Threat target classification using ANN and SVM based on a new sensor array system

Abstract

Electromagnetic imaging is based upon the fundamentals of electromagnetic (EM) fields and their relationship with the material properties under evaluation. A new system based on a Giant Magneto-Resistive (GMR) sensor array was built to capture the scattered EMsignal returned by metallic objects. This paper evaluates the new system's capabilities through the classification of metallic objects based on features extracted from their response to EM fields. A novel amplitude variation feature as well as the combinations of typical features is proposed to obtain high classification rates. The selected features of metallic objects are then applied to well-known supervised classifiers (ANN and SVM) to detect and classify 'threat' items. A collection of handguns with other commonly used metallic objects are tested. Promising results show that a high classification rate is achieved using the proposed new combination features and classification framework. This novel procedure has the potential to produce significant improvements in automatic weapon detection and classification.