摘要：In this article, we describe a methodology for the estimate of bathimetry using satellite imagery (IKONOS II) based on the neural network approach. The input variables of the model are the digital values of two spectral bands and the position of the pixel, given by its N, E coordinates. The proposed model consists of an artificial feed forward neural network with two hidden layers. The study reveals that the proposed methodology is able to produce results that reach technical specifications of Diretoria de Hidrologia e Navegação (DHN), in charge of for the bathimetric surveys in Brazil, for class 1 surveys, as the maximum error lies bellow 0,5m. However, it was also verified that the methodology is effcient only for restricted depths, from 0,80 to 3,00 meters, where the spectral response of the water column prevails on the spectral response of the bottom and it is not strongly affected by absorption.
其他摘要：In this article, we describe a methodology for the estimate of bathimetry usingsatellite imagery (IKONOS II) based on the neural network approach. The inputvariables of the model are the digital values of two spectral bands and the positionof the pixel, given by its N, E coordinates. The proposed model consists of anartificial feed forward neural network with two hidden layers. The study reveals thatthe proposed methodology is able to produce results that reach technicalspecifications of Diretoria de Hidrologia e Navegação (DHN), in charge of for thebathimetric surveys in Brazil, for class 1 surveys, as the maximum error lies bellow0,5m. However, it was also verified that the methodology is effcient only forrestricted depths, from 0,80 to 3,00 meters, where the spectral response of the watercolumn prevails on the spectral response of the bottom and it is not stronglyaffected by absorption.
