A preliminary study of alteration mapping from airborne geophysical and remote sensing data using feed‐forward neural networks

Abstract

A preliminary experiment was performed to predict altered and non-altered volcanic rocks from airborne geophysical and Landsat data using feedforward neural networks. The experiment was performed separately in areas of felsic and mafic volcanic rocks. About two third of the field stations in felsic volcanic rocks and less than one third of the field stations in mafic volcanic rocks were used to train the neural networks. The trained neural networks was then used to classify pixels of the volcanic rocks. A classification accuracy of 79% for all field stations in areas of volcanic rocks was achieved. This experiment demonstrated that the neural network approach is a potentially very useful tool in alteration mapping from airborne geophysical and remote sensing data.