Evaluation of an embedded unscented Kalman filter for soil tomography

Marcos A. M. LaiaPaulo E. Cruvinel

With the flexibility of using the Field Programmable Gate Arrays (FPGAs) is the possibility of using algorithms that were previously validated in general-purpose computers, as well as the use of most modern processors using proper management of energy expenditure and use of resources like memory and instruction cycles, and allows the use of parallel processing to eliminate bottlenecks and computational analysis of dynamic problems of the agricultural environment. The problem of estimating a noise-free signal involves foreknowledge of system variables, which is not always perfect due to the behavior of systems are influenced by the environment in which it is and the quality of the components of sampling equipment, and may be a optimal solution for control and measurement systems. Agricultural soil tomography aims at investigating soil proprieties as water and solute transport, soil porosity, soil contents, root growing and humidity. For a better analysis about these proprieties, an image quality is required. Previous works focused on image filtering or in the use of filters specialized in Gaussian process estimation and in an implementation in general purpose computers that have high processing power and memory but with a processing time depends on the resolution of the data to be acquired and the number of neurons in the neural network. This paper presents formulations for the use of unscented Kalman filter with neural networks in a joint estimation filtering (a filter for state and weight estimation) implemented in an embedded system with the objective of obtaining better quality in the signal / noise relation of the tomography projections. The aim of this work is to use the filter on a dedicated system as an invisible module to receive and filter the projections.

Caso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web

Biblioteca Digital Brasileira de Computação - Contato:
     Mantida por: