Using EO1 hyperspectral images for rock units mapping

Abstract

The issue of mapping geological units during an evolving process has now reached a point
where the detection and classification of geological units is carried out with the aid of
hyperspectral sensing. In this study, using hyperspectral the image of the Hyperion sensor,
related to the Khorramabad area in Lorestan province, and using Spectral Angle Mapper
(SAM) and SVM (Support Vectors Machine) algorithms for detecting and separating geological
units after performing the necessary preprocesses, the MNF conversion and PPI algorithm
were used to reduce data and extract pure pixels on the image, respectively. From the
overlapping of pure pixels with geological units and ground data, the average range for each
member was extracted. Field surveys performed at the points provided by the Spectral Angle
Mapter (SAM) confirm the superiority of the SVM method in separating geological units.
Finally, by verifying the accuracy of the algorithms by calculating the error matrix, the
accuracy of the classification of each method are (68.83) and (81.70) is for SVM and SAM
respectively, it was found that at the end of the SVM algorithm with a total accuracy of 81.70
was introduced as the best classification algorithm.