Multi-scale feature-based matched filter processing

Using the extremely feature difference of self-similarity and kurtosis at large level scale of wavelet transform approximation between the PTFM (Pulse Trains of Frequency Modulated) signals and its reverberation, a feature-based matched filter method using the classify-before-detect paragriam is proposed to improve the detection performance in reverberation and multipath environments. Processing the data of lake-trails showed that the performance of detection in reverberations of the proposed method is better than that of matched filter about 10 dB. In multipath environments, detection performance of matched filter become badly poorer, while that of the proposed method is improved a little. It shows that the method is much more robust with the effect of multipath.