CEAS EuroGNC 2022
|
Attention-based DeepMoon for Crater Detection
|
| Jianing Song |
Postdoctoral Research Fellow, City University of London, Department of Electrical and Electronic Engineering, EC1V 0HB, London, United Kingdom. | | Nabil Aouf |
Professor of Robotics and Autonomous Systems, City University of London, Department of Electrical and Electronic Engineering, EC1V 0HB, London, United Kingdom. | | Christophe Honvault |
European Space Agency, Keplerlaan 1, 2201AZ Noordwijk, The Netherlands. |
|
| Abstract:
Aiming at the potential application of explainable artificial intelligence techniques in space tasks, we propose an attention-based deep network for crater detection during Lunar landing scenarios. A methodology combining a fully convolutional neural network and self-attention modules is developed to explore the explainability of automatically crater detection from images. By applying the transfer learning technique, the DeepMoon model is selected as the backbone of the proposed pipeline, and an encoder-decoder based architecture is therefore established and evaluated. The crater images generated using the Blender platform are trained and tested to estimate the performance of the mythology. The self-attention module and data augmentation techniques are applied to the dataset to enhance the segmentation results and improve the generalisation of the implementation. The experiment results on the synthetic greyscale dataset show that the precision, recall, and F1 scores of the crater detection results achieve 0.86, 0.84, and 0.83, respectively. The explainability of the proposed network is achieved by visualisation of each attention map of the attention modules, showing that the deeper attention module pays more attention to the craters.
|
| Keywords: Attention Mechanism; DeepMoon; Crater Detection; Deep Learning |
 View PDF CEAS-GNC-2022-059 |
| Jianing Song, Nabil Aouf, Christophe Honvault: Attention-based DeepMoon for Crater Detection. Proceedings of the 2022 CEAS EuroGNC conference. Berlin, Germany. May 2022. CEAS-GNC-2022-059.
|
| BibTeX entry:
@Incollection{CEAS-GNC-2022-059,
authors = {Song, Jianing and Aouf, Nabil and Honvault, Christophe},
title = {Attention-based DeepMoon for Crater Detection},
booktitle = {Proceedings of the 2022 {CEAS EuroGNC} conference},
address = {Berlin, Germany},
month = may,
year = {2022},
note = {CEAS-GNC-2022-059}
}
|