الفهرس 
AbstractOnly 14 pages are availabe for public view
 |  | from | 119 |  |  |
| | | from | 119 | | |
Abstract
Images classification is an important task for many aspects of global change studies and environmental applications. This study aims to improve the accuracy of classification for satellite images by comparing the spectral signature that got from the laboratory measurements of spectral reflectance obtained for 18 samples of different land covers and the spectral signature of the selected regions in this classification. In this study three different classifications for the same satellite image have been compared. The first classification is unsupervised classification while the second and third classifications are supervised classifications. The first and second classifications are traditional in ERDAS software while the third classification depends on the laboratory measurements. The samples include most identified land covers from the previous supervised classifications. All collected samples were prepared and then its spectral response were investigated using the spectrometer device (tec 5, Oberursel) which cover the wavelengths range (302 – 1148) nm in 2 nm steps of electromagnetic radiation. These data used to derive the spectral signature for different land covers. This spectral signature is used in the third classification. The accuracy of the first classification was 60% and the second classification was 70% while the third classification accuracy was 85%. This increase in accuracy of classification was lower than expected due to the random regions in the marine area. It is a known problem in LANDSAT7 sensors. This problem has been solved by masking the seawater through making a mask vector by manual digitizing of that class in ARCGIS software which led to an improved accuracy classification of 94.3%.