SAR IMAGE SEGMENTATION USING TEXTURAL INFORMATION AND NEURAL CLASSIFIERS

This paper describes the application of neural network techniques to the automatic segmentation of Synthetic Aperture Radar (SAR) images. Two techniques have been studied : the Multilayer Perceptron (MLP) and the Learning Vector Quantization (LVQ). For an efficient classification, the neural network is applied to a set of features extracted from the SAR image. Examples of suitable features are the radiometric information, the spatial frequency and so on. Two techniques are studied for feature extraction : the Gabor filter and the Principal Component Analysis (PCA). The paper reports the study in terms of correct classification rates related to four classification schemes obtained by a suitable combination of the feature extraction stage and neural network stage.