Deep learning selection of analogues for Mars landing sites in the Qaidam Basin, Qinghai-Tibet Plateau

TL;DR

This paper develops a deep learning method using CNNs to identify and compare landforms in Earth and Mars images, aiding in selecting optimal Mars landing sites based on analogues in the Qaidam Basin.

Contribution

It introduces a novel similarity metric trained on Earth and Mars surface images for landform recognition to assist Mars landing site selection.

Findings

CNN-based model effectively distinguishes landforms.

Similarity metrics improve landing site identification.

Method aids in prioritizing Mars exploration sites.

Abstract

Remote sensing observations and Mars rover missions have recorded the presence of beaches, salt lakes, and wind erosion landforms in Martian sediments. All these observations indicate that Mars was hydrated in its early history. There used to be oceans on Mars, but they have now dried up. Therefore, signs of previous life on Mars could be preserved in the evaporites formed during this process. The study of evaporite regions has thus become a priority area for Mars' life exploration. This study proposes a method for training similarity metrics from surface land image data of Earth and Mars, which can be used for recognition or validation applications. The method will be applied in simulating tasks to select Mars landing sites using a selecting small-scale area of the Mars analaogue the evaporite region of Qaidam Basin, Qinghai-Tibet Plateau. This learning process minimizes discriminative loss function, which makes the similarity measure smaller for images from the same location and larger for images from different locations. This study selected a Convolutional Neural Networks (CNN) based model, which has been trained to explain various changes in image appearance and identify different landforms in Mars. By identifying different landforms, priority landing sites on Mars can be selected.