Skip to main content
SearchLoginLogin or Signup

Automatic crater detection and classification using Faster R-CNN

Presentation #104.06 in the session Mission-supporting Practices, Modeling, and Data (Oral Presentation)

Published onOct 23, 2023
Automatic crater detection and classification using Faster R-CNN

Planetary surfaces undergo constant transformation driven by geological phenomena such as volcanic eruptions and impacts. Understanding the distribution of impact craters is crucial for comprehending the geological processes, including tectonic deformation, volcanism, erosion, transport, and impact cratering itself [Hartmann & Neukum. 2001]. By analysing this distribution, valuable insights into the geological history and evolution of planetary surfaces can be gained. This research project focuses on the development of a computer vision algorithm that automatically detects and characterises craters in planetary surface images. Our primary objective is to create a comprehensive database of encompassing crater positions, sizes, and characteristics. This database will serve as a valuable resource for planetary scientists, enabling the study of relative age and facilitating a deeper understanding of the geological history and evolution of planetary surfaces.

To achieve this, we used a dataset of high-resolution planetary surface images, including images from the Mars Reconnaissance Orbiter ConTeXt camera (CTX). The CTX mosaic covers 99.5% of Mars [Dickson & al. 2018]. We also use an existing crater database [Lagain & al. 2021] which provide a list of more than 376,000 craters of 1km and more in diameter. We utilise a computer vision algorithm call Faster R-CNN, which has shown promise in detecting objects in images [Ren & al. 2016]. We train the algorithm using a subset of the images, and then test its performance on a separate subsets.

Our preliminary results gave excellent results comparable to the most recent work found in literature [La Grassa & al. 2023, Benedix & al. 2015]. Indeed when we test our model we obtain a mean average precision with an Intersection over Union criterion ≥ 0.5 (mAP50) > 0.8. Therefore, our algorithm consistently performs well regardless of the size of the crater, as long as it exceeds 10 pixels which is the minimum crater size in our training database.

In summary, this research project aims to develop a computer vision algorithm for automatic crater detection and characterisation in planetary surface images. The objective is to create a comprehensive database with crater positions, sizes, and characteristics. Promising results were obtained using the Faster R-CNN algorithm, showcasing its consistency and effectiveness regardless of crater size. This valuable database will enhance the study of relative surface ages and deepen our understanding of planetary geological history and evolution.

Comments
0
comment
No comments here