Performance evaluation of state-of-The-Art CNN architectures for the on-board processing of remotely sensed images

Abstract

Over the last few years, Convolutional Neural Networks (CNNs) have been extensively used in different remote sensing applications. However, for large networks the computation and memory requirements have brought many challenges into this field. Additionally, the computational capabilities of hardware devices available on-board satellites is limited, being this another constraint for these implementations. In this paper, the authors present the evaluation of nine different CNN architectures for ship and airplane detection, taking into consideration that the final use-case application will be an on-board system with target detection capabilities.