cut fresh cross-section for ATR

VI. Chemometrics Enhances Discrimination Accuracy

6.1 Why Chemometrics?

Food IR spectra are information-rich but overlapping, single-peak methods are inadequate [9][14]:

• Characteristic peaks of sugars, proteins, and fats all lie in 1000–1700 cm⁻¹
• Different components interfere with each other
• Adulterant signals are weak and often masked by the matrix

Chemometrics extracts full-spectrum information through multivariate analysis [9][14]:

6.2 Overview of Main Methods

① Data preprocessing [9][14]

• SNV (Standard Normal Variate): Eliminates particle scattering
• MSC (Multiplicative Scatter Correction): Baseline correction
• First/second derivatives: Enhance weak peaks, remove baseline tilt
• Savitzky–Golay smoothing: Noise reduction

② PCA (Principal Component Analysis) [9]

   PC2
    ▲
    │      ● Pure honey
    │   ●●●●
    │  ●●  
    │   ●     × Adulterated
    │           ××
    │            ×
    └─────────────► PC1

• Reduces multidimensional spectra to 2–3 dimensions
• Visualizes sample clustering
• Identifies "anomalous" samples

③ PLS-DA (Partial Least Squares Discriminant Analysis) [9]

• Training set: samples with known classes (e.g., pure honey vs. adulterated)
• Model learns relationship between spectra and classes
• Unknown samples can be classified

④ PLS (Quantitative Analysis) [9]

• Training set: samples with known concentrations (e.g., urea 0–5%)
• Model learns relationship between spectra and concentration
• Unknown samples can predict concentration

6.3 Model Validation

Model validation is critical to avoid overfitting [9][14]:

• Cross-validation: Leave-one-out within training set
• External validation: Independent test set
• RPD (Residual Predictive Deviation): RPD > 3 indicates usable model, > 5 indicates excellent

Identifying overfitting:

• High R² in training set, low R² in test set → overfitting
• Too many principal components → fitting noise
• Insufficient training samples → model not robust

6.4 Recommended Open-Source Tools

• Orange-Spectroscopy: Graphical interface, zero coding (see Ep 54)
• SpectroChemPy: Python framework (see Ep 53)
• HyperSpy: Advanced analysis (see Ep 54)

🔗 Extension: Systematic study of chemometrics will be detailed in Ep 43 (PCA, PLS) and Ep 44 (machine learning).

VII. Industry Standards and Method Validation

7.1 Regulatory Status of FTIR Detection of Food Adulteration

Currently, FTIR is mainly used as a screening method, not as a legal confirmatory method [3][4][5]:

7.2 Method Validation Requirements

FTIR screening methods should be validated according to AOAC or ISO 17025 [3]:

7.3 "Dual-Track" Confirmation of Positive Samples

Practice convention [3][9]:

• FTIR screening positive → mass spectrometry/isotope method confirmation
• Official report only issued after confirmation
• Legal liability must be based on legal methods

VIII. Industry Frontiers and Development

8.1 Portable FTIR On-Site Testing

• Handheld FTIR (e.g., Bruker ALPHA II): can be used for market spot checks
• Built-in models: directly display "pure honey / adulterated" judgment
• Results in 1 minute, suitable for high-throughput screening [4]

8.2 Quantum Cascade Laser Infrared (QCL)

Quantum cascade lasers (QCL) offer 1000× higher brightness than traditional FTIR [14]:

• Greatly reduced detection limits
• Can detect < 0.1% melamine
• Already applied to liquid foods such as milk and wine

8.3 Machine Learning + FTIR

• Deep learning (CNN): Direct discrimination of adulteration from spectra
• Transfer learning: Trainable with few samples
• Explainable AI: Identifies spectral regions most important for discrimination

8.4 Multi-Sensor Fusion

• FTIR + Raman + NIR + fluorescence
• Multimodal information complementarity
• Improved discrimination accuracy

IX. Industry Experience and "Pitfall" Records

9.1 Pitfall 1: Ignoring Sample Pretreatment

Case [6]: Direct ATR measurement of crystallized honey gave abnormal spectra. Reason: poor contact between crystal particles and ATR crystal.

Lesson:

• Honey must be melted in a 40°C water bath
• Cool to room temperature before measurement
• Ensure liquid covers the crystal uniformly

9.2 Pitfall 2: Baseline Interference from Aqueous Solutions

Case [4]: The amide I band at 1650 cm⁻¹ was almost invisible in the ATR spectrum of a milk sample. Reason: the H-O-H bending peak of water at 1640 cm⁻¹ overlaps with amide I, causing severe interference.

Solution:

• Use Ge crystal (shallower penetration)
• Or use difference spectroscopy to subtract pure water spectrum
• Or use PLS model (automatically handles baseline)

9.3 Pitfall 3: Model Extrapolation

Case [9]: The PLS model developed for liquid milk from one brand failed to detect adulteration in milk powder from another brand.

Reason: Differences in matrix, particle size, etc.

Lesson:

• The model is only applicable to the scope defined by the training set
• New sample types must be added to the model
• Periodic model updating and validation required

Case [9]: A PLS-DA model built with honey from one region gave high misclassification rates when applied to honey from another region. The reason is that the background spectra of honey from different floral sources vary.

