Preliminary communication
https://doi.org/10.31298/sl.142.1-2.6
Image fusion influence on forest area change using unsupervised classification
Luka Rumora
orcid.org/0000-0002-2362-6685
; University of Zagreb, Faculty of Geodesy
Mario Miler
orcid.org/0000-0003-3883-8493
; University of Zagreb, Faculty of Geodesy
Damir Medak
; University of Zagreb, Faculty of Geodesy
Abstract
Demand for high quality free satellite data is increasing. Currently the most popular and known mission is Landsat satellite mission. This mission ensures ground resolution of 30 m x 30 m. For some application, this ground resolution is not sufficient. Landsat mission, starting from the Landsat 7 satellite, collects panchromatic band that is used to increase resolution of images.
This paper analyzes the impact of multispectral and panchromatic image fusion on unsupervised classification. Based on original recordings NDVI (Normalized difference vegetation index) is calculated. This indexed image is used as reference image for the purpose of further comparison. The original images of first four bands (blue, green, red and near-infrared) are sharpened using eighth (panchromatic) band gathered with Landsat 7 satellite. From this bands, based on forth and third band, NDVI is calculated. With this calculation it is conducted pansharpening of reference NDVI image. Images for classification was chosen by comparing obtained images. Selected images was classified with K-means unsupervised classification algorithm, and it was determined that image calculated with bicubic interpolation and sharpened with fast intensity-hue-saturation (FIHS) algorithm on previously sharpened bands represents the best solution.
Keywords
image fusion; normalized difference vegetation index; Landsat 7
Hrčak ID:
194636
URI
Publication date:
28.2.2018.
Visits: 2.424 *