Gpu parallel implementation for real-time feature extraction of hyperspectral images

Abstract

As the application of real-time requirements gradually increases or real-time processing and responding become the bottleneck of the task, parallel computing in hyperspectral image applications has also become a significant research focus. In this article, a flexible and efficient method is utilized in the noise adaptive principal component (NAPC) algorithm for feature extraction of hyperspectral images. From noise estimation to feature extraction, we deploy a complete CPU-GPU collaborative computing solution. Through the computer experiments on three traditional hyperspectral datasets, our proposed improved NAPC (INAPC) has stable superiority and provides a significant speedup compared with the OpenCV and PyTorch implementation. What’s more, we creatively establish a complete set of uncrewed aerial vehicle (UAV) photoelectric platform, including UAV, hyperspectral camera, NVIDIA Jetson Xavier, etc. Flight experimental results show, considering hyperspectral image data acquisition and transmission time, the proposed algorithm meets the feature extraction of real-time processing.