Soup's On! How Chemical Imaging and Chemometrics are Revolutionizing Food Analysis
"Uncover hidden ingredients and ensure food safety with the power of hyperspectral chemical imaging."
Applications of hyperspectral chemical imaging (HCI) are rapidly expanding. Its integration into analytical chemistry relies heavily on chemometrics. A single hyperspectral image holds vast amounts of data, requiring chemometric techniques to extract meaningful chemical information. Chemometrics unlocks the understanding of HCI, and conversely, HCI enhances our understanding of chemometrics.
When exploring chemometric principles, visual perception plays a crucial role, particularly in interpreting scatter plots and histograms. HCI takes this further by introducing spatial images, speeding up analysis through visual processing. This allows for pattern recognition, spatial distribution analysis, and outlier detection. HCI's "image advantage" allows us to observe the effects of processing on familiar scenes.
This paper explores the fundamental properties of basic chemometric techniques by applying them to an example image: a near-infrared (NIR) hyperspectral image of a powdered soup mix contaminated with foreign objects like string, wood, and a paperclip. The HCI hypercube has spatial dimensions of 418 pixels × 318 pixels and 207 wavelength channels, acquired with a BurgerMetrics HyperPro NIR imaging system.
A Data-Rich Revolution
Hyperspectral imaging produces exceptionally rich measurements: a typical system with 580 x 580 pixel spatial resolution generates a 'hypercube' containing 336,400 spectra, each with more than 100 wavelength variables. This combination of spatial and chemical information enables HSI to outperform classical spectroscopic measurements and vision systems based solely on color. That richness comes at a price, however, as HSI requires powerful data analysis tools for interpretation and to facilitate its implementation in process analytical settings. To manage this complexity, researchers have proposed exploratory analysis approaches that summarize the main characteristics of hyperspectral imaging data and guide the choice of the most suitable modelling method.
Where Classical Tools Fall Short
Conventional food analysis has typically relied on classical spectroscopic techniques and standard color-based vision systems, which capture either chemical or visual appearance information but generally not both together. Such accepted methods are well established, yet they can overlook how chemical properties vary across a sample rather than just at a measured spot. As a result, quality decisions have often depended on sampling regimes or destructive tests that may not fully represent an entire batch. Newer chemical imaging approaches are being explored precisely because of these limitations, though they are not yet as universally adopted as the traditional toolkit.
From Remote Sensing to the Lab Bench
Hyperspectral imaging has evolved tremendously over the past fifty years, transforming from a specialized technique into one considered indispensable across a range of industries. At its foundation, the technology collects and processes information from across the electromagnetic spectrum with the goal of obtaining a spectrum for every pixel in an image, enabling the identification of objects, materials, and processes, and it comes in three general types of spectral imagers. A sign of the field's maturation is a Nature Reviews Methods Primers article introducing HSI through an imaging-centric perspective that links sensor platforms, data types, and representative datasets across application domains. Among the milestones along the way, the technology has found use in health, playing an instrumental role in cancer surgery by distinguishing tumour margins through blood flow differences between tumorous and healthy tissue.
Unveiling Food Secrets: Principal Components Analysis (PCA) with HCI
PCA is a cornerstone of chemometrics, ideal for exploring and predicting data. In NIR spectroscopy, some wavelength regions hold key information, while others primarily add noise. PCA compresses data, projecting high-dimensional spectral data into a lower-dimensional space, separating signal from noise. This process yields principal components (PCs) consisting of loading and score vectors, ordered to explain decreasing amounts of signal variance.
- PC-1: Captures maximum variance, often highlighting physical variations (e.g., soup mix surface).
- PC-2: Shows reduced contrast, making contaminants (e.g., paperclip) more distinguishable.
- Later PCs: Gradually decrease in contrast, dominated by specular reflection or noise, indicating diminishing signal content.
A Fast-Moving Research Landscape
Recent reviews show hyperspectral imaging expanding rapidly in scope and ambition. A 2024 review specifies the fundamentals of HSI and its familiar technologies while focusing on novel applications in food safety and quality assessment, medical analysis, agriculture, water resource management, plant stress identification, weed and crop discrimination, and flood management. In parallel, a December 2025 publication by Anna de Juan and Rodrigo Rocha de Oliveira of the Chemometrics Group at the Universitat de Barcelona argues that HSI, combined with chemometrics, represents a leap forward for industrial process analytical technology beyond classical spectroscopic PAT tools. Together, these works signal momentum on both the breadth of applications and industrial readiness.
