Objects classification based on UWB scattered field and SEM data using machine learning algorithms
Résumé
This paper addresses the target classification problem using supervised learning techniques to discriminate spherical objects from their scattered field. The main goal is to demonstrate that pre-processed data provide a higher accuracy for classification purposes in comparison with raw data. To this end, we compare the classification performances when using raw data, in time and frequency domains, and pre-processed data by the Singularity Expansion Method (SEM). The first step is to build 3 datasets from mono- and bi-static Ultra Wide Band scattered fields; each containing 5 sphere classes with different materials. Then, we evaluate the performances of several classifiers based on Machine Learning algorithms trained using those constructed datasets. Applying these algorithms on the data resulting from pre-processing the scattered field with SEM proved to be more successful and allows the use of simpler classifiers.