Research on remote sensing image classification based on Kohonen neural network
基于Kohonen神经网络的遥感影像监督分类

Kohonen algorithm has defects that initial weights are randomized, learning rate and neighborhood are difficult to determine, and as a clustering algorithm, kohonen algorithm is difficult to be used directly for supervised classification. In this paper, we used max-min distance method to determine Kohonen Network' initial weights, introduced weight error as the convergence condition of network, and improved Kohonen network' unsupervised-learning algorithm. We then used the improved Kohonen' learning algorithm combined with penalty formula to carry out supervised classification of remote sensing data. Experiments showed that higher accuracy resulted from this improved algorithm.