MMW radar target identification is a hot spot. Character extraction is key of it. So we must filter the noise of echo in order to extract steady character. In this dissertation study MMW radar echo simulation, denoising and character extraction. MMW radar locates optics section, submits to scatter dot model. We simulate range profile of three kinds of tank target shined by high-resolution frequency-stepped MMW radar. Because echo signal is stationary, and wavelet transformation (WT) possess favorable time-frequency localize property, so we use wavelet transformation denoising. Through testing we find db6 wavelet base for echo signal, then improves from thresholding function and selecting thresholding respectively, gain better denoising effect compared to soft and hard thresholding. WT character extraction can compress character dimension and number, then present energy character based Beylkin algorithm, hold low character dimensions and translation-invariant, make design of classification simple. Bispecra is a translation-invariant character, contain abundant phase information, but it is two dimension, and misfit board match. Bring up a bispectra complex character, contain part phase information, and possess translation-invariant and scale-variant, fit target identification. In order to validate character validity, use NearestNeighbor(NN) and probabilistic neural network(PNN) classification identify target, gain content identification probability.
|