From manual and reactive checks to advanced and preventive methods, integrated with artificial intelligence and the IoT for real-time analysis, thanks to hyperspectral sensors capable of detecting various contaminants, ensuring food safety and reducing waste.

There are numerous inspection techniques available to food companies to ensure product quality and safety, including X-rays, thermal methods, ultrasound, fluorescence and infrared imaging. These are used to identify foreign contaminants in food packaging and assess food quality during processing. Each method has advantages and limitations. X-ray imaging, for example, offers high-resolution, rapid food quality inspection.

However, it has difficulty detecting soft materials such as plastic, wood fragments and insects in food products. In addition, X-ray radiation can ionise food products, raising concerns about genetic mutations and posing a significant risk to operators. Infrared imaging offers high scanning speeds and ensures the safety of both operators and food products.

However, it has limited penetration capabilities. Thermal imaging, on the other hand, is only suitable for cooked food during processing and is not applicable to raw food. Ultrasonic imaging boasts excellent penetrability but requires a medium, making it unsuitable for use in open spaces. Fluorescence imaging is mainly practical for examining fluorescent compounds.

Non-destructive spectral systems

Real-time inspection is critical for identifying defects, optimising processes and minimising costs. Non-destructive spectral systems enable in-line analysis, surpassing off-line and at-line methods. Off-line measurements require an operator or automated system to extract samples from the process line or location for analysis in a separate laboratory. At-line analysis, on the other hand, is performed close to the production line, but samples still need to be physically removed before testing. In-line systems, on the other hand, instantly monitor entire production lines.​​

The traditional inspection methods, which rely on the manual expertise of the operator, are slow, subjective and prone to errors due to fatigue or inexperience, resulting in inconsistent quality and high defect rates.​​ r early automated systems, such as colour imaging based on red, green and blue (RGB) filters, improved speed but were limited to surface properties such as colour or shape, unsuitable for assessing chemical composition or internal defects critical to modern standards. Non-destructive spectral systems, integrated into process analytical technologies, utilise spectroscopic techniques in the ultraviolet (UV), visible (Vis), near-infrared (NIR) and mid-infrared (MIR) ranges to provide rapid, non-invasive analysis with high accuracy.

More innovative methodologies, such as hyperspectral imaging (HSI) and artificial intelligence-based data processing (e.g., convolutional neural networks), surpass traditional methods in terms of speed, repeatability, and analytical depth. Among the advantages of automated methods is their ability to perform repetitive tasks without errors, regardless of workload, and to minimise the possibility of errors caused by operator fatigue, distraction or subjectivity. Compared to RGB systems, which are limited to the visible spectrum (approximately 400-700 nm), non-destructive spectral sensors probe chemical and structural properties through NIR and MIR, enabling applications such as the detection of environmental contaminants in water or the assessment of food quality.

Hyperspectral imaging simultaneously acquires spatial and spectral data, such as mapping the distribution of sugars in fruit, a task impossible for older single-point spectroscopy, which lacks spatial resolution. On-The-Fly Processing (OTFP) allows data (images, data streams, 3D scans) to be processed in real time or near real time without having to save it completely first, enabling immediate modifications, analysis and efficient visualisations. This real-time processing of data streams without storing large data sets overcomes the memory constraints of traditional batch processing methods. These capabilities make non-destructive spectral systems valuable for real-time quality control and production process monitoring.  However, challenges remain, such as high costs (US$50,000 to US$200,000 for HSI systems), calibration complexity and computational demands. Despite the advantages and significant advances, the widespread adoption of spectral technologies in real-time industrial applications is uneven and still limited.

How hyperspectral imaging works for food quality control

Hyperspectral imaging, also known as chemical or spectroscopic imaging, combines conventional imaging, radiometry and spectroscopy technologies to acquire both spatial (images) and spectral information from agricultural material samples. Many studies have focused on the application of hyperspectral imaging for the quality analysis of liquid and semi-liquid food products, meat, poultry and fish, fruit and vegetables, ham, cereals and cheese, and for the detection of microorganisms in food stored at refrigerated temperatures. Hyperspectral imaging is a non-destructive tool designed with the integration of digital imaging technology, radiometry and optical spectrometry principles.

Radiometry is the estimation of the amount of electromagnetic energy emitted per unit of time (usually expressed in Watts) at a specific wavelength range. A radiometer is configured with a single sensor that includes a filter installed to select the projected wavelength range. Spectrometry refers to the amount of light intensity (usually expressed in ^W/m²) at a specific wavelength range. Unlike radiometers, spectrometers use diffraction gratings or prisms or multiple sensors to divide the wavelength range into different wavelength bands.

The ‘hyper’ in hyperspectral means ‘many’ or ‘multiple’ and includes the assembly of data in hundreds of bands, with each band or channel covering a narrow, contiguous portion of the electromagnetic spectrum. Hyperspectral imaging can generate both a spatial map and a spectral variation. The data collected from a hyperspectral imaging system forms a three-dimensional structure consisting of one spectral dimension and two spatial dimensions called “hypercubes” or “data cubes”. The combination of imaging technology with spectroscopy provides information about the food being tested (size, geometry, appearance, colour) through image feature extraction, as well as the chemical properties or constituents of the food through spectral analysis.

In addition to being a non-invasive technology, hyperspectral imaging is safe, as no chemicals are used, allows for a better understanding of the chemical elements of food, saves time compared to traditional or chemical methods of food quality control and evaluation, allows for the correct selection of the area of critical interest for image analysis, and simultaneously obtains spectral and spatial information to provide more accurate data on the chemical constituents of the sample.

As mentioned above, despite its advantages, hyperspectral imaging also has some limitations, first and foremost being its cost. In addition, due to the large size of hyperspectral imaging data, large-capacity units are required for data storage and high-speed computers for data processing.  During image acquisition, signals may be affected by the surrounding environment, such as lighting, scattering, etc., resulting in a poor signal-to-noise ratio. Future improvements in the precision, accuracy and speed of hyperspectral imaging could come from better lighting systems, advanced photometric sensors and faster hardware.

Integration of artificial intelligence into food quality control technologies

The application of artificial intelligence (AI) and machine learning (ML) to food quality control helps improve accuracy in data analysis and monitoring throughout production. These technologies use machine learning algorithms to analyse large data sets and detect patterns, increasing the accuracy of food quality detection to over 90%. Deep learning-based computer vision enables detailed defect detection and product classification based on attributes such as size, color and shape.

Hyperspectral imaging has also proven effective, detecting chemical compositions and deterioration at the molecular level, while IoT-enabled sensors provide real-time data on crucial parameters such as temperature and humidity for immediate intervention. Electronic nose (e-nose) technology has proven highly effective in quality inspection, achieving classification accuracies of up to 96% in distinguishing, for example, between fresh and overripe fruit. The combination of biosensors with AI/ML technology further amplifies their capabilities, improving food safety monitoring by being efficient and reliable in detecting pathogens.

These biosensors employ highly sensitive optical detection techniques, utilising a large surface area, often enhanced with nanoparticles to increase signal detection. The data generated by biosensors can be further processed using artificial intelligence, enabling rapid identification of patterns and anomalies for faster and more accurate results. The real-time analysis provided by optical biosensors is critical for food safety and quality assurance. Improving sensor stability, reducing costs, and optimising the interpretability of AI models will be key. Miniaturisation and standardisation of devices will play a key role in making these technologies more accessible and operationally efficient.