Generalized target description and wavelet decomposition (sonar)

Abstract

Generalized target description by means of colored bright spots (Altes [Z]) is very attractive for recognition or classification tasks in active sonar applications. In this correspondence, we show how such a description can be achieved directly from the impulse response to identify. It turns out that the resulting procedure is in close connection with the recently introduced technique of wavelet decomposition.

<title>Target classification using wavelet decomposition features</title>

“…The wavelet needed however, is the autocorrelation of the Altes signal u(t). Moreover, it is shown that such a wavelet is admissible [6] and its shape partially resembles the Grossman Monet wavelet [6] for specific choices of the constants n,g, and c. The multiscale representations of the impulse response by wavelet decomposition correspond to different levels of integration and differentiation and hence different geometrical substructures of the target. It is shown in [81 that by using the autocorrelation of u(t) as a wavelet, then the scale parameter (a) is defined as a=g.…”

“…Flandrin et al [6], noted that the matched filter based system proposed by Altes [29] is equivalent to evaluating the wavelet transform of the target impulse response. The wavelet needed however, is the autocorrelation of the Altes signal u(t).…”

“…Also, wavelet decomposition provides the needed balance between time and frequency localization without violating the Heisenberg uncertainty principle. The signal f(t) can be recovered from its redundant wavelet transform expansion using f(t)=-i-ff W/a,b)4*!-) db (5) a5b a assuming that fp(t)dt=O (6) and f IF(w)I2 < (7) J Fourier transform and Fourier series methods require many Fourier expansion coefficients corresponding to high frequency components to characterize a transient signal and perform the necessary cancellation, thereby permitting the inclusion of high frequency noise. In contrast to that, wavelet analysis permits a selective mode representation of the signal due to the compact nature of the analyzing wavelet (limited duration of J, ( t) ), and therefore is particularly suited for analyzing SPIE Vol.…”

“…The extracted information are basis for target identification in a maximum likelihood sense. Recently, the problem of estimating the location of target scattering centers and their respective amplitudes, in the framework of spectral analysis, has received considerable attention [2,3,6,29,30,31]. Methods using Fourier Transform and high resolution spectral estimation algorithms have been used to estimate the parameters The Fourier transform based techniques are limited in resolution and are often incapable of resolving two closely related scatterers, but perform adequately when used for target recognition particularly under low signal-to-noise ratios.…”

“…On the other hand, high resolution techniques provide highly accurate estimates of the target scattering centers under high SNR but their performance deteriorates severely at low SNR. High resolution spectral estimation techniques such as Prony method, MUSIC, ARMA modeling, Maximum Entropy method have been used for target signature analysis [2,3,6,29,30,311, and scattering mechanisms identification. These methods produce accurate estimates of the location of scatterers based on a point scattering model and hence ignore the effects of differentiation and integration associated with each scattering component.…”