MALICIOUS UAVS CLASSIFICATION USING VARIOUS CNN ARCHITECTURES FEATURES AND MACHINE LEARNING ALGORITHMS

Abstract

Aircraft are used in many fields such as engineering, logistics, transportation and disaster management. With the development of drones, aerial vehicles have become more widely used for entertainment purposes. However, in addition to its useful applications, its malicious use is also becoming widespread. It has become a necessity to eliminate this problem, especially since it poses a significant danger to other aircraft. In order to identify the aircraft and solve this problem quickly, in this study, five different aircraft were classified based on images. In the study, a five-class dataset containing aeroplane, bird, drone, helicopter and malicious UAV (Unnamed Aerial Vehicle) images was used. Three different CNN (Convolutional Neural Network) models were employed to extract the images of features. Image features extracted with SqueezeNet, VGG16, VGG19 models were classified with Artificial Neural Network (ANN), Support Vector Machine (SVM) and Logistic Regression (LR) machine learning methods. As a result of the experiments, the most accuracyful result, 92%, was obtained from the classification of the features extracted with the SqueezeNet model with ANN. The models proposed in the study will be integrated into various systems and used in the field of aviation to detect malicious UAVs and take necessary precautions.