Recuperação de perfis de temperatura e umidade da atmosfera a partir de dados de satélite - abordagens por redes neurais artificiais e implementação em hardware / Atmospheric temperature and humidity retrieval from satellite data - artificial neural networks and hardware implementation approaches

AUTOR(ES)
DATA DE PUBLICAÇÃO

2007

RESUMO

This thesis presents Artificial Neural Networks for inverse problem solution to recover atmospheric temperature and moisture profiles from satellite data. The Artificial Neural Networks are presented as alternative methods in the solution of inverse problems in the atmospheric data retrieval, considered ill-posed problems and requiring advanced numerical techniques to solve them, e. g. regularization methods, when solving by classic methods. Different neural networks are studied, the MultiLayer Perceptron, Radial Basis Function, Hopfield neural networks, and also a variation of the Radial Basis Function. Different kinds of atmospheric sensors of different satellites, and also tested with Global and Brazil data. The analysis of neural networks models is done presenting the data not used in training phase, adding gaussian noise and satellite data. The temperature and moisture profiles obtained by the neural networks are near of radiosonde measurements and are compared with the ones obtained by classic method of inverse problem solution, the regularization method. A hardware implementation in programmable logic device is done and the real-time and on-board atmospheric temperature and moisture retrieval are enable.

ASSUNTO(S)

temperatura e umidade da atmosfera redes neurais artificias

ACESSO AO ARTIGO

Documentos Relacionados