Near-infrared spectroscopy, almost always shortened to NIRS, is a technique that shines light in the near-infrared wavelength range (roughly 700 to 2500 nanometers) through or onto a sample and reads what comes back. Because different chemical bonds absorb specific wavelengths, the returning light carries a fingerprint of what the sample contains. That simple principle has made NIRS one of the most versatile analytical tools in science and industry, turning up in hospital operating rooms, pharmaceutical factories, Mars rovers, and even the gym. The technology is fast, usually non-destructive, and increasingly portable, which explains why its reach keeps expanding.
What the Light Actually Tells You
Near-infrared light interacts primarily with bonds between hydrogen and heavier atoms like carbon, nitrogen, and oxygen. When a molecule absorbs NIR photons, those bonds vibrate at overtone and combination frequencies of their fundamental vibrations. Because these overtone absorptions are weaker than the fundamental absorptions you would see in the mid-infrared range, NIR light penetrates deeper into a sample before being fully absorbed. That deeper penetration is a practical advantage: you can analyze the inside of a tablet, a piece of fruit, or a slab of muscle tissue without cutting anything open.
The trade-off is that NIR spectra are broad and overlapping. You rarely see a single clean peak belonging to just one compound. Instead, the spectrum is a dense landscape of merged signals. Extracting useful information from that landscape almost always requires mathematical modeling, which is why NIRS and chemometrics (the statistical science of chemical data) are essentially joined at the hip. More on that below.
Brain Imaging With fNIRS
One of the most high-profile uses of NIRS is functional near-infrared spectroscopy, or fNIRS, which measures brain activity. The technique works because near-infrared light passes through the skull well enough to reach the outer layers of the brain, and oxygenated and deoxygenated hemoglobin absorb NIR wavelengths differently. When a brain region becomes active, blood flow to that area increases, changing the local ratio of these two forms of hemoglobin. By placing light sources and detectors on the scalp, fNIRS tracks those changes in real time.1PubMed Central. Application of Functional Near-Infrared Spectroscopy to the Study of Brain Function in Humans and Animal Models
fNIRS does not replace fMRI, which provides much finer spatial resolution. But it fills gaps that fMRI cannot. An MRI scanner is a multi-ton machine bolted to a room; an fNIRS headset can be worn while walking, talking, or interacting with other people. It tolerates motion far better than fMRI, making it especially useful for studying infants, children, surgical patients, and anyone who cannot lie still inside a loud tube for extended periods.2PubMed Central. Functional Magnetic Resonance Imaging and Functional Near-Infrared Spectroscopy: Insights from Combined Recording Studies Researchers have used fNIRS to study everything from language processing in newborns to executive function in older adults performing cognitive tasks.3PubMed. Age dependency of the hemodynamic response as measured by functional near-infrared spectroscopy
Monitoring Oxygen in Surgery and Intensive Care
Outside the research lab, NIRS has become a bedside tool in hospitals. The most common clinical application is monitoring cerebral oxygen saturation during surgery. In cardiac operations, for example, the brain can be vulnerable to drops in oxygen supply, and a pulse oximeter on your finger only tells you what is happening in the blood generally, not in the brain specifically. NIRS sensors placed on the forehead provide a continuous, real-time read of how well the brain tissue is oxygenated.4PubMed Central. Monitoring tissue oxygenation by near infrared spectroscopy (NIRS): background and current applications
Neonatal intensive care units have adopted the technology for a similar reason. Blood oxygen levels measured at the wrist or foot do not always reflect what is happening inside a premature baby’s brain. NIRS monitoring can flag when cerebral oxygenation drifts outside a safe range, prompting clinicians to adjust ventilator settings or medications.5PubMed Central. Monitoring Cerebral Oxygenation in Neonates: An Update Clinical trials have tested protocols in which interventions are triggered when cerebral oxygenation drops below about 55% or rises above 85% in extremely premature infants during the first days of life.6Pediatric Research. Cerebral near-infrared spectroscopy monitoring (NIRS) in children and adults: a systematic review with meta-analysis
Exercise and Sports Science
Athletes and sports scientists have embraced wearable NIRS devices that clip onto a muscle and report its oxygenation level during exercise. The principle is the same one used in the brain: NIR light penetrates the skin and muscle, and the sensor reads how much oxygenated versus deoxygenated hemoglobin is present. During intense effort, muscles consume oxygen faster than the blood can deliver it, and the sensor captures that decline in real time.7PubMed. Reliability of near-infrared spectroscopy in measuring muscle oxygenation during squat exercise
