A seafloor classification methodology, based on a parametrization of the reverberation probability den sity function in conjunction with neural net classifiers, is evaluated through computer simulations. Different seafloor "provinces" are represented by a number of scatterer distributions exhibiting various degrees of de parture from the nominal Poisson distribution. Using the computer simulation program REVGEN/SST, these distributions were insonified at different spatial scales by varying the transmitted pulse length. The statisti cal signature obtained consists of reverberation kurto sis and coherent component estimates as a function of pulse length. The adaptive neural network algorithms are trained through supervised learning to recognize each statistical pattern and are presented with the task of dis criminating among the various scatterer distributions. The initial results indicate that this approach offers con siderable promise for practical, realizable solutions to the problem of remote seafloor classification.
[1]
Julius T. Tou,et al.
Pattern Recognition Principles
,
1974
.
[2]
Yoh-Han Pao,et al.
Adaptive pattern recognition and neural networks
,
1989
.
[3]
A. A. Mullin,et al.
Principles of neurodynamics
,
1962
.
[4]
Geoffrey E. Hinton,et al.
Learning internal representations by error propagation
,
1986
.
[5]
Carl Eckart,et al.
The Scattering of Sound from the Sea Surface
,
1953
.
[6]
Richard P. Lippmann,et al.
An introduction to computing with neural nets
,
1987
.