Multimedia & Artificial Intelligence

With hyper/multi-spectral sensor technology evolving and becoming more cost-effective, it is likely we will see these spectral cameras replace standard RGB cameras in a multitude of applications beyond the traditional niches of medical, remote sensing, and precision agriculture [5] in the near future. The rich spectral information contained in hyperspectral images can be used to characterize the objects/materials in the scene with great precision and detail. The availability of more bands than the usual three RGB bands has been shown advantageous in disambiguating objects [1,2,3,4,5,6]. Recently, spectral cameras have been deployed to a wide range of applications, such as in mammography, single-shot spectral imaging is used for breast tumour and developing cancer detection [8], similarly cameras in cars exploit V-NIR range for pedestrian detection in night vision [1], and more. In this talk, we will discuss the hyperspectral imaging pipeline from computer vision [4,7] standpoint in terms of exploiting the multi/hyper-spectral sensor information for a more accurate classification, recognition, and detection tasks. So that the

imaging system is compact, computationally efficient, cost-effective, and a perfect fit for real-time applications.