Image classification algorithm based on the RBF neural network and K-means

This Letter presents the implementation of an algorithm using the radial basis function (RBF) neural networks combined with the technique of K-means for the classification of optical and radar remote sensing images. During the convergence of RBF networks, K-means serves for the initialization of class centres. An automatic self-organizing classification algorithm is constructed based on the RBF networks and split and merge technique. Experimental results show that this algorithm is an effective classifier compared to some conventional methods.