Lesson:

• The model training set should cover all variations of the target samples
• Different origins/varieties require independent models
• Regularly update the model with new samples

9.4 Pitfall Four: Ignoring Additive Interference

Case [4]: Vitamin A (fat-soluble) was added to milk, enhancing the 1740 cm⁻¹ peak, which was misidentified as "added vegetable oil."

Solution:

• Understand the sample formulation and additives
• Subtract the additive spectrum if necessary
• Combine with sensory and other methods for judgment

9.5 Pitfall Five: Solely Relying on HQI Library Search

Case [6]: A honey sample library search hit "honey" with a match of 0.92, and QC passed it directly. In reality, it contained 15% HFCS—honey was still the main component, so the library search naturally hit.

Lesson:

• Library search is insensitive to adulteration
• Adulteration detection must use discriminant models (PLS-DA), not mere similarity matching
• A match score only indicates "contains honey," not "only honey"

Summary of This Episode

Review Questions

1. You receive a milk powder sample suspected of melamine adulteration. Design an ATR-FTIR screening protocol including sample preparation, spectrum acquisition, decision criteria, and indicate when LC-MS confirmation is needed. Refer to ftir.fun amide page to select key peaks.

2. In an ATR spectrum of a honey sample, the 1650/1540 cm⁻¹ doublet is very weak (almost absent), while the 1050–1150 cm⁻¹ sugar peak is normal. Identify the possible adulterant and explain the reasoning.

3. For olive oil adulterated with soybean oil, the single-peak method uses the ratio A3007/A1746. Pure olive oil gives 0.08; with 30% soybean oil, the ratio becomes about 0.11. Using data from ftir.fun ester page, explain the assignment of the 1746 cm⁻¹ C=O stretch and discuss factors affecting this ratio.

4. You need to build a PLS-DA model to discriminate pure honey from honey adulterated with HFCS. Design the training set, validation method, and performance metrics, and discuss the risk of overfitting. For characteristic peak regions, consider both hydroxyl and sugar C–O (ether/carbohydrate).

5. When measuring milk directly by ATR-FTIR, strong water absorption (3300, 1640 cm⁻¹) severely interferes. List at least three solutions and compare their pros and cons.

References

[1] State Administration for Market Regulation. Compilation of Food Sampling Announcements and Typical Cases 2019.
https://www.samr.gov.cn/

[2] Downey G, McIntyre P, Davies A N. "Detecting Honey Adulteration." Food Chemistry, 2014, 149: 39–48. DOI:10.1016/j.foodchem.2013.06.045.

[3] USP (United States Pharmacopeial Convention). Food Fraud Database. 2023.
https://www.foodfraud.org/

[4] Paradkar M M, Irudayaraj J. "A Rapid FTIR Spectroscopic Method for Estimation of Caffeine in Soft Drinks." Food Chemistry, 2002, 78(2): 261–266. DOI:10.1016/S0308-8146(01)00392-5.

[5] PÚBLICO E DE LIMA K M et al. "FTIR Spectroscopy for Identification of Adulteration in Milk." LWT - Food Science and Technology, 2016, 76: 119–124. DOI:10.1016/j.lwt.2016.09.015.

[6] Sivakesava S, Irudayaraj J. "Detection of Inverted Beet Sugar Adulteration of Honey by FTIR Spectroscopy." Journal of the Science of Food and Agriculture, 2001, 81(8): 683–690. DOI:10.1002/jsfa.859.

[7] International Olive Council (IOC). Trade Standard Applying to Olive Oils and Olive Pomace Oils. COI/T.15/NC No 3/Rev. 19, 2023.
https://www.internationaloliv…

[8] ftir.fun. "Amide — FTIR absorption peaks and assignments." FTIR Functional Group Database.
https://ftir.fun/ir/group/ami…

[9] Cocchi M et al. "Chemometric Methods for the Analysis of Food." Analytica Chimica Acta, 2018, 1015: 1–14. DOI:10.1016/j.aca.2018.02.039.

[10] General Administration of Quality Supervision, Inspection and Quarantine of the People's Republic of China. GB/T 22388—2008 Determination of Melamine in Raw Milk and Dairy Products. China Standard Press.

[11] ftir.fun. "Hydroxyl (O-H) — FTIR absorption peaks and assignments." FTIR Functional Group Database.
https://ftir.fun/ir/group/hyd…

[12] Guillén M D, Cabo N. "Characterization of Edible Oils and Lard by FTIR Spectroscopy." Journal of Agricultural and Food Chemistry, 1997, 45(11): 4495–4503. DOI:10.1021/jf970253t.

[13] ftir.fun. "Ester — FTIR absorption peaks and assignments." FTIR Functional Group Database.
https://ftir.fun/ir/group/est…

[14] Beć K B, Grabska J, Huck C W. "Quantitative Infrared Spectroscopy of Food: Principles and Applications." Spectroscopy, 2021, 36(4): 26–34.

[15]AOAC International. Official Methods of Analysis. 21st ed. 2023. Chapter 33 (Food Analysis).

Next episode preview: Ep 26 — Food Industry: Food Packaging Migration and Microplastic Detection
We will shift from "food adulteration" to "food contamination" — focusing on FPA-FTIR full filter membrane analysis of microplastics in takeaway containers, verification methods for microplastics in drinking water, and multi-city LDIR study of microplastics in carbonated beverages. Microplastic detection is a hot topic in food and environment fields.