Application of Machine Learning Techniques for Remote Sensing in Lithological Mapping: a Review

EasyChair Preprint 15804

Abstract

Identification of the geological features with the help of maps plays a crucial role in mineral exploration. The traditional way of geological data collection in remote areas is time-consuming and also challenging, thus the images extracted from the Remote Sensing technique are followed by image processing, which assists in classifying and improving geological mapping more accurately. Remote Sensing technology can now map different litho-units and associated structural features with greater precision and speed. In this paper, a comparative review is done based on remote sensing and image processing using different machine-learning methods. Images obtained from many satellites popularly used by geologists for geosciences have been explored and compared. Secondly, different machine-learning methods for processing these images are analyzed and their accuracy is compared. The comparative result concludes that the satellites that are first and foremost for such studies are of multispectral types (e.g. ASTER, Sentinel-2 and Landsat) due to the historical coverage. The survey concludes that when Landsat-8 images are used with the SVM give output accuracy of more than 80%. At the same time, the Random Forest is a technique, that uncovers the potential of remote sensing to address the emerging problems in Geographic Information Science. With only a sizable number of experiments performed through deep learning, provides results with more than 90% accuracy highlighting the supremacy of deep learning in geographical and remote sensing applications.

Keyphrases: Lithological Mapping, Mineral Exploration, machine learning, remote sensing, spectral reflectance

@booklet{EasyChair:15804,
  author    = {Tanushree Parmar and Romit S. Beed},
  title     = {Application of Machine Learning Techniques for Remote Sensing in Lithological Mapping: a Review},
  howpublished = {EasyChair Preprint 15804},
  year      = {EasyChair, 2025}}