Near-Infrared Spectroscopy: Principles and Applications

Near-Infrared Spectroscopy: Principles and Applications

Introduction

Near-infrared spectroscopy (NIRS) is a rapid, non-destructive analytical technique whose wavelength range covers 780-2500 nm (wavenumbers of about 12820-4000 cm⁻¹), between the visible and mid-infrared regions. Since Karl Norris first applied NIR to moisture analysis of agricultural products in the 1960s, and with advances in chemometrics, optical components and detector technology, near-infrared spectroscopy has become an indispensable process-analysis and quality-control tool in fields such as food, agriculture, pharmaceuticals, chemicals and the environment.

Unlike mid-infrared spectroscopy (which reflects the fundamental vibrations of molecules), near-infrared spectroscopy mainly reflects the overtone and combination band absorption of molecular vibrations; its peaks are broad and heavily overlapped, making direct assignment of functional-group characteristic peaks difficult in the way that is possible in the mid-infrared. However, it is precisely this "fingerprint-like" overall response, combined with multivariate calibration methods, that allows NIR spectroscopy to rapidly and simultaneously predict the content of multiple components in a sample, showing unique practical value.

1. Basic Principles of Near-Infrared Spectroscopy

1.1 Molecular Vibration and Near-Infrared Absorption

Atoms in a molecule undergo periodic vibrational motion around their equilibrium positions. Under the harmonic-oscillator approximation, the molecular vibrational energy levels are equally spaced, and the allowed transitions occur only between adjacent levels (i.e. $\Delta v = \pm 1$); in this case the absorption frequency equals the fundamental vibrational frequency of the molecule - which corresponds to absorption in the mid-infrared region.

However, the vibration of a real molecule is not strictly harmonic. Under the anharmonic potential-energy function of a real molecule:

• Overtone absorption: the transition occurs at $\Delta v = \pm 2, \pm 3, \dots$, corresponding to integer multiples of the fundamental frequency, with the intensity decreasing rapidly as the overtone order increases;
• Combination band absorption: the quantum numbers of two or more normal modes change simultaneously, and the absorption frequency equals the sum or difference of the corresponding fundamental frequencies.

These overtone and combination band absorptions fall exactly in the near-infrared region (780-2500 nm), constituting the physical basis of near-infrared spectroscopy.

1.2 Main Absorbing Groups

The absorption intensity in the near-infrared region is closely related to the anharmonicity constant of the chemical bonds. Because the hydrogen atom has a small mass and a large vibrational amplitude, hydrogen-containing groups exhibit the most significant anharmonicity; therefore near-infrared spectra mainly reflect the information of the following hydrogen-containing groups:

Among these, the strong absorption peaks of water near 1450 nm and 1940 nm are among the most prominent spectral features in near-infrared analysis, so that measuring trace components in water-containing samples requires higher-sensitivity instruments and more refined modeling.

1.3 Measurement Modes

Depending on the sample form and optical characteristics, near-infrared spectroscopy has four main measurement modes:

• Transmission measurement: suitable for transparent liquids or thin-film samples; light passes through the sample to reach the detector, with a fixed optical path length, obeying the Beer-Lambert law;
• Diffuse reflectance measurement: suitable for opaque samples such as solid powders and granules; the detector collects the diffuse light reflected from the sample surface, commonly used for the analysis of grains, feed, soil, etc.;
• Diffuse transmittance measurement: the light source and detector are on opposite sides of the sample, and the detector receives the scattered light transmitted through the sample, suitable for semi-transparent samples of a certain thickness (such as internal fruit quality inspection);
• Transflectance measurement: light enters the sample, is scattered, then returns through the incident surface carrying sample information and is detected, commonly used for turbid liquids, emulsions, etc.

2. Dispersion Methods of Near-Infrared Spectrometers

The core function of a near-infrared spectrometer is to disperse polychromatic light by wavelength and measure the light intensity at each wavelength. Depending on the dispersion principle, common schemes include the filter type, the grating-scanning type, the DLP/DMD digital micromirror type and the Fourier-transform type.

