Machine Learning Based Classification of Ducted and Non-Ducted Propeller Type Quadcopter

Abstract

Fast and efficient means of classifying a multi-copter(drone) according to its radar cross-section (RCS) is important. These multi-copters are characterized as LSS target which is an acronym for low altitude, slow speed, and small RCS, hence identifying and classifying them is quite difficult. In this paper, we tried to classify between ducted and non-ducted propeller quad-copter drones. For this we used Machine Learning techniques. Here we proposed the use of three models namely fine k-NN (k-nearest neighbor), fine Gaussian SVM (Support Vector Machine) and a two layered feed forward neural network. In each of the models four parameters were considered frequency, angle of elevation ($\Theta$), azimuth angle ($\Phi$) and measured radar cross-section (RCS) values. The classification accuracy in case of fine k-NN keeping the distance metric as Chebyshev it varies from 75.2% to 76.6% depending on the number of neighbors. In case of fine Gaussian SVM accuracy is 76.2% and for feed forward Neural Network (NN) it is 75.1% to 75.5%.