The advent of affordable and easily deployable hyperspectral imaging systems have extended us the ability to collect images and study large swathes of vegetation. Increased use of hyperspectral imaging brings with it a plethora of challenges in both the pre-processing and ecological knowledge extraction realms. This talk will discuss our approach toward calibrating large sets of hyperspectral images, which is an important pre-processing step. We propose a cloud computing based architecture that uses a software stack that largely automates the image calibration process and greatly reduces the human effort required in comparison to currently available tools. In the heart of the software stack lies our HSCal application, which is a scriptable and OMP parallel hyperspectral image processing tool. HSCal uses factory provided radiometric calibration data and can exclude shaded pixels from calibration reference samples for reflectance calibration. After calibration, spectral transformation such as spectral indices can be computed, or image areas sampled. The main design goals of HSCal are supporting longitudinal hyperspectral analysis and the creation of hyperspectral panoramas. We will discuss the radiometric and reflectance calibration algorithms, panorama stitching process, and our software implementation. The talk will conclude with a description of our cloud architecture for processing multiple streams of continually captured sensor data and time-series data-sets, with emphasis on stream processing hyperspectral images.

Bio:

Erik Keever is a third year PhD student in the Department of Physics at the University of Oregon studying with advisor James Imamura. He is generally interested in the application of computers to simulation of physical systems, with specific research interest in GPU acceleration of simulation programs and the simulation of plasma dynamics.