This gives coaches and physiologists a window into what is happening metabolically inside the working muscle, not just at the heart or lungs. You can see how quickly oxygen drops when a sprint starts, how quickly it recovers between intervals, and how different training loads stress different energy systems. The devices are now inexpensive enough that university physiology courses use them as teaching tools, letting students visualize the concept that all energy production in the body ultimately depends on oxygen.8PubMed. All about oxygen: using near-infrared spectroscopy to understand bioenergetics
Pharmaceutical Manufacturing
Drug manufacturing is one of the oldest industrial homes for NIRS. When you blend powders to make a tablet, the active ingredient needs to be evenly distributed before you compress the mix. Traditionally, you would stop the blender, pull samples at intervals, and run lab tests. NIRS sensors mounted inside the blender can do the same job continuously, predicting blend uniformity in real time and flagging the moment the mix is homogeneous.9PubMed. A Process Analytical Technology approach to near-infrared process control of pharmaceutical powder blending. Part III: Quantitative near-infrared calibration for prediction of blend homogeneity and characterization of powder mixing kinetics
The U.S. FDA’s Process Analytical Technology initiative specifically encouraged this kind of shift from after-the-fact testing to real-time monitoring. Pattern recognition algorithms trained on NIR spectra from known good and bad blends can classify new batches on the fly, reducing waste and catching problems before a million defective tablets roll off the line.10PubMed. A Process Analytical Technology approach to near-infrared process control of pharmaceutical powder blending: Part II: Qualitative near-infrared models for prediction of blend homogeneity
NIRS is also used to check finished tablets. However, the measurement mode matters. Shining light onto the surface (diffuse reflectance) only captures information from the outer layer, which can be a coating that tells you nothing about the drug inside. Sending light through the entire tablet (transmission mode) interrogates a much larger volume and tends to give better predictions of the active ingredient content, especially for coated or multilayer tablets.11PubMed. Assessment of diffuse transmission and reflection modes in near-infrared quantification, part 2: DIFFuse reflection information depth
Food Authentication and Safety
Food fraud is a surprisingly large problem. Spices get bulked up with cheaper dyes, milk powder gets mixed with starch, and honey gets diluted with sugar syrup. Laboratory tests can catch these tricks, but they are slow and expensive. NIRS offers a fast screening alternative: scan a sample, compare its spectral fingerprint to a database of known pure and adulterated samples, and flag anything suspicious.
A global inter-laboratory study tested whether handheld NIR devices could detect adulterated food samples. Across 34 portable instruments from different manufacturers, 33 correctly classified five out of six test samples, and every adulterated sample was identified after applying appropriate data-processing steps.12PubMed. The potential of handheld near infrared spectroscopy to detect food adulteration: Results of a global, multi-instrument inter-laboratory study In a focused study on paprika powder, a portable NIR device distinguished pure paprika from samples spiked with cheaper colorants at specificity above 90% and error rates below 2%.13Journal of Food Composition and Analysis. Portable near-infrared spectroscopy for rapid authentication of adulterated paprika powder
Beyond fraud detection, NIRS finds internal defects in fruit that are invisible from the outside. Pears, for example, can develop a disorder called brown heart that you would never notice until you cut one open. Visible and near-infrared spectroscopy in transmission mode classified affected pears with about 91% accuracy, and during validation no defective pear was missed.14Journal of Food Engineering. Comparison of diffuse reflectance and transmission mode of visible-near infrared spectroscopy for detecting brown heart of pear These kinds of applications are being explored for everything from detecting bruised apples on a packing line to estimating sugar content in grapes before harvest.15PubMed Central. Non-Destructive Detection of Fruit Quality: Technologies, Applications and Prospects
Agriculture and Soil Analysis
Soil scientists have used visible and NIR spectroscopy to estimate soil organic carbon content for decades, because organic matter absorbs NIR light in characteristic ways. A meta-analysis pooling results from 134 studies published between 1986 and 2022 found statistically significant differences in model performance depending on the quantitative metrics and modeling approaches used, confirming that while the technique works, the details of how you build the prediction model matter a great deal.16SpringerLink / Pleiades Publishing (Eurasian Soil Science). Vis-NIR Spectroscopy for Soil Organic Carbon Assessment: A Meta-Analysis The appeal is speed: a single NIR scan of a soil sample can replace wet chemistry analyses that take hours.
