LWIR Identifier Classification using Convolutional Neural Network

Abstract

Combat Identification aims to distinguish friends against foes in combat environment. Utilizing special symbolic identifiers and imagers, automatic identification of combat assets can assist quick and robust decision making. Especially in long-wave infrared (LWIR) band, improved covertness can be expected as the need of luminance diminishes. We have classified, using a CNN, identifiers extracted from LWIR videos. Gaussian noise addition and contrast stretching improved the classifier’s generalization performance and classification accuracy. The proposed method showed 76% accuracy. Especially in challenging conditions, caused by ambient lighting distorting the identifiers, the proposed method still performed with 64% accuracy.