Application of the Constructive Mikado-Algorithm on Remotely Sensed Data

摘要: Finding an optimal architecture for a neural network, i.e. number and size of layers, is open problem. With the Mikado-algorithm we present new method to construct network in course learning. two examples, classification structure detection from remotely sensed data, demonstrate capabilities Mikado-algorithm. This algorithm provides good generalisation presence mixed pixels, delivering small networks high-dimensional problems way interpreting ability.