Project Description
The Flagstaff Campus of USGS serves the nation, general public, and science community by providing new knowledge about our solar system. Dr. Ryan Anderson at USGS conducts research aimed at allowing us to better understand the origins, evolutions, and history of Mars through geological processes. The goal of our senior capstone project was to provide USGS with a more efficient approach to map terrains such as valleys, canyons, sinuous ridges, and sand dunes on Mars’ surface. Currently the method of terrain mapping used at USGS is a manual process that requires annotating terrains by hand, which can take months to complete. This process is not only inefficient, but is also inconsistent due to time constraints and human fatigue-, which can lead to mistakes.
Solution Overview
We developed a customized software solution that automates the annotation process by taking in an orbital data set with a terrain type of interest (e.g., canyons) and applying a neural network that will detect similar terrain in an efficient amount of time. More specifically, the designed computer program will allow human users to automate the task of identifying characteristic terrain types on Mars’ surface by loading HiRISE (i.e. high resolution) images into the system for processing, have the neural network learn to recognize certain terrain types, then produce the results as an image. The key pieces of functionality for the planned system involve pre and post processing of a data set and using a Machine Learning algorithm to recognize patterns on the input orbital data.
Future Work
Our sponsor Dr. Ryan Anderson plans on hiring a student to continue the development and and maintaning of the automated terrain mapping system we developed for our capstone. The student worker will continue to test the system as well as continuing to add features to create a more diverse system. An example of a feature to be added to the system can include training the convolutional neural network to map other terrains, since we were only provided training data to map sand dunes.