Texture Classification of SAR image by Neural Network parameterised by Wavelet Transform

摘要

The parameterisation of the image by the wavelet coefficients permitted to return the not correlated data. These data were used as entries of Neural Network for the learning and the classification of Synthetic Aperture Radar images. The developed method basing on the parameterisation of the Neural Network entries was applied to an airborne SAR image which presents various types of textures : bare ground, batis+vegetation, olive trees on ground bare and olive trees on grounds place setting vegetation. The classification with this type of architecture allowed attaining assessment identification of the order of 92.5 %. These results were compared in term of assessment identification with those obtained by the method of classification by a Neural Network and by a Neural Network parameterised by the energies of sub-bands born from redundant wavelet decomposition.