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.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **To achieve crater detection and attitude estimation in autonomous lunar landing navigation, and improve the interpretability of deep - learning models used in these tasks**. ### Problem Background In recent years, with the increasing interest in lunar exploration, autonomous lunar landing navigation has become a key research area. Although artificial intelligence (AI) has achieved great success in various tasks and is expected to play an important role in autonomous and intelligent space missions, human experts have questioned the reliability of AI solutions. Therefore, this paper pays special attention to **explainable artificial intelligence (XAI)**, aiming to provide transparent and understandable predictions for vision - based lunar landings. ### Specific Problems 1. **Crater Detection**: In order to achieve autonomous navigation, it is necessary to accurately identify craters on the lunar surface. Craters, as natural landmarks, can be used for matching and navigation. 2. **Attitude Estimation**: During the landing process, accurate attitude estimation is crucial for ensuring a safe landing. This involves extracting features from a series of RGB images and estimating the relative attitude. 3. **Model Interpretability**: Traditional deep - learning models are usually regarded as "black boxes" and it is difficult to explain their decision - making processes. This paper improves the interpretability of the model by introducing an attention mechanism, enabling researchers to understand how the model makes predictions. ### Solutions 1. **Introduction of Attention Mechanism**: - Use Darknet53 with an attention mechanism as a feature extraction structure. - For the crater detection task, a YOLOv3 network based on the attention mechanism is proposed. - For the attitude estimation task, a Darknet53 - LSTM network combined with an attention mechanism is proposed. 2. **Evaluating Interpretability**: - Introduce the Pearson correlation coefficient (PCC) to quantitatively evaluate the interpretability of the model. - Visualize the attention maps to show the focus points of the model at different layers. ### Experimental Results The experimental results show that the proposed networks have competitive performance in crater detection and attitude estimation tasks. Especially in terms of interpretability, the attention mechanism makes the decision - making process of the model more transparent, which helps to enhance the trust in the AI system. ### Formula Summary - **Loss Function**: \[ L_{obj} = -S^2 \sum_{i = 0}^{S^2} \sum_{j = 0}^{B} \left[ 1_{obj}^{ij} \left( \hat{C}_i \log(C_i)+(1 - \hat{C}_i) \log(1 - C_i) \right)+\lambda_{noobj} 1_{noobj}^{ij} \left( \hat{C}_i \log(C_i)+(1 - \hat{C}_i) \log(1 - C_i) \right) \right] \] \[ L_{ciou} = S^2 \sum_{i = 0}^{S^2} \sum_{j = 0}^{B} 1_{obj}^{i,j} \left[ 1 - IoU+\frac{\rho^2(b, b_{gt})}{c^2}+\alpha v \right] \] where \( v=\frac{4}{\pi^2} \left( \arctan \frac{w_{gt}}{h_{gt}}-\arctan \frac{w}{h} \right)^2 \), \(\alpha = v/(1 - IoU)+v\). - **Pearson Correlation Coefficient**: \[ \rho(a, b)=\frac{\sum_{i = 1}^{n} (a_i - \bar{a})(b_i - \bar{b})}{\sqrt{\sum_{i = 1}^{n} (a_i - \bar{a})^2} \sqrt{\sum_{i = 1}^{n} (b_i - \bar{b})^2}} \] Through these methods.

Explainable Convolutional Networks for Crater Detection and Lunar Landing Navigation

Lunar Crater Detection on Digital Elevation Model: A Complete Workflow Using Deep Learning and Its Application

A Pixel-Level Explainable Approach of Convolutional Neural Networks and Its Application

KrNet: A Kinetic Real-Time Convolutional Neural Network for Navigational Assistance

An Effective Lunar Crater Recognition Algorithm Based on Convolutional Neural Network

Coupling Complementary Strategy to U-Net Based Convolution Neural Network for Detecting Lunar Impact Craters

LCDNet: An Innovative Neural Network for Enhanced Lunar Crater Detection Using DEM Data

An Image Retrieval Method for Lunar Complex Craters Integrating Visual and Depth Features

Optimized Landing Site Selection at the Lunar South Pole: A Convolutional Neural Network Approach

Split-Attention Networks with Self-Calibrated Convolution for Moon Impact Crater Detection from Multi-Source Data

Cross-attention induced multilayer domain adaptation network for extraction of sub-kilometer craters from HiRIC images

Moon Impact Crater Detection Using Nested Attention Mechanism Based UNet++

Automated Crater Detection from Co-registered Optical Images, Elevation Maps and Slope Maps using Deep Learning

Topographic knowledge-aware network for automatic small-scale impact crater detection from lunar digital elevation models

Lunar ground segmentation using a modified U-net neural network

YOLOLens: A Deep Learning Model Based on Super-Resolution to Enhance the Crater Detection of the Planetary Surfaces

Bayesian Network-Based Extraction of Lunar Impact Craters from Optical Images and DEM Data

Recognition and Mapping of Landslide Using a Fully Convolutional DenseNet and Influencing Factors

Automatic Detection for Small-Scale Lunar Impact Crater Using Deep Learning

Lunar features detection for energy discovery via deep learning

Automated Lunar Crater Identification with Chandrayaan-2 TMC-2 Images using Deep Convolutional Neural Networks