A neural-network-based classifier applied to real-world aerial images
Shlomo Greenberg, Hugo Guterman · 2002
The classification and recognition of real-world aerial images, independently of their position and orientation, by using neural network are discussed. Invariance feature spaces which have been used in conjunction with neural nets are not invariant to all possible transformations and required an extensive computational preprocessing. In the proposed method the invariance is achieved by training a neural network (NN) with a large number of appropriate distorted scene samples. The performance of the neural network classifier is compared with the classical correlation based techniques. Invariant classification of shifted and rotated real scene image is shown to be feasible.>