Using Artificial Neural Networks for ultrasonic signals processing from simple geometric shapes

Abstract

The aim of this paper is the recognition of simple plane geometric shapes using Artificial Neural Networks and ultrasound echoes obtained from appropriate objects situated in front of one ultrasonic sensor. For us, the two much more simple plane geometric shapes are: segments and corners. Once the echoes were obtained by the sensor and stored in the computer, we preprocessed them by Fast Fourier Transform Algorithm and then we fed to an Artificial Neural Network with the first one hundred coefficients of the Fourier Serie of each echo obtained from the sensor. The Network is trained by Backpropagation algorithm.