Evaluating the Close Range Hyperspectral Data for Feature Identification and Mapping

Abstract

Exploring enhanced spectral information embedded in hyperspectral remote sensing (HSRS) datasets is a frontier field of remote sensing and bears the potential of detailed information extraction. This study involves the utility of high spatial resolution HSRS data for enhanced feature identification and classification based on spatial and spectral information. For the purpose of the study, close range hyperspectral dataset was used that was acquired using HySpex VNIR-1800 camera developed by Norsk Elektro Optikk. The image was converted to surface reflectance, and classification was employed on a total of 9 land cover features identified. Pixel-based classification was performed using support vector machine and compared with the classification result of object-based classification. For performing object-based classification, multiresolution segmentation along with fuzzy rule-based classifier was applied on the optimal spectral bands and minimum noise fraction components generated from the HSRS dataset. The results indicate an enhanced capability of object-based classification (92.5%) over pixel-based classification (71.24%). It was observed that the resultant classified images show misclassification of natural features in pixel-based classified output. Object-based classification approach has advantage over other approach in terms of accuracy with Kappa score of 0.89 over pixel-based kappa score of 0.62.