UST035

92 less spatial and spectral resolution. On the other hand, smaller pixels can mean higher spatial and spectral resolution but potentially at the expense of frame rate, because more data needs to be processed. That’s why different sensors are optimised for particular uses. The pixels vary from a CMOS FPA with 1600 pixels, 369 spectral bands and a frame rate of 250 fps to one with 640 pixels, 270 spectral bands and a frame rate of 345 fps for UAVs, or an FPA with 1020 spatial pixels and 340 spectral bands. The FPA array is selected according to the wavelength range needed and the dimensions and pixel layout. For example, in the NIR (900-1700 nm) range, sensors use InGaAs FPAs, while in the SWIR (900- 2500 nm) range they use MCT. Another approach to simplifying the design of a hyperspectral camera is to use spectral filtering. This takes advantage of the VNIR wavelengths to detect and label the properties of different objects in the environment. Machine learning can be used to interpret this environmental spectral data in images. Limiting the range of the sensor to NVIR allows for a simpler 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 can bring the cost of a VNIR hyperspectral sensor hardware operating in the 600-900 nm range down to $150 using micro-machined MEMS tuneable filter technology. The sensor cost, including the camera optics, could be less than $20, and the core component, the micro-opto-electromechanical chip, could cost less than $1. Space applications Image sensors are of course essential for unmanned systems in space, but the technology takes up to 10 years to be designed and deployed in systems. For example, each of the cameras on the Mars 2020 rover uses a 20 MP CMOS sensor with an RGB colour filter array. The rover’s navigation cameras, which are mounted on a pan/tilt remote sensing mast, acquire colour stereo images and panoramas from a height of about 2 m above the Martian surface. Separate cameras used to detect hazards are hard-mounted to the rover body at a height of about 0.7 m above the surface and acquire colour stereo images of the areas immediately in the front and rear of the rover. Acknowledgements The author would like to thank Joseph Notaro at ON Semiconductor; Naoya Sato, Satoko Iida and Yuichi Motohashi at Sony Semiconductor, Devang Patel and Mat Arcoleo at Omnivision Technologies, Luca Verra at Prophesee, and Russ Nakatsuji at Headwall Photonics for their help with researching this article. December/January 2021 | Unmanned Systems Technology Focus | Image sensors AUSTRIA AMS +43 3136 500 0 www.ams.com BELGIUM Imec +32 16 28 12 11 www.imec-int.com CHINA SmartSens +86 21 6485 3570 www.smartsenstech.com FRANCE Prophesee +33 1 85 73 27 33 www.prophesee.ai GERMANY Framos +49 89 710 6670 www.framos.com IGI +49 27 325 5250 www.igi-systems.com JAPAN Sony Semiconductor +86 755 8258 1661 www.sony-semicon.co.jp SPAIN Qurv Technologies – www.qurv.tech UK Teledyne e2v +44 1245 493493 www.teledyne-e2v.com USA Analog Devices +1 781 329 4700 www.analog.com Headwall Photonics +1 978 353 4100 www.headwallphotonics.com Omnivision Technologies +1 408 567 3000 www.ovt.com ON Semiconductor +1 602 244 6600 www.onsemi.com Examples of image sensor suppliers