Adaptive Identification of Landmine Class by Evaluating the Total Degree of Conformity of Ring-CSOM Weights in a Ground Penetrating Radar System

In the world demining field, ground penetrating radar (GPR) systems are expected to visualize antipersonnel plastic landmines and distinguish them from iron fragments and soil clods. We previously proposed an adaptive texture-classification GPR based on a complex-valued self-organizing map (CSOM). In this paper, we propose a landmine-class identification method utilizing CSOM-space topology that reflects the total similarity of feature-vector values including the SOM-space structure constructed through self-organization. Experiments demonstrate that the proposed method can identify the landmine class to show where a landmine is buried even under a low-resolution and high-noise observation condition.