Mark Williamson, MV Pro European Editor at Large takes a deeper look at how, in the world of machine vision, lighting isn’t just an afterthought it’s a critical enabler for defect detection, material classification, and process control.
From manipulating visible light with specialized angles and filters to exploiting the hidden insights of the infrared spectrum, modern imaging systems are evolving rapidly. Technologies like SWIR and hyperspectral imaging now reveal what the human eye cannot.
In many vision applications, setting up the right lighting angle, light colour, and lens filter is crucial for detecting defects. It’s not just about snapping a picture. Experts understand the differences between dark field, bright field, collimated, on-axis, diffuse, and backlighting with various illumination wavelengths, each highlighting different types of defects. When done correctly, defects that are nearly invisible under normal light become very obvious. A well designed vision system will control the illumination and light using filters to increase the applications reliability.
Visible light typically covers wavelengths from 380nm (violet) to 700nm (red). However, most standard image sensors are sensitive up to 1000nm, which includes the near-infrared (NIR) range. In colour cameras, manufacturers fit an NIR block filter to mimic the human eye, but by filtering out the visible spectrum, this extra 300nm can reveal hidden details or simplify tasks by removing colour information.
Spectroscopy, which historically measures wavelengths at a point, has shown that objects reflect light waves outside the visible spectrum. Over time, different camera technologies have been developed to detect ultraviolet (UV) light (<380nm), short-wave infrared (SWIR) (1000 to 2500nm), medium-wave infrared (MWIR) (3-5µm), and long-wave infrared (LWIR) (8-12µm).
One of the first areas of inspection beyond the visible was the introduction of thermal imaging cameras which used the LWIR. First used to enable search and rescue, surveillance and law enforcement, thermal imaging enabled warm objects to be identified through smoke, fog or under vegetation. As prices have fallen the adoption has expanded in to areas such as process control, infrastructure maintenance and building surveying. Companies such as Teledyne FLIR have grown to deliver a variation of camera for almost any application. Calibrated cameras can now accurately measure temperatures remotely and identify poorly insulated properties, poor conductivity in power distribution systems. Smart thermal camera such as the IRSX from AT sensors now connect directly to factory control systems monitoring processes and even providing process control feedback.
In the past decade, there has been significant interest in what the SWIR wavelength band can reveal. By analysing the spectrum through hyperspectral imaging, we can unlock new insights. Just as materials in the visible spectrum absorb and reflect different wavelengths, giving objects their colour, the same happens in the non-visible spectrum. Spectroscopy research shows that each material has a spectral signature with different wavelengths being absorbed or reflected. A common demonstration at trade shows involves using a SWIR camera to show clear water appearing black when illuminated with SWIR light, highlighting that water absorbs light in the 1-10µm band.
Hyperspectral imaging companies are increasingly entering the machine vision market. The concept is similar to the image on Pink Floyd’s famous album cover, “The Dark Side of the Moon,” where white light is diffracted into the full colour spectrum. By illuminating an object with broadband SWIR light and refracting the light onto a 2D SWIR image sensor, each pixel captures a signature of wavelengths. Using a 512 x 512 pixel sensor provides a 512-wavelength profile for each pixel along a 512-line. Scanning an object passing the sensor (like a fax machine or line scan camera) builds up a data cube. However, this data is 512 times the size recorded by an equivalent monochrome line scan sensor, requiring significant processing power and fast memory.
When coupled with hyperspectral analysis software, the capabilities are game-changing. It enables the detection of the chemical composition of materials. For example, white pills can be classified as paracetamol, aspirin, or ibuprofen, which is not possible with the human eye.
Real-world applications for SWIR cameras in factories include detecting subsurface damage, such as bruising and ripeness of fruit. In food safety, they can detect contamination and disease in products like fish and meat, and identify stones in pulses or grains that are indistinguishable using visible light. Outside the factory, drones and high-flying survey aircraft can analyse crops, highlight weeds and water content, providing farmers with key information to water and apply chemicals only when needed. Law enforcement can detect illegal crop cultivation from afar.
Despite these advantages, the adoption of this technology has remained niche due to cost. While hyperspectral imaging works at any wavelength, most applications require SWIR or MWIR, which are more expensive. Currently, a SWIR camera costs over $12,000, a calibrated hyperspectral camera over $30,000, and with additional costs for SWIR analysis software, computing power, and illumination, the total system cost exceeds $50,000 plus integration costs.
Suppliers in this area include Xenics and Allied Vision for raw cameras, SPICIM and Innospec for calibrated cameras, and Headwall for complete systems. These cameras use InGas sensors, which are expensive due to limited manufacturing volumes and process complexity. To mainstream this technology, innovation is needed to reduce costs.
One approach to reduce costs is to identify a few wavelengths with clearly different absorption reflection points to differentiate materials. Reducing the data cube to a few bands eliminates the need for diffraction, saving costs. If the signatures are in the VISIR waveband, lower-cost multispectral cameras can be used, such as custom prism multi-chip cameras by JAI or 2D sensors with multispectral pixel filter patterns provided by IMEC and integrated into cameras by companies like Lucid and Ximea. Each application needs validation testing, which can be costly for one-off projects, making the SWIR hyperspectral approach less manpower demanding if the budget allows.
Fortunately, significant work is being done to reduce costs. New technologies like quantum dot engineering promise to lower costs by an order of magnitude, increasing adoption. Startups like Quantum Solutions and QDI are working on products in this area.
The market for this technology is growing, with key players emerging from small innovation companies. For example, Headwall Group has made several acquisitions, including Innospec in 2024 and EVK in early 2025, and SPECIM was acquired by Konica Minolta in late 2020.
Having worked with hyperspectral over the last 7 years or so, its clear there are many use cases where the technology will be able to solve previously impossible applications. The level of interest is clearly there. So many of the applications just could not justify the cost needed for hardware coupled with the limited availability of spectroscopy knowledge needed to design the system correctly. With AI helping simplify hyperspectral signal detection and with many companies working towards significant cost reductions in the SWIR sensor cost its clear the adoption of hyperspectral imaging will further accelerate over the next 5 to 10 years. The first company able to scale a lower cost SWIR sensor is sure to do very well.
