Unmanned aerial vehicle intrusion detection: Deep-meta-heuristic system

Abstract

The UAV (Unmanned Aerial Vehicles) is an automatic aircraft, widely used several applications like emergency management, wildlife conservation, forestry, aerial photography, etc. The communication among the UAV is susceptible to security threats with several diverse attacks. The data sharing among the UAV and other vehicles is vulnerable to jamming and suspicious activities that disturbs the communication. To tackle the issue, IDS (Intrusion Detection System) is the significant system that monitors and identifies the suspicious activities in the communication network. To attain this, several conventional researchers attempted to accomplish better intrusion detection. However, classical models are limited by accuracy, noise and computation. To overcome the limitation, proposed method employs particular set of procedures for the intrusion detection in UAV with Intrusion UAV dataset. The dataset comprise of features like drone speed, height, width, velocity etc. Initially, in the respective approach, GG (Greedy based Genetic) algorithm for feature selection, which maintains the exact balance between the greediness and diversified population. Greedy approach enhances Genetic algorithm in combinatorial optimisation problems. Further, the study proposes Modified Deep CNN-BiLSTM (Deep Convolutional Neural Network and Bi-Long Short Term Memory) with attention mechanism for classification of intrusion in UAV. The deep CNN is utilized for the ability of handling larger datasets and accuracy. Conversely, it is limited by computation and speed. To tackle the problem, Bi-LSTM is used for the capability of enhancing the computation and speed. Moreover, attention mechanism is used for handle the complexity and to permit the presented system to focus on the significant and relevant data. Correspondingly, proposed approach performance is calculated using performance metrics such as accuracy, specificity, sensitivity, R2 (R-Squared), execution time, RMSE and precision. Furthermore, comparative analysis of the proposed method and classical model exposes the efficacy of the respective system.