Wine is another agricultural product where NIRS has been tested. Researchers built models to predict mineral content (calcium, iron, potassium, and others) from the NIR spectra of wine samples, achieving strong correlations for several elements.17PubMed. Analysis of elements in wine using near infrared spectroscopy and partial least squares regression The fact that NIRS can pick up signals related to inorganic elements, even though it primarily interacts with organic bonds, is somewhat counterintuitive. The explanation is that mineral ions are often associated with organic molecules, so the technique is really detecting those organic matrix effects rather than the ions themselves.
Transmission Versus Reflectance
How you deliver and collect the light shapes what NIRS can see. In reflectance mode, the light source and detector sit on the same side of the sample. Light enters, bounces around, and some of it scatters back out toward the detector. In transmission mode, the source and detector face each other on opposite sides, and the light passes all the way through. Both approaches have their uses, but they are not interchangeable.
Reflectance mode is convenient for surfaces and irregularly shaped objects, but it only probes a thin layer. For pharmaceutical tablets, reflectance barely penetrates past a coating.18PubMed. Assessment of diffuse transmission and reflection modes in near-infrared quantification, part 2: DIFFuse reflection information depth Transmission mode interrogates the full depth of the sample, which explains why multiple studies have found it produces better predictions for internal quality measurements. For metoprolol tablets, transmission-mode NIR outperformed reflectance, largely because reflectance was too sensitive to the physical unevenness of the tablet surface, while transmission averaged those effects over a larger volume.19Journal of Pharmaceutical and Biomedical Analysis. Vibrational spectrometry for the assessment of active substance in metoprolol tablets: a comparison between transmission and diffuse reflectance near-infrared spectrometry A similar advantage showed up when detecting internal browning in pears.20Journal of Food Engineering. Comparison of diffuse reflectance and transmission mode of visible-near infrared spectroscopy for detecting brown heart of pear
The practical takeaway: if you care about what is on or near the surface, reflectance can work well. If you need to know what is inside, transmission is usually the stronger choice.
Why Raw Spectra Are Not Enough
A common misconception is that NIRS works like a barcode scanner: point, scan, get an answer. In reality, raw NIR spectra are messy. Even though the underlying physics follows the Beer-Lambert law relating absorbance to concentration, real-world spectra are contaminated by scattering effects, baseline shifts, temperature changes, and noise.21PubMed Central. Preprocessing NIR Spectra for Aquaphotomics Water is a particularly stubborn interference. It absorbs strongly across much of the NIR range, its scattering properties change with concentration, and when water content varies between samples, the relationship between absorbance and analyte concentration becomes nonlinear.22PubMed. Relating Near-Infrared Light Path-Length Modifications to the Water Content of Scattering Media in Near-Infrared Spectroscopy: Toward a New Bouguer-Beer-Lambert Law
Getting around these problems requires preprocessing: mathematical treatments applied to the spectra before any prediction model is built. Techniques like standard normal variate correction, derivatives, and wavelet-based filters strip away unwanted variation while preserving the chemical information. One hybrid approach, combining wavelet-based water-signal removal with orthogonal signal correction, has been shown to substantially improve prediction accuracy in aqueous NIR data by eliminating the major interference sources before the model even sees the spectra.23PubMed. Removal of major interference sources in aqueous near-infrared spectroscopy techniques
After preprocessing, a statistical model links the cleaned spectra to the property you want to predict. The workhorse algorithm for decades has been partial least squares regression, which handles the high-dimensional, heavily correlated nature of spectral data well.24PubMed. A step-by-step tutorial to local partial least squares analyses for near-infrared spectroscopic data analysis Building the model requires a calibration set of samples whose properties have been measured by a reference method, and the model’s accuracy is only as good as that calibration data. This is a point that sometimes gets lost in the enthusiasm around NIRS: the technique does not eliminate the need for conventional analysis; it automates the prediction once the hard work of calibration is done.