The Skeptic's View
Critics observe that the very data richness that makes hyperspectral imaging attractive also creates practical burdens, since enormous hypercubes demand significant computing power, storage, and specialist chemometric expertise to interpret correctly. Outcomes can be sensitive to preprocessing and modelling choices, feeding concerns about reproducibility when methods move from controlled lab conditions to messy production environments. Instrument cost and the difficulty of transferring laboratory-built models onto factory lines are other recurring objections. Advocates respond that many of these hurdles are being actively addressed, but caution about real-world robustness persists in some quarters.
How HSI Stacks Up
Set against classical point spectroscopy, which measures chemistry at a single spot, hyperspectral imaging captures a full spectrum at every pixel and thereby reveals spatial variation that single-point methods cannot see. Compared with ordinary color-based machine vision, it looks beyond surface appearance toward underlying chemical composition. The trade-offs lie in speed, data volume, and equipment complexity, areas where simpler and cheaper techniques keep the advantage. Consequently, many practitioners regard HSI less as a wholesale replacement than as a powerful complement to established analytical methods.
HCI: A Powerful Tool for Chemometric Insights
HCI serves as an intuitive tool for investigating chemometrics. Our visual processing skills are crucial when understanding spatial scene details, while the detection of subtle differences in score images and prediction maps optimizes models and parameters, like latent variables or classification thresholds.
What the Experts Agree On
Expert commentary converges on the idea that hyperspectral imaging derives its value from uniting space and chemistry in a single measurement. As framed in 'A chemometric view of hyperspectral images,' hyperspectral images provide spatial and structural information on samples, and the main chemometric techniques have been assessed for their application and feasibility across different foodstuffs. Field syntheses likewise emphasize that hyperspectral image analysis research centers on adapting multivariate methods to data carrying both spatial and spectral dimensions. For food analysis in particular, commentators view the pairing of HSI and chemometrics as a natural fit whose full potential is still being systematically mapped.
Smaller, Smarter, Closer to the Line
The forward-looking consensus points to deep learning models for automated feature extraction and decision-making, alongside lightweight, portable HSI devices that would enable more on-site applications in agriculture, healthcare, and environmental monitoring. Industry observers additionally flag miniaturization, edge AI, and multi-modal fusion as defining trends, with one market outlook describing hyperspectral imaging as turning commercial scale in 2026. Advances in imaging sensors and data processing algorithms, coupled with integration of AI and machine learning, are expected to shape the market, aided by rising interest in early disease detection and life sciences research. On the commercial side, one market research report values the hyperspectral imaging market at $630.4 million in 2025 and projects growth to $1,825.3 million by 2035 at an 11.3% CAGR — figures that remain projections rather than settled outcomes.
Beyond the Sensor
Wider adoption of chemical imaging in food analysis depends on more than instrument performance, touching systemic issues such as standardization across laboratories, regulatory acceptance of image-derived evidence, and a workforce skilled in chemometrics and data science. Capital costs and data infrastructure requirements risk creating a divide between large producers who can invest and smaller operations that cannot. Open questions around model validation, data sharing, and interoperability between platforms remain unresolved at the industry level. How these ecosystem-level challenges play out may prove as decisive for adoption as progress in the underlying sensor technology.
Keeping Humans in the Loop
Behind the algorithms, researchers are working to keep people meaningfully involved in interpreting hyperspectral data. A 2022 Scientific Reports study on hyperspectral imaging for chemical identification took a human-centred approach and successfully applied its model to real data from the Cuprite mining site in Nevada, USA, collected in 1997 using an airborne visible-infrared imaging spectrometer (AVIRIS). Work of this kind underscores that interpretable outputs, not just raw predictions, are what allow scientists and inspectors to act on chemical images with confidence. In food analysis, the same human-machine partnership is what ultimately converts hypercubes of data into trustworthy decisions about safety and quality.
Techniques applied to PCA and PLS-DA are easily extended to other chemometric methods. As a powerful technique for quickly amassing large sets of spectral data, HCI is ideal for training and validating chemometric models.
The incorporation of HCI as a teaching aid in introductory chemometrics courses should be further developed, enhancing the understanding of chemical processes and food analysis techniques among emerging scientists and engineers.