2.1 Filter Type

Fixed-wavelength filter-type instruments use a set of narrow band-pass filters (interference filters) to select specific wavelengths for measurement. They are simple in structure and low in cost, and are suitable for dedicated online instruments, but have limited wavelength points and poor flexibility. Although a filter wheel can hold multiple filters, its mechanical movement affects repeatability and speed.

2.2 Grating-Scanning Type

The grating-scanning type is the most classical dispersion scheme for near-infrared spectrometers, using the diffraction effect of the grating to disperse polychromatic light into monochromatic light arranged spatially:

• Plane blazed grating: through a precisely designed blaze angle, the diffracted energy is concentrated into a specific order and wavelength range; combined with a slit and a rotating mechanism it realizes wavelength scanning, and is widely used in benchtop high-precision laboratory instruments;
• Concave grating: combines dispersion and imaging functions, simplifying the optical path, reducing the number of reflecting surfaces and lowering stray light, suitable for portable and miniaturized devices.

The groove density of the grating (usually 300-1200 lines/mm) determines the dispersion capability and wavelength coverage: a high groove density provides higher resolution but narrows the free spectral range, requiring a trade-off.

2.3 DLP/DMD Digital Micromirror Type

The application of Digital Light Processing (DLP) technology in near-infrared spectrometers is an important development direction in recent years. Its core lies in using Texas Instruments' (TI) Digital Micromirror Device (DMD) as a programmable wavelength selector - the spectrum dispersed by the grating is projected onto the DMD array, and the on/off state of each micromirror is controlled by software to select specific wavelengths or wavelength combinations for reflection to the detector.

The essential difference between the DLP scheme and the grating-scanning type lies in the detector configuration:

Another prominent advantage of the DLP scheme is its programmability: users can flexibly adjust the scanned wavelength range and step size through software, shrinking the step size when high resolution is needed, and enlarging the step size while increasing throughput when fast measurement is needed. In addition, Hadamard encoding turns on multiple micromirror columns simultaneously for encoded measurement and recovers the spectrum through a matrix transform, greatly shortening the measurement time while maintaining the same signal-to-noise ratio.

2.4 Fourier-Transform Type (FT-NIR)

Fourier-transform near-infrared spectrometers use a Michelson interferometer structure, generating an interferogram through moving-mirror scanning and obtaining the spectrum via Fourier transform. FT-NIR has the multiplex (Fellgett) advantage and the throughput (Jacquinot) advantage, with high wavelength accuracy and adjustable resolution, and is a common choice for high-precision laboratory analysis. However, it is sensitive to vibration and ambient temperature, making miniaturization difficult.

3. Key Performance Specifications

3.1 Wavelength Range

The near-infrared region is usually divided into:

• Short-wave NIR (780-1100 nm): silicon-based detectors can be used, with a large penetration depth, suitable for transmission and diffuse-transmittance measurements;
• Long-wave NIR (1100-2500 nm): InGaAs or extended InGaAs detectors are required, with rich information content; this is the mainstream analysis range.

Complete 780-2500 nm coverage can obtain the higher-order and lower-order overtone and combination band information of hydrogen-containing groups, giving stronger discrimination between different components.

3.2 Spectral Resolution

Resolution refers to the spectrometer's ability to distinguish adjacent absorption peaks, usually expressed in nanometers (nm) or wavenumbers (cm⁻¹). Near-infrared spectral peaks are relatively wide (usually tens of nanometers), so the resolution requirement is less stringent than for the mid-infrared - a resolution of 5-20 nm already meets most quantitative analysis needs. However, excessively high resolution increases noise and scanning time; in practice a balance must be struck among resolution, signal-to-noise ratio and measurement speed.

