A highly cost-effective, hyperspectral imaging technology that allows innovative artificial intelligence applications to be introduced into consumer devices, has been developed at the VTT Technical Research Centre of Finland.
Spectral filtering technology taps the benefits of the very-near-infrared (VNIR) wavelengths, which can be detected by even inexpensive mobile phone cameras. Artificial intelligence can be adopted to understand these environmental spectral data within images, which cannot be observed by the naked eye.
Small VNIR hyperspectral camera can show raw avocados from ripe. (Image credit: VTT Technical Research Centre)
The designed hyperspectral camera uses VNIR wavelengths that surpass the red color viewed by the human eye and are commonly filtered out of standard camera images. The spectral data included in the VNIR range allows detection and labeling of the materials and characteristics of various objects within the environment.
Spectral data of image objects produces information associated with, for instance, food safety or freshness, differentiating between fake and real products, security camera recordings or medicines. This information can be also adopted by wireless mobile applications developed for interpreting sensor data. The facility can be merged into the everyday environment to render them highly intelligent, and can be integrated with smart home systems and appliances, robots, mobile devices and autonomous vehicles, which have to discern visual information to function securely.
In future, an increasing share of vehicles and systems will become autonomous, and the need for reliable visual camera information for automated decision-making will increase. Adding the third spectral dimension to images could provide more safety and security for autonomous systems relying on machine vision and artificial intelligence to make decisions based on visual camera data.
Anna Rissanen, Research Team Leader at VTT
At present, the cost of majority of the hyperspectral imagers accessible from the market, ranges from thousands to tens of thousands of dollars, indicating that they cannot be incorporated, for instance, into a smart fridge to evaluate food freshness. Other spectral imaging techniques that intend to achieve mass-producible volume scaling for reducing the final cost of the product, generally process filters of fixed wavelength directly into individual camera sensor pixels. Yet, this strategy has the disadvantage that it necessitates high-cost telecentric optics.
VTT’s technology has a simple optical path, making it compatible even with the very compact and low-cost optics used in mobile cameras, which is not possible with other spectral imaging technologies. This is a huge advantage because it enables very cost-efficient mass production for these hyperspectral camera sensors.
The bill-of-material cost for the new VNIR range, 600-900 nm, hyperspectral camera sensor hardware has set a new low record of 150 dollars. By adopting the mass-producible MEMS technology, the tunable filter technology developed by VTT can also be combined with any camera sensor without raising its cost or size much. If high-volume production and calibration techniques are adopted, the sensor cost, including the camera optics, can be less than 20 dollars and the cost of the core component (i.e. the micro-opto-electro-mechanical, MOEMS, chip) can be less than one dollar.
In the recent past, the VTT researchers have pioneered the development of hyperspectral imagers for the most challenging, innovative application fields, by launching the world’s first visible and SWIR hyperspectral imagers on space CubeSat missions, as well as diagnostic instruments for rapid skin cancer screening. Spectral sensing techniques devised by the researchers are already creating business for various companies, which have also developed award-winning photonic sensor products.
Over the coming years, the goal of VTT is to commercialize cost-effective hyperspectral imaging techniques in collaboration with companies functioning in this field.