Automatic Aircraft Recognition using DSmT and HMM

Abstract

In this paper we propose a new method for solving the Automatic Aircraft Recognition (AAR) problem from a sequence of images of an unknown observed aircraft. Our method exploits the knowledge extracted from a training image data set (a set of binary images of different aircrafts observed under three different poses) with the fusion of information of multiple features drawn from the image sequence using Dezert-Smarandache Theory (DSmT) coupled with Hidden Markov Models (HMM). The first step of the method consists for each image of the observed aircraft to compute both Hu's moment invariants (the first features vector) and the partial singular values of the outline of the aircraft (the second features vector). In the second step, we use a probabilistic neural network (PNN) based on the training image dataset to construct the conditional basic belief assignments (BBA's) of the unknown aircraft type within the set of a predefined possible target types given the features vectors and pose condition. The BBA's are then combined altogether by the Proportional Conflict Redistribution rule #5 (PCR5) of DSmT to get a global BBA about the target type under a given pose hypothesis. These sequential BBA's give initial recognition results that feed a HMM-based classifier for automatically recognizing the aircraft in a multiple poses context. The last part of this paper shows the effectiveness of this new Sequential Multiple-Features Automatic Target Recognition (SMF-ATR) method with realistic simulation results. This method is compliant with realtime processing requirement for advanced AAR systems.