What should you know about hyperspectral imaging?

Hyperspectral imaging (HSI) is a powerful tool for machine vision quality inspection. Thanks to its high spectral resolution and rapid data acquisition and processing capabilities, it is ideally suited for the most demanding inspection applications.

HSI components

The HSI technology consists of both data acquisition and data processing. It relies on appropriate illumination, software, AI and machine learning algorithms, system integration, and data acquisition devices, including:

• hyperspectral cameras (operating in the near-infrared (NIR) and short-wave infrared (SWIR) spectral ranges),
• point spectrometers,
• line-scan hyperspectral scanners,
• hyperspectral microscopes,
• integrated hyperspectral sensors.

Strengths and limitations

Its key advantages include application flexibility, high sensitivity to physical and chemical variations, full spectral information for every pixel, high-speed data acquisition (exceeding 1,000 frames per second in some systems), and seamless integration with AI and machine learning algorithms. However, hyperspectral imaging also has its limitations:

• high implementation costs compared to other inspection technologies;
• requires significant computing power, spectral libraries, and advanced machine learning (ML) and artificial intelligence (AI) models;
• sensitive to environmental variations (such as lighting and temperature), requiring stable acquisition conditions and, where necessary, periodic retraining of AI and ML models;
• requires substantial data storage capacity due to the large volume of generated data;

requires substantial storage capacity to accommodate the large volumes of data generated;

Where is hyperspectral imaging used?

Simply put: wherever the investment is justified by the application. Industries that make the most extensive use of hyperspectral imaging include:

• defense and security – detection of camouflage and hazardous chemical substances;
• industrial manufacturing – machine vision quality inspection, material analysis, waste sorting, and quality inspection in the food and pharmaceutical industries;
• mealthcare – diagnosis of skin lesions and tissue analysis;
• agriculture – crop health assessment, nutrient analysis, and weed detection.

Applications of hyperspectral imaging in industrial manufacturing

Available solutions

The most widely used solutions are push-broom (line-scan) hyperspectral cameras, which are ideally suited for conveyor-based inspection, providing high spectral resolution at high production speeds. For applications involving randomly moving objects, more advanced snapshot systems are used, capturing the full spectral dataset in a single frame.

Hyperspectral cameras typically operate in the VNIR (400–1000 nm) and SWIR (1000–2500 nm) spectral ranges, enabling material differentiation as well as the assessment of properties such as moisture and fat content. Illumination is most commonly provided by halogen lamps, which offer continuous spectral output in the NIR/SWIR range, although LED-based lighting with application-specific spectral characteristics is becoming increasingly popular. Since measurement accuracy depends heavily on lighting conditions, stable illumination and effective reflection suppression are essential, often achieved using light integrators or inspection tunnels.

The vast amounts of data generated by hyperspectral cameras are processed using AI and machine learning algorithms, ranging from traditional methods such as PLSR and SVM to advanced deep neural networks. These algorithms enable not only material classification (e.g. PET vs. PVC in recycling applications), but also quantitative analysis through regression models (e.g. determining the water content of fruit). Increasingly, data processing is performed using edge computing, reducing the need to transfer large volumes of data while enabling real-time system response.

Example: Quality inspection of meat and meat products

In the first example application, Specim SWIR line-scan hyperspectral cameras and halogen illumination mounted above a conveyor belt were used for data acquisition. The objective was to detect eight types of foreign materials across five different types of meat. Spectral analysis revealed that adding an RGB camera improved the detection performance for three types of foreign materials.

In the second example application, the inspected products were finished meat products—kabanos sausages, frankfurters, and smoked sausages. The inspection focused on their visual quality attributes, including discoloration and the degree of smoking. Tests conducted by Scanway showed that, in this case, RGB machine vision provided superior inspection performance. The reason was that the evaluated quality attributes showed no significant deviations in the infrared spectrum, whereas they were clearly distinguishable in the visible spectrum.

Hyperspectral imaging can reveal far more than conventional cameras, yet it is not always the right choice for every application. Although it involves higher costs and generates large volumes of data, its flexibility and analytical capabilities make it one of the key technologies shaping the future of automated quality inspection.