An Adaptive Chirplet Filter for Range Profiles Automatic Target Recognition

An adaptive Chirplet filter approach is proposed to deal with the aircraft recognition problem based on high-resolution range profiles. The Chirplet filter is a joint feature extraction and target identification method derived from the feed-forward neural networks, which consists of two layers: the Chirplet-atoms transform in the input layer is used for replacing the conventional sigmoid function, and the weights between the input and the output layer are taken as the linear classifier. The Chirplet-atoms parameters and the weights are adaptively adjusted by using the nonstationarity degree as the measurement of the features. The simulation results suggest that the adaptive Chirplet filter has advantages especially in noisy conditions.