To search, Click below search items.


All Published Papers Search Service


Classification Based Approach for Spectral Signature of Remotely Sensed Temporal Data


H.Farouk, B.M.Abdel Latif


Vol. 13  No. 11  pp. 8-17


Techniques based on multi-temporal, multi-spectral and satellite-sensor have demonstrated potential as a means to detect, identify, map and monitor ecosystem changes. Multi-temporal images processing becomes more and more important in monitoring earth surface. The large collection of past and present remote sensing imagery makes it possible to analyze spatio-temporal and spectro-temporal pattern of environmental elements. However most existing multi-temporal classification methods use the spectral information alone, ignoring the spatial and temporal correlation between images acquired from different dates, in spite of this represents an amount of information far greater than the individual images. However, their analysis is complex and difficult. This enables to extract evolutions of the same geographic area over time to create a generic spectral signature. In this paper a database of spectral signatures was created for the three main features of the earth, water, vegetation and soil. As large free archives of Landsat 7 ETM+ has been created over time. A temporal series of Landsat 7 ETM+ scenes of different training sites were used to extract the spectral signatures in a reflectance representation by accumulating the individual signatures collected form the individual scenes. The signatures are collected in a statistics form, the mean, minimum, maximum and standard deviations of the training pixels values in the reflectance representation. The database was developed as windows application. Searching has been included to allow users quickly search for the signature of interest. The database was only filled by Landsat 7 ETM+ signatures for the three main features of the earth, water, vegetation and soil which are the features of interest of this study, but it is designed to accommodate other spectral signatures of different satellites and features. Temporal series of Landsat 7 ETM+ scenes of five different lakes located in Egypt are used. The first lake is Burullus Lake at path 177 and raw 38 located in the Nile Delta. The second lake is Qaroun Lake at path 177 and raw 40 located in Fayoum governorate. The third lake is Nasser Lake at path 174 and raw 44 located in Aswan governorate. The selected lakes are typical areas that represent the three main features of the earth. They typically include water, vegetation and soil. That’s why they are selected as study areas. All images are converted to reflectance representation form in order to be independent of the illumination and atmospheric characteristics. In this paper a new spectral classification method is introduced based on the spectral signature database. The classification process is done based on the mean, minimum and maximum of the spectral signature from the database. This paper introduces classification for only the three main classes of the earth, water, vegetation and soil for only Landsat 7 ETM+. However, it is supposed that, classification of any other feature for any other satellite can similarly be done after extracting its corresponding spectral signature and filling in the spectral signature database. A series of tests was applied on the same scenes but with different date to extract the three main classes, water, vegetation and soil. Accuracy assessment has proven that the introduced classifier that uses the spectral signature gives nearly the same results of the supervised classification that uses the signature that was extracted from each image individually. It is supposed that any classifier that uses the proposed spectral signature should give better results than using the individual signature extracted from a single image that is to be classified. Moreover, it is supposed that, the more enhanced classifiers should give better classification results than the introduced classifier.


Classification, Spectral Signature, Signature Database, Satellite Images and Landsat.