Filtragem de projeções tomográficas da ciência do solo utilizando Kalman e redes neurais

Abstract

This work presents the space variant noise filtering of tomographic projections based on the Kalman filter. For development and filter selection it was evaluated different modalities of the Kalman filter, as well as included the use of Ascombe transform and neural network. Results were analyzed by means of Improvement in Signal to Noise Ratio (ISNR) measurements, which were obtained in a region of interest (ROI) on the resultant images, reconstructed with the use of a backprojection algorithm. In this context the results qualified the unscented Kalman filter with a neural network as the best configuration for filtering of soil tomographic projections.