Learn more about how Opsyne’s technology can be implemented to improve safety, quality, and workflows across various applications.
Explore our FAQ’s about hyperspectral imaging, Opsyne products, and how we can help improve your processes.
Hyperspectral imaging is a powerful sensing and detection technology for performing non-invasive material analysis. Where a mass spectrometer directly counts atoms, hyperspectral imaging, also known as imaging spectroscopy, can predict the content of materials based on the interaction of light with the molecules. The instrument detects and counts photons at specific wavelengths across a specified and broad spectrum of light to see things invisible to the human eye. With hyperspectral we can find hidden value such as gold concentration in core samples, antioxidant levels and flavour profiles of fresh produce, and cancer presence in skin or brain tissue
The principles of hyperspectral imaging technology are not new. Historically the effectiveness of the technology has been limited by the hardware cost and the image data processing power. There is also the issue of correlating the spectral response with direct “wet” chemical analysis. In line with other developments in the world of sensing, detection and the internet-of-things, this has all changed. The primary components of the camera, detection chips and camera filters have declined significantly, opening the doors to many new applications. The power of graphics processing units (GPUs) cloud computing and machine learning, including the use of centralized image data classification libraries and neural networks, have made it possible to classify an image in an instant, supporting real-time surgical guidance, on-site core sampling, and on-line classification and sorting of food components and products in real-time. The cameras cannot penetrate objects beyond a relatively thin layer, so depending on the application, particularly in food manufacturing, are best used in concert with x-ray systems, which we can also provide. The initial scientific research to develop the application remains the most significant barrier, though we can work closely with your operations team to complete the up-front development work as appropriate.
This may also be referred to as “spectral sampling.” Let’s start with the concept of megapixels. If you take an image of a landscape with two cameras of different megapixels, one being low, the other high, as you zoom in with the first camera, you will struggle to see specific features such as a house or a hiker. With the higher resolution, you will not only be able to identify these features, but you will see the colour and texture, perhaps even the material, of the roof of a house and the brand of jacket worn by our hiker. A highest possible spectral resolution is critical for advanced high-value material analysis. If you want to find features at the molecular level such as pigments, salts, beta-carotenoids, flavonoids, and low concentrations of gold, quartz, or early indications of cancer and disease, it is impossible without the appropriate spectral resolution. We offer the highest spectral resolution in the very-near (350-100nm) and short-wave (1000-2200nm) because our imagers are designed using the most advanced chips and sophisticated optical engineering to find the slightest, most nuanced features in materials and objects of interest. If you are familiar with the term sensitivity, this is what spectral resolution provides. The higher the spectral resolution, the lower the probability of false positive, and the higher the probability of true positives.
In the world of sensing and detection, not only do we need to find (in absolute terms) what we are looking for, say gold and quartz in remote sensing or a specific flavonoid in coffee beans, but we need to distinguish from similar elements. The similar elements, though distinct may emit a similar spectral response, or spectral signature. This is one source of noise. Depending on the application, there may also be environmental interference. For an aerial survey to find rare tree species or tree disease, or specific mineralogy it may be a cloudy or otherwise low-light environment. A higher signal-to-noise ratio (SNR) may be crucial to pull out, or distinguish the feature or profile of interest. In short, the SNR helps to distinguish between similar but different molecules, colours, textures, attributes and so on. If you are familiar with the term specificity, this is what the SNR provides. The higher the signal-to-noise ratio, the lower the probability of false negative, and the higher the probability of true negatives. Find what you want, avoid finding what you do not want.
A regular RGB (red-green-blue) camera captures only these three colours. The camera has a filter that split the incoming light, by each of these three visible colours. Recall that colour is light. The colour in each pixel is recreated through a process of interpolation. If you ever looked into the back of a projection television (ca. 1997), you will have seen this process in reverse. Hyperspectral imaging captures colours beyond the visible spectrum, across many individual colour bands or wavelengths. The hyperspectral camera detects the photons, which are the units of light emitted from the object, within each and many (over 1000 for our VNIR camera) individual colour bands. As each molecule or object or feature, depending on the application, reflects light differently, we can interpolate or more accurately reconstruct the spectral signature for each individual pixel, just as for the RGB camera, to not only know the colour and recognise the object, but to also see what our eyes cannot see: the texture, the freshness, the flavour, the minerals. If we want to know the freshness or pesticide content of lettuce, hyperspectral imaging can do this. High and low concentrations of pesticides will absorb and reflect light differently. If you can count photons carefully, you will see this and be forgiven for thinking its magic!
The set of applications is broad, varied and can be highly specific. In general, hyperspectral imaging shines where the surface or near-subsurface needs to be analysed rapidly or in real-time, non-invasively and non-destructively. A great example is in brain or gastrointestinal surgery. The surgeon may have a map from previous medical imaging (CT, X-Ray, MRI) but otherwise is reliant upon active visual inspection. Often the surgical bed can become flooded with blood. How can the surgeon see beneath or through the layer of blood into the surgical bed and the tissue of interest. With the appropriate configuration, hyperspectral imaging can see through the blood. In brain surgery, this form of surgical guidance is crucial to improve patient outcomes by distinguishing and delineating cancerous or bad tissue from good tissue. It can be fatal or at least damaging if good tissue is removed, or if bad cancerous tissue is left in place. Real-time hyperspectral imaging can distinguish the good from bad in very clear terms with the visual power of a heat map. These principles are obviously extended throughout biomedical screening and diagnosis, but also throughout the agrifood supply chain and natural resources development project lifecycles, where the key value drivers are not available at the breadth, depth and precision required using the human eye or discrete chemical analysis.
There are three sources of business value: risk mitigation, workflow efficiency and automation, and value mining. In the agrifood industry vertical, hyperspectral imaging can detect disease onset, crop stress, and produce variability. It can also be used to select the best seeds automatically at scale. Post-harvest imaging can select for specific origin or flavour profiles to detect fraud or otherwise ensure uniformity according to specific attributes. In food manufacturing, hyperspectral imaging can detect plastic, metals and other foreign objects in low concentrations. But it can also measure protein, fat, and moisture content as a quality control mechanism to ensure consistent shelf-life, component distribution and flavour profiles. Another interesting application is sorting coffee and cocoa beans according to their flavour profiles to unlock value in the highest-margin chocolate markets. By moving beyond standard aggregation and averaging of raw materials, mapping the distribution and classifying according to key value drivers unlocks value. Contamination is a standard part of business in the food industry, and hyperspectral can help you directly measure rather than merely estimate which units have been affected, to reduce waste.
Yes. Depending on the application it may be best to work in concert with other sensors. X-Ray can be a strong complement in food, and we can work to combine the power of multi-sensor datasets using machine learning algorithms. We can also offer multi-spectral solutions which can be more powerful for finding specific parameters such as a pesky leaf disease in grapes, particularly in remote sensing. Thermal cameras, Raman spectroscopy and NIR spectroscopy are other complementary sensors that we can consider and consult.