Satellite images fusion using possibility theory

Bouakache, Abdenour; Tahraoui, Ahmed; Khedam, Radja; Belhadj-aissa, Aichouche

Bulletin des sciences géographiques de l'INCT
Volume 21, Numéro 1, Pages 28-35
2017-10-31

Satellite Images Fusion Using Possibility Theory

Authors : Bouakache Abdenour . Tahraoui Ahmed . Khedam Radja . Belhadj-aissa Aichouche .

Abstract

The aim of this work is the improvement of land use map and the detection of changes in a region of globe. To this end, the database used consists of multisource multitemporal satellite images using possibility fusion process. The crucial problem in the development of this process is the estimation of possibility functions. We have, for this purpose, applied transformations from probability distributions to possibility distributions. Thus, we propose two methods to implement the fusion process:In the first method, we implemented a model of possibility fusion: we estimated the probability distribution of each spectral class samples of the training set. This estimate is based on the histogram analysis of each class. Then we estimated the possibility distribution from probability distributions using the transformations: Dubois and Prade, improved Dubois and prade, Klir and variables transformation (VT). Next, we determined the possibility of each pixel with linear interpolation method. For the combination operator, we opted for the conjunction operator (severe) and disjunction operator (indulgent). Finally, we applied the decision rule based on the maximum of possibility. In the second method, we implemented the fusion process monoband and multiband, whose mono-band fusion does not require a combination step. The results obtained represent maps containing classes and relatively well discerned different each other. Therefore, we used the fusion process multiband by exploiting the complementarity of their spectra and we obtained maps with predefined classes in the training set on which the spectral bands emit the same decision and a class confusion on which the spectral bands differ. The plot of the spectral signature of the confusion class is a curve which has an intermediate form between the signatures of predefined classes.Finally, in the third method, we implemented the multisources multisensor multitemporal process of fusion where we have combined two multispectral images from the two sensors HRV of SPOT satellite and ETM + of LANDSAT 7 satellite, acquired respectively in 1997 and 2003 (different dates). For invariant zones, the result is a map containing predefined classes of sensors which on emit the same opinion and a confusion class which on the sensors are different. However, for variants zones, the result is a change map containing predefined stable classes and a change class represented by the class of confusion.