Explosive detection system based on Leddar sensor and Self-Organizing Maps in controled environments

Abstract

An explosives detection/identification system is presented to be used by a humanoid robot that must manipulate them, using a Leddar sensor and a SOM (Self-Organizing Map) network as data acquisition and processing tools, respectively. By creating a sensor/PC interface, a database with 16 distance measurements is created for each sample. These samples were of 2 kinds: explosives (object of detection, cylindrical) and a test object (rectangular). In total, the SOM network was trained with 100 vectors of 16 distances, achieving the separation of the 2 clusters, which is evidenced in the validation where 100% of the samples of the “explosive” pattern are grouped to the upper right side of the neurons output.