A spectral-spatial model used for classification of remotely sensed images is presented in this paper. The proposed method benefits from Statistical Region Merging (SRM) algorithm for object extraction as well as Support Vector Machines (SVM) for classification of the objects. A Partial Differential Equation (PDE) based preprocessing step is also provided in order to smooth each band of the multi/hyperspectral data. The SRM-based object extraction step is then applied to the smoothed data and the spectral features of the outcomes are used in the classification process. The parameters of the PDE-based smoothing algorithm are adaptively tuned according to the data texture. The adaptation process is based on Genetic Algorithm (GA) and an innovative fitness function. The efficiency of the proposed GA-PDE-based method as well as the effect of using different values of SRM and PDE parameters on the gained results are evaluated using different criteria. The first measure is the enhancement level in the classification ratios after applying the object-based method. The other measures which are used in this paper are the computational time of the proposed method, the object-to-pixel ratio and finally a recently developed metric which assesses the segmentation efficiency via defining an upper bound for the overall accuracy of classification.