3.3 Signal-to-Noise Ratio (SNR)

The signal-to-noise ratio is the core indicator of instrument stability and detection sensitivity. A high SNR is crucial for detecting low-concentration components and building robust prediction models. SNR is affected by multiple factors such as light-source stability, detector sensitivity, throughput and circuit noise. The DLP/DMD scheme, using a large-area single-point detector and Hadamard encoding, has a natural advantage in signal-to-noise ratio.

4. Chemometrics - From Spectra to Results

Near-infrared spectroscopy is an indirect analytical technique: it cannot directly obtain concentration or content as titration or chromatography can, but must use chemometric methods to build a correlation model between the spectrum and the property to be measured. A typical modeling workflow is as follows:

1. Sample spectrum acquisition: select a representative calibration set of samples (usually 50-200) and acquire their near-infrared spectra;
2. Reference method determination: use standard wet-chemical or instrumental methods (such as Kjeldahl nitrogen determination, oven-drying, HPLC, etc.) to determine the content or property of the target component in each sample;
3. Spectral preprocessing: apply preprocessing such as smoothing, derivation (Savitzky-Golay), standard normal variate (SNV) or multiplicative scatter correction (MSC) to the raw spectra to eliminate interferences such as baseline drift and scattering effects;
4. Build the prediction model: use multivariate calibration methods to build a mathematical model; commonly used quantitative methods include:
   - Principal component regression (PCR): first decompose the spectral matrix into principal components, then regress against concentration;
   - Partial least squares regression (PLSR): decompose the spectral matrix and the concentration matrix simultaneously and extract latent variables; currently the most widely used method;
   - Support vector regression (SVR), artificial neural networks (ANN) and other machine-learning methods, suitable for nonlinear systems;
5. Model validation and evaluation: use an independent validation set to evaluate the model's predictive ability; commonly used indicators include the coefficient of determination ($R^2$), the root mean square error of prediction (RMSEP) and the relative prediction deviation (RPD);
6. Unknown sample prediction: acquire the spectrum of an unknown sample, substitute it into the established model, and output the prediction result in real time.

An excellent chemometric model is the key to the success of near-infrared analysis - the "garbage in, garbage out" principle applies here as well; the representativeness of the calibration samples and the accuracy of the reference method directly determine the usability of the model.

5. Application Areas

5.1 Agriculture and Food

Agriculture is the earliest and most mature application area of near-infrared spectroscopy:

• Grain analysis: rapid determination of the moisture, protein, starch, fat and fiber content of grains such as wheat, corn and rice; a standard method in grain purchase, storage and trade;
• Oilseed crops: oil content and fatty-acid composition analysis of soybean, rapeseed and peanut;
• Feed industry: simultaneous determination of multiple indicators such as crude protein, crude fiber, ash and moisture;
• Fruit quality: use of the diffuse-transmittance mode to detect the soluble solids (sugar content), acidity and internal defects of fruit;
• Dairy products: online and offline detection of protein, fat, lactose and total solids in milk powder and liquid milk;
• Meat: prediction of moisture, fat and protein content, and even tenderness and water-holding capacity.

The core advantage of near-infrared spectroscopy in the food field is that it is fast (seconds to tens of seconds), requires no sample preparation (or only simple grinding), and can output results for multiple components simultaneously, significantly reducing laboratory labor and consumable costs.

5.2 Pharmaceutical Industry

In the pharmaceutical field, near-infrared spectroscopy is widely used for:

• Raw and auxiliary material identity verification: quickly confirm the type and purity of raw and auxiliary materials upon warehousing to prevent confusion;
• Blend uniformity monitoring: online monitoring of the uniformity of active-ingredient distribution during powder blending to determine the blending endpoint;
• Moisture determination: determination of the drying-process endpoint and finished-product moisture quality control;
• Tablet content uniformity: non-destructive testing of whether the active pharmaceutical ingredient (API) content in tablets is uniform;
• Coating thickness and endpoint determination: online monitoring of film thickening during the film-coating process.

The pharmaceutical industry has also been a significant driver of near-infrared technology - the promotion of the process analytical technology (PAT) concept and the recognition of regulatory agencies (such as the FDA) have made near-infrared spectroscopy an indispensable part of modern pharmaceutical quality-control systems.

