|
|
|
Application of Machine Learning Technology to Martian Geology
|
|
|
| New and more powerful machine learning technologies have always been sought to effectively analyze multi/hyperspectral remote sensing data, which have been rapidly accumulating by the ever-increasing number of spacecrafts launched by NASA. In particular, algorithms for more sensitive detection and accurate classification are necessary to help identify, map, and characterize materials with subtle spectral differences such as a diverse suite of rock types that may exist on the surface of Earth or other planets such as Mars. With ever-increasing data from ongoing and near-term missions, such tested and proven techniques will greatly benefit the planetary community, as traditional means of data analysis may not produce the required sensitivities for remote detection and accurate classification of various surface materials. We propose to develop and test an intelligent system composed of a suite of the latest cutting-edge algorithms for the detection and classification of surface materials of weak and/or similar spectral signatures that are otherwise difficult to detect and distinguish by the existing conventional methods. The algorithms include, for example, the kernel-based methods, such as support vector machines (SVMs), which have demonstrated their superiority over many conventional classifiers in terms of classification accuracy and the independent component analysis (ICA) methods, which is capable of unmixing the observed signals to find the independent components with minimum prior knowledge and additional information, and is therefore a powerful tool to separate mixed materials contained in remote sensing image data. While such methods have already found applications in different scientific fields, here we propose to develop an intelligent system by coupling new methods and algorithms with the strength of these methods for the purpose of addressing a variety of space science related problems. In particular, we will use the system to more accurately assess the rock types in the ancient mountain ranges of Mars, Thaumasia highlands and Coprates rise, which includes determining whether the ranges consist of more silicic-rich, mountain building rock. The identification of such a suite of rocks would ultimately lead to an improved understanding of the geological evolution of early Mars. |
|
|
