Foundations: Spectroscopy Basics¶

This chapter introduces vibrational spectroscopy for food science: what spectra are, how wavenumbers are used, and how Raman, FTIR, and NIR differ. It anchors the physics so later preprocessing and ML chapters have a common starting point.

1. What is a spectrum?¶

A spectrum plots intensity vs wavenumber (cm⁻¹). Wavenumber ( \tilde{\nu} = 1/\lambda ) is preferred because it scales linearly with energy.
Peaks correspond to vibrational modes of molecules (stretching, bending). Food matrices contain lipids, proteins, carbohydrates, water—each with characteristic bands.
Always store axes in ascending cm⁻¹ for computational pipelines.

2. Raman vs FTIR vs NIR¶

Raman (inelastic scattering): Measures shifts relative to laser line (Stokes/anti-Stokes). Good for aqueous samples; sensitive to symmetric stretches (e.g., C=C, CH).
FTIR (absorption): Measures molecular absorption; ATR-FTIR is common in food labs. Strong for polar bonds (C=O, O–H).
NIR (overtones/combination bands): Broader, weaker features; useful for bulk composition and rapid QC.

Typical food spectral regions (examples)¶

Fingerprint (600–1800 cm⁻¹): C–C, C–O, C=O; unsaturation bands (≈1655–1745 cm⁻¹) in oils; amide bands (protein) around 1650/1550 cm⁻¹.
CH stretching (2800–3100 cm⁻¹): Lipid/protein CH2/CH3 bands.
OH/NH (3200–3600 cm⁻¹): Water/protein hydrogen bonding (FTIR).

3. Peak shapes, baselines, and artifacts¶

Peaks/bands: Can be sharp (Raman) or broad (NIR). Shoulders often encode overlapping modes.
Baseline & fluorescence: Raman often has fluorescence backgrounds; FTIR can show sloping baselines due to ATR contact or scattering.
Atmospheric lines: Water/CO₂ in FTIR; remove or account for them in preprocessing.
Noise & scatter: Instrument noise, cosmic rays (Raman spikes), path-length/contact variation.

3a. Vibrational modes and spectral signatures¶

Stretching vs bending: stretching changes bond length; bending changes bond angles. Raman favors polarizability changes (e.g., C=C), FTIR favors dipole changes (e.g., O–H).
Food-relevant bands (cm⁻¹, illustrative):
FTIR synthetic example (generated via generate_synthetic_ftir_spectrum): O–H stretch (~3300), C–H stretches (2800–3000), ester C=O (~1740), CH₂ bend (~1450), C–O stretch (~1050), fingerprint 800–1500. Plot wavenumber vs absorbance and label each band with the mode and a food interpretation (e.g., ester C=O in lipids).
Raman synthetic example (generated via generate_synthetic_raman_spectrum): discrete peaks at ~717 (C–C stretch), 1265 (cis =C–H bend), 1440 (CH₂ bend), 1655 (C=C stretch). Annotate peaks and note how intensity shifts relate to unsaturation/saturation.
Interpretation: shifts or intensity changes in these bands map to composition (unsaturation, ester content, moisture). Synthetic plots (see plotting helpers) mirror real bands observed in oils/fats.
For notation/abbreviations, see the Glossary. For a practical bands/ratios guide, see Feature extraction.

4. Sampling and instrument notes¶

Laser wavelength (Raman) affects fluorescence and penetration; ATR crystal choice (FTIR) affects depth of penetration.
Resolution: finer spacing yields more data points but may increase noise.
Export formats: vendor-specific to TXT/CSV. FoodSpec standardizes via CSV → HDF5; see CSV → HDF5 pipeline.

5. Choosing a modality for food tasks¶

Authentication/adulteration: Raman/FTIR fingerprint region for oils, spices; NIR for rapid screening.
Heating/oxidation studies: Track unsaturation bands (1650–1750 cm⁻¹) and CH stretches.
Protein-rich samples (dairy/meat): Amide bands (FTIR/Raman); CH stretches.
Water-dominated matrices: Raman often preferred to avoid strong water absorption in FTIR.

6. Links to computation¶

Baseline drift and fluorescence → Baseline correction.
Scatter/contact effects → Normalization & smoothing and Scatter & cosmic-ray handling.
High dimensionality → PCA.

Summary¶

Wavenumber in cm⁻¹ is the standard axis; keep spectra monotonic.
Raman, FTIR, and NIR emphasize different vibrational modes; choose modality by matrix and question.
Baselines, fluorescence, atmospheric lines, and scatter are common artifacts to mitigate in preprocessing.

When Results Cannot Be Trusted¶

⚠️ Red flags for spectroscopy basics application:

Spectral resolution insufficient for features of interest (broad bands analyzed as if sharp peaks)
Overlapping peaks unresolved; chemical assignment ambiguous
Information loss
Fix: Use higher-resolution spectrometer; deconvolve overlapping peaks; document resolution limits
Peak assignments based on single literature source without validation
Literature assignments may be context-dependent (different matrix, conditions)
Misassignment common
Fix: Cross-reference multiple sources; validate with isotopic substitution or known standards
Temperature not controlled (samples measured at varying room temperatures)
Temperature affects peak positions and intensities
Introduces uncontrolled variability
Fix: Control sample temperature; document temperature; use thermostated stage
Spectral saturation undetected (detector saturated; spectra clipped)
Saturated spectra lose quantitative information
Ratios biased by clipping
Fix: Check detector counts; reduce integration time or laser power if saturated; re-measure
Fluorescence not distinguished from Raman/FTIR signal (strong background mistaken for chemical information)
Fluorescence dominates weak Raman; obscures peaks
Can create false features
Fix: Use longer-wavelength excitation; subtract fluorescence baseline; validate with fluorescence measurement
No replicate measurements (single spectrum treated as ground truth)
Measurement noise unquantified; reproducibility unknown
Single outlier can bias analysis
Fix: Measure ≥3 replicates; report SD; average replicates for analysis