5.3 Petrochemical

• Rapid determination of gasoline octane number, distillation range, benzene content, aromatic/olefin content and other indicators, replacing traditional time- and labor-intensive standard methods;
• Simultaneous prediction of multiple indicators of diesel cetane number, density, distillation range and sulfur content;
• Polymers: online monitoring of the density, melt index and comonomer content of products such as polyethylene and polypropylene;
• Chemical intermediates: reaction-process monitoring, endpoint determination and byproduct content control.

5.4 Environmental Monitoring

• Rapid detection of soil fertility indicators such as organic matter, total nitrogen and available phosphorus, serving precision agriculture;
• Rapid screening of water pollution indicators such as chemical oxygen demand (COD), total nitrogen and total phosphorus;
• Determination of the calorific value of solid waste for feed management in waste-incineration power plants.

5.5 Other Applications

• Textiles: rapid identification of cotton/polyester blend ratio and fiber type;
• Wood and paper: wood density, moisture, and pulp kappa number;
• Bioenergy: fatty-acid methyl ester content of biodiesel, and sugar/alcohol concentration monitoring in ethanol fermentation processes;
• Clinical and biomedical: non-invasive blood-glucose detection, blood oxygen saturation, and tissue oxygenation assessment.

6. Advantages and Limitations of Near-Infrared Spectroscopy

Advantages

• Fast: a single measurement usually takes only seconds, suitable for large batches of samples and online real-time analysis;
• Non-destructive: samples remain unchanged after measurement, with no chemical reagent consumption and no waste generation - truly green and environmentally friendly;
• No sample preparation: solid, liquid, powder and slurry samples can be measured directly;
• Simultaneous multi-component analysis: a single spectrum can predict the content of multiple components, far more efficient than traditional methods that measure them one by one;
• Suitable for online/field analysis: remote and continuous monitoring can be achieved through fiber probes or online flow cells.

Limitations

• Model-dependent: chemometric models must be built and maintained; the calibration workload is large and model transfer must be done with care;
• Limited sensitivity: the detection limit is usually on the order of 0.1%, unsuitable for trace analysis (ppm and below);
• Water interference: water absorbs strongly with a broad spectral band and may mask the information of other components; trace-component analysis in high-moisture samples is difficult;
• High model specificity: a model usually applies only to a specific type of sample; changing the sample type requires rebuilding the model.

7. Summary and Outlook

After more than half a century of development, near-infrared spectroscopy has moved from the laboratory to the field, from offline to online, and from single-indicator to simultaneous multi-parameter analysis. Its fast, non-destructive, green and efficient characteristics make it play an irreplaceable role across the entire agriculture-food industry chain from field to table, the pharmaceutical manufacturing process from raw materials to finished products, and the petrochemical processing from crude oil to polymers.

Looking ahead, the development trends of near-infrared spectroscopy include:

• Miniaturization and portability: MEMS technology, miniaturized spectrometer chips and smartphone spectral accessories will make field rapid testing more widespread;
• Intelligent modeling: deep learning and big-data technologies will improve model prediction accuracy and reduce the reliance on manually screened calibration samples and professional modeling knowledge; model auto-updating and transfer technologies will also mature;
• Multi-spectral fusion: the fusion of near-infrared with Raman spectroscopy, laser-induced breakdown spectroscopy (LIBS), hyperspectral imaging and other technologies will provide more comprehensive, multi-dimensional sample information;
• Cloud platforms and the Internet of Things: cloud storage, analysis and sharing of spectral data, combined with the Internet of Things, will enable remote modeling, model distribution and instrument health management.

As the performance of core components continues to improve and data-modeling methods continue to advance, near-infrared spectroscopy will surely serve scientific research and industrial production in ever wider fields and to ever greater depths, providing the key sensing capability for intelligent manufacturing and precision quality inspection.

This article was compiled by Pynect.