An Onboard Hyperspectral Image Processing System Based on Deep Belief Network Using FPGA

Abstract

Abstract Real-time processing of hyperspectral images has been widely adopted in the field of remote sensing applications. Deep learning methods have been proved that it has high accuracy compared to traditional algorithms like Support Vector Machines (SVMs). It is very challenging to achieve real-time performance in hyperspectral imagery with deep learning algorithms due to its computational complexity and high dimensionality of hyperspectral images. Deep Belief Network (DBN) is an emerging deep learning algorithm that involves unsupervised pretraining and supervised fine-tuning. Gaussian– Bernoulli Restricted Boltzmann Machines (GBRBMs) are used to construct the layers of DBN. This work presents a novel methodology for the implementation of the DBN algorithm on the Field-Programmable Gate Array (FPGA) platform. In experimental analysis, a real hyperspectral image is considered for evaluation and the proposed algorithm is implemented on the Virtex-6 FPGA board. The experimental results show that the proposed implementation shows promising processing speed, high accuracy and low power consumption.