Deep Learning based Terrain Classification of Mars Raw Images using UNet and FCN Models

The exploration of Mars has consistently been of utmost significance in understanding planetary development, suitability for life, and the possibility of establishing civilizations. Precise classification of Martian terrain is essential for effective navigation, identification of resources, and prev...

Full description

Saved in:
Bibliographic Details
Published in2024 11th International Conference on Computing for Sustainable Global Development (INDIACom) pp. 855 - 860
Main Authors Samhitha, D Sai, Ashwini, K
Format Conference Proceeding
LanguageEnglish
Published Bharati Vidyapeeth, New Delhi 28.02.2024
Subjects
Online AccessGet full text

Cover

Loading…
More Information
Summary:The exploration of Mars has consistently been of utmost significance in understanding planetary development, suitability for life, and the possibility of establishing civilizations. Precise classification of Martian terrain is essential for effective navigation, identification of resources, and prevention of potential dangers, to enhance the effectiveness of Mars missions. Traditional methods of terrain categorization frequently depend on manual examination and have constraints in terms of scalability and precision. This paper introduces an innovative method that utilizes deep learning algorithms to completely transform the process of classifying Martian terrain. This study employed the Unet and FCN models to categorize the Martian surface by utilizing a wide variety of images obtained from the AI4MARS dataset, from which a collection of 5K to 15K images are extracted. The results show the effective performance, the FCN model attained an Intersection over Union (IoU) score of 73%, while the Unet model acquired a score of 60%. The results emphasize the capacity of deep learning to revolutionize the classification of the Martian landscape, providing a more precise and adaptable answer in contrast with traditional approaches.
DOI:10.23919/INDIACom61295.2024.10498424