Explainable Convolutional Networks for Crater Detection and Lunar Landing Navigation

TL;DR

This work addresses the need for explainable, vision-based autonomous lunar landing by coupling an attention-enabled Darknet53 backbone with an attention-based YOLOv3 detector for craters and an attention-Darknet53-LSTM RCNN for relative pose estimation. The authors show that incorporating soft attention through CBAM-like modules yields improved crater detection metrics and enables visualization of where the network concentrates its attention, while the temporal RCNN-LSTM component captures motion dynamics for robust pose estimation. They quantify explainability using Pearson correlation coefficients between attention maps and ground-truth masks, and with affine-transform analyses in the navigation module, revealing how different layers contribute to predictions. The approach demonstrates competitive performance on synthetic data generated by the PANGU simulator and offers a path toward more trustworthy, end-to-end autonomous lunar landing systems with interpretable AI components.

Abstract

The Lunar landing has drawn great interest in lunar exploration in recent years, and autonomous lunar landing navigation is fundamental to this task. AI is expected to play a critical role in autonomous and intelligent space missions, yet human experts question the reliability of AI solutions. Thus, the \gls{xai} for vision-based lunar landing is studied in this paper, aiming at providing transparent and understandable predictions for intelligent lunar landing. Attention-based Darknet53 is proposed as the feature extraction structure. For crater detection and navigation tasks, attention-based YOLOv3 and attention-Darknet53-LSTM are presented respectively. The experimental results show that the offered networks provide competitive performance on relative crater detection and pose estimation during the lunar landing. The explainability of the provided networks is achieved by introducing an attention mechanism into the network during model building. Moreover, the PCC is utilised to quantitively evaluate the explainability of the proposed networks, with the findings showing the functions of various convolutional layers in the network.

Figures (14)

• Figure 1: Attention module of
• Figure 2: Proposed attention-based Darknet53 framework
• Figure 3: Proposed attention-based Yolov3 framework for Lunar crater detection during landing
• Figure 4: Block diagram of a recurrent memory unit.
• Figure 5: Proposed attention-based framework for relative lunar landing navigation