Semi-automatic land cover classification using ADEOS/AVNIR multispectral data
Nguyen Dinh Duong
Institute of Geography,
National Center for Natural Sicence and Technology of Vietnam
Nghiado, Tuliem, Hanoi, Vietnam
Fax: 84-4-8361192, 84-48352483
E-mail:duong@igg.ac.vn
Abstract
AVNIR is multispectral high resolution sensor on board of Japanese satellite ADEOS that was successfully launched to orbit in August 1996. In the paper the author presents application of GASC (Graphical Analysis of Spectral reflectance Curve algorithm for 4 bands data set of AVNIR in land cover classification. Validation result of proposed algorithm is submitted in the paper for discussion and further improvement of the algorithm).
Introduction
Land cover is one of the most frequently extracted information from Remote Sensing data. It represents a complex mixture of natural and anthropogenic ground classes (Kalensky 1996). Land cover map is used in capturing of current physical situation of land surface for natural resources management and environmental monitoring. Up to now satellite land cover map is made usually by supervised or unsupervised classification method. The first approach requires interactive by any way. The second approach is carried out mainly by clustering technique. This is done automatically based on a number of chosen classes and separability measure among them. To get meaning full final result it requires generally very much manual reclassification with ground truth data collection using different information sources such as thematic maps, field survey and photos etc. The AVNIR sensor on board the Advanced Earth Observation Satellite produces high resolution data in four spectral bands offers new information source for land cover mapping in large scale. In order to speed up classification the author proposed to use GASC algorithm (Graphical Analysis of Spectral reflectance Curve) that is being developed under framework of NASDA ADEOS-II RA for automatic classification of future GLI data. Instead of 6 bands of GLI sensor, the GASC algorithm will be applied for only 4 bands of AVNIR to produce interim product that will be refined by manual classification to meet requirement of land cover mapping in large scale. By application of this algorithm the author expects a shortening of multispectral classification and achievement of more objective and standard interim result that should derive more reliable final land cover map. In order to demonstrate this algorithm the author used ADOES/AVNIR Osaka scene observed on Oct., 17, 1996 level 1B2-GD distributed on CD EORC-011 by NASDA.
