A SOM Approach on Processing Multispectral Acoustic Imaging Data

Abstract

Analysis of three-dimensional acoustic data is a very important step in object detection, especially in robotic detection. It provides rich information of object classification on the region of interest. In this study, an approach on displaying multispectral acoustic imaging (MSAI) data was proposed based on a neural network called Self-Organizing Maps (SOM), which effectively realized dimensional reduction of high-dimensional data. The multispectral acoustic imaging data was obtained from an example that a rigid surface with 9 different holes was illuminated by sound waves sweeping over the frequency range from 1 to 20 kHz with a 30 Hz step. The results showed that the profiles of the holes were identified by their colors in the reconstructed image.