Lunar Ejecta Pattern Detection and Analysis Using Faster R-CNN and GAN Augmentation

Gaining understanding of the Moon's geological and impact history requires a thorough analysis and inspection of its surface. Understanding the history of impact events, planetary development, crater formation mechanisms, geological composition, lunar regolith, planetary defense strategies, spa...

Full description

Saved in:

Bibliographic Details
Published in:	2024 14th International Conference on Cloud Computing, Data Science & Engineering (Confluence) pp. 652 - 656
Main Authors:	Ghadekar, Premanand, Rathad, Chanchal, Jangral, Sourav, Unde, Rushikesh, Mali, Adwait, Karande, Aryan
Format:	Conference Proceeding
Language:	English
Published:	IEEE 18-01-2024
Subjects:	Analytical models Computational modeling Ejecta Pattern Faster R-CNN Generative adversarial networks Geology Lunar Craters Moon Radiometry Vanilla GAN Weathering
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	Gaining understanding of the Moon's geological and impact history requires a thorough analysis and inspection of its surface. Understanding the history of impact events, planetary development, crater formation mechanisms, geological composition, lunar regolith, planetary defense strategies, space weathering effects, and the broader field of comparative planetary science depend greatly on the analysis of ejecta patterns. This study examines the automated recognition and interpretation of complex lunar patterns from high-resolution lunar pictures using region-based convolutional neural networks. The model is customized and fine-tuned to identify a wide range of craters with prominent ejecta patterns. An essential component for enhancing the model's effectiveness and reliability is the integration of Generative Adversarial Networks. GAN s, a class of generative models, are employed to generate additional lunar images that closely resemble real lunar surface features. This augmentation strategy serves multiple technical purposes, such as improving resistance to noise and artifacts, expanding the training dataset, and enhancing overall robustness. The initial step encompasses essential tasks such as orthorectification and radiometric calibration, aimed at optimizing image quality and ensuring accurate geometric and radiometric characteristics. Subsequently, this study trained the model for feature extraction and region proposal generation. This is followed by region classification and bounding box regression to accurately pinpoint and categorize lunar crater patterns. The analysis encompasses characterizing ejecta patterns based on their size, shape, distribution, and their relationship with other lunar surface features. Proposed model has achieved an outstanding accuracy rate of 85.35% in the automatic recognition and interpretation of intricate lunar patterns.
ISSN:	2766-421X
DOI:	10.1109/Confluence60223.2024.10463225