Portable Devices and Where They Struggle
Miniaturization has pushed NIRS out of the laboratory. Handheld spectrometers now weigh less than a smartphone and cost a fraction of benchtop instruments. Farmers can scan grain in a silo, customs officials can check a suspicious powder at a border crossing, and quality control inspectors can test incoming ingredients on a loading dock.25PubMed Central. Recent Advances in Portable and Handheld NIR Spectrometers and Applications in Milk, Cheese and Dairy Powders
But smaller hardware means compromises. Portable devices typically cover a narrower wavelength range, have lower spectral resolution, and use less sensitive detectors than their benchtop cousins. A model built on a laboratory spectrometer may not transfer cleanly to a handheld device without additional calibration steps. The global food-adulteration study mentioned earlier showed that appropriate preprocessing routines could bring portable instruments into line, but that word “appropriate” does a lot of heavy lifting: someone still needs to develop, validate, and maintain those routines for each instrument and each application.26PubMed. The potential of handheld near infrared spectroscopy to detect food adulteration: Results of a global, multi-instrument inter-laboratory study Comparing benchtop and portable NIR performance is an active area of research in food authentication, and the gap is narrowing but has not closed.27PubMed Central. Detection of Adulterants in Powdered Foods Using Near-Infrared Spectroscopy and Chemometrics: Recent Advances, Challenges, and Future Perspectives
Deep Learning as a Replacement for Traditional Modeling
The classic workflow for NIRS involves manually selecting which wavelengths to use and which preprocessing steps to apply before feeding data into a regression model. That pipeline requires expertise and judgment calls at every step. Convolutional neural networks offer an alternative: hand the raw spectrum to the network and let it learn which features matter on its own. A study testing this end-to-end approach found that a CNN trained on full raw spectra outperformed traditional partial least squares models by a statistically significant margin. When the traditional models were combined with carefully chosen wavelength selection, however, the gap disappeared.28Journal of Chemometrics. End‐to‐end quantitative analysis modeling of near‐infrared spectroscopy based on convolutional neural network
The practical implication is that deep learning could lower the skill barrier for building NIRS models. You no longer need a chemometrics specialist to hand-pick wavelengths and preprocessing routines if a neural network can learn those transformations from data. That said, CNNs require large training datasets and their internal workings are harder to interpret, which matters in regulated industries like pharmaceuticals where you may need to explain why the model makes a given prediction.29Journal of Near Infrared Spectroscopy. Review: The evolution of chemometrics coupled with near infrared spectroscopy for fruit quality evaluation. II. The rise of convolutional neural networks
Hyperspectral Imaging and Spatial Information
Standard NIRS measures a single spectrum from a spot on the sample. Hyperspectral NIR imaging collects a full spectrum at every pixel of an image, combining chemical fingerprinting with spatial mapping. Instead of getting one average reading for a whole tablet or food sample, you get a map showing where each component is distributed across the surface or cross-section. This is valuable when uniformity is the question: Is the drug evenly spread through the tablet? Are contaminants localized in one area of a grain sample?30PubMed. Near infrared hyperspectral imaging in quality and safety evaluation of cereals
Calibrating hyperspectral images is more complex than calibrating single-point spectra because the instrument introduces spatial variation across its field of view. Pixel-level calibration methods that account for both column position and wavelength have been shown to reduce prediction errors and preserve the spatial detail that makes hyperspectral imaging worthwhile in the first place.31PubMed. Hyperspectral near infrared image calibration and regression
NIRS on Other Planets
The same physics that lets a wearable sensor read your muscle oxygenation works on the surface of Mars. NASA’s Opportunity rover carried a multispectral camera covering visible and near-infrared wavelengths, using them to characterize the mineral composition of rocks at Meridiani Planum.32Journal of Geophysical Research: Planets. Visible and near‐infrared multispectral analysis of rocks at Meridiani Planum, Mars, by the Mars Exploration Rover Opportunity China’s Zhurong rover carries a dedicated NIR spectrometer as part of its surface composition detector, measuring reflectance spectra that, after careful in-flight calibration, are used to identify minerals on small scales and to cross-check orbital remote sensing data.33Earth and Space Science. In‐Flight Calibration of Near‐Infrared Reflectance Spectra Measured by the Zhurong Mars Rover
Planetary NIRS faces calibration challenges that terrestrial labs never worry about. There is no standard reference tile sitting on the Martian surface, temperature swings are extreme, and dust accumulation changes the optical properties of the instrument over time. The calibration pipelines developed for rover data share conceptual DNA with the preprocessing steps used in food or pharmaceutical NIRS, but the stakes and constraints are dramatically different. Getting the calibration right is not about optimizing profit on a production line; it is about correctly reading the geological history of another world.