Classification of similar shaped objects from scattered electromagnetic waves is a difficult problem to solve, as it heavily depends on the aspect angle. Eliminating the effects of the aspect angle is possible by extracting distinguishable features from the scattered signals. These features should be robust to noise effects especially at SNR levels, where noise effects become dominant on the scattered signal. In this paper, we propose a target classification method, which uses a structural feature set extracted from scattered signal. Prior to feature extraction, a multi-scale approximation is performed using hierarchical radial basis function network topology to suppress the effects of noise on scattered signal. After principle component analysis, k-fold cross validation based experiments is performed. Results show that spherical targets are recognized successfully up to -10dB SNR.