Neural Network Classifier for Fighter Aircraft Model Recognition

Abstract

Fighter aircraft recognition is important in military applications to make strategic decisions. The complexity lies in correctly identifying the unknown aircraft irrespective of its orientations. The work reported here is a research initiative in this regard. The database used here was obtained by using rapid prototyped physical models of four classes of fighter aircraft: P51 Mustang, G1-Fokker, MiG25-F, and Mirage 2000. The image database was divided into the training set and test set. Two feature sets, Feature Set1 (FS1) and FS2, were extracted for the images. FS1 consisted of 15 general features and FS2 consisted of 14 invariant moment features. Four multilayered feedforward backpropagation neural networks were designed and trained optimally with the normalized feature sets. The neural networks were configured to classify the test aircraft image. An overall accuracy of recognition of 91% and a response time of 3 s were achieved for the developed automatic fighter aircraft model image recognition system.

Keywords

• rapid prototyping
• fighter aircraft
• feature set
• neural network
• image recognition system