Embedded high performance computing for on-board hyperspectral image classification

Abstract

Jetson TK1 is a recently launched embedded application development platform from NVIDIA, which features the Tegra K1 processor and Kepler Graphics Processing Unit (GPU). We envisage that such a system has huge potential for deploying an embedded system for on-board classification of hyperspectral images. We used a convolutional deep neural network for designing a unified model for hyperspectral image classification. Deep convolutional model hierarchically extracts spectral-spatial features from hyperspectral imagery and these features are used by the fully connected layer of neural network to perform pixel level classification of hyperspectral imagery. Our experimental results show that Jetson TK1 based hyperspectral image classification gives promising results and the possibility of having Jetson based embedded platform for on-board classification of hyperspectral images.