Parallel discriminative subspace for city target detection from high dimension images

Abstract

City Target Detection is an enduring problem that intrigues the researchers all over the world. The great success of existing Target Detection algorithm appears in ubiquitous scenarios: Pedestrian Detection, Vehicle Tracking, etc. However, as for the city target detection in the remote sensing, we are facing with two inevitable problems: Complex Environment and Massive Information. The complicated environment encumbers the accurate extraction of the target profile, and the huge amount of information turns it into a heavy workload to get the final outcome for the conventional CPU- compiler architecture. In this paper, we propose a binary hypothesis framework based on adaptive dictionary and discriminative subspace for hyperspectral city target detection (BHADDS). Furthermore, we have also implemented it on other hardware platform alongside with CPU, such as FPGA. FPGA is a low-power portable and programmble SoC, and also the protocol model for potential massive production of the SoC chipset. Our eventual aim is heading for the high-performance processor with strong instant processing ability for remote sensing. In the final part of the paper, we have given a comprehensive performance comparison over the different platforms and summarized their applicable scenarios.