Design and implementation of real time hyperspectral target detection using fpga

Abstract

Hyperspectral imaging is an emerging research area in the field of newlineremote sensing. Hyperspectral sensors capture hundreds of continuous newlinespectral bands as a three-dimensional cube from the earth s surface. The newlinetransmission of this large amount of data to the station on earth may exceed newlinethe maximum data rate of the communication system. The high dimensional newlinedata may also burden the ground processing system. To address this issue, newlinemany researchers suggest the onboard processing of hyperspectral images to newlineachieve real-time performance. Field Programmable Gate Arrays (FPGAs) are newlinethe preferred hardware platform for onboard processing of hyperspectral newlineimages. FPGAs are light, small in size and have parallel processing systems newlinewith low power consumption. They have a reconfigurable capability and can newlinealso handle ionizing radiation in space. newlineThe main contribution of this research is to design and implement newlinean efficient, unsupervised technique for hyperspectral image analysis. In this newlineresearch, three unsupervised strategies are proposed for hyperspectral target newlinedetection and classification applications. Due to the limited availability of newlinelabelled data, unsupervised methods are suggested for remotely sensed newlinehyperspectral imagery. newlineIn the first strategy, an automatic hyperspectral target detection newlinetechnique was adopted and the performance of the hardware implementation newlineanalysed. The conventional Automatic Target Generation Procedure (ATGP) newlineuses Orthogonal Subspace Projection (OSP) which has complex matrix newlineinverse calculation. This algorithm is difficult to implement in hardware and newlinealso does not meet the real-time requirements. To overcome this limitation, a newlinemodified version of ATGP is adopted with Gram-Schmidt Orthogonalization newlinewhich uses only inner products. newline