Up until now traditional hyperspectral cameras have been large, expensive and slow. Therefore, they can often not be used for time-critical or high-throughput applications, and their use is thus limited to highly specialized research environments. Imec’s chip-based technology paves the way towards the industrial application of hyperspectral imaging, as it enables small, cost-efficient and fast solutions.
Imec’s fully integrated CMOS compatible hyperspectral sensor consists of a set of spectral filters that are directly post-processed at wafer level on top of a commercially available CMOSIS CMV4000 image sensor (a 4 megapixel sensor with a maximum framerate of 180fps). The hyperspectral filter has100 spectral bands between 560nm and 1000nm. The filter bandwidth (Full Width Half Max) ranges from 3nm at 560nm to 20nm at 1000nm, and the transmission efficiency of the filters is around 85%. Typical integration times used in the current prototype setup are between 2 and 10 miliseconds under halogen light illumination of 450 Watt. This indicates maximal frame rates can be achieved of up to 500fp.
Due to its integrated filter design and high spectral filter efficiency, imec’s solution can obtain scanning speeds that are compatible with industrial requirements. The speed of the demonstrated system corresponds to an equivalent speed of 2,000 lines per second and significantly exceeds current state-of-the-art hyperspectral sensors. Moreover, the technology can be adopted to a range of industrial vision applications. To match to specific application requirements, the image sensor can be selected (a commercially available sensor or even a custom-designed sensor), determining pixels sizes, maximal frame rate, etc. The hyperspectral filter can be tuned by changing the number of spectral bands and their spectral resolution.
Imec’s prototype hyperspectral camera can capture all relevant data enabling automatic classification of different objects using state-of-the-art image processing methods. Classification results of imec’s solution are equivalent to state-of-the-art hyperspectral references and recorded spectra of e.g. plant material.