A Deep Learning-Based Enhancement to OLSR for Robust Attack Detection and Secure Multimedia Transmission in MANETs

Abstract

Wireless Multimedia Sensor Networks (WMSNs) and Mobile Ad Hoc Networks (MANETs) share important features, such as their decentralized topology and multimedia transmission without infrastructure. However, their dynamic nature subjects them to internal routing attacks, including but not limited to, selective packet drop attacks in which the malicious nodes block data flows without detections. This paper presents SA-DCBiGNet which is a hybrid deep learning model that implemented a combination of Dense Convolutional Neural Networks (DCNN), Bidirectional GRUs (Bi-GRU), and Channel Attention as a defense approach in the Optimized Link State Routing Protocol (OLSRP). Using behavioral logs and network metrics, the model identifies adversarial nodes with high precision. The model was evaluated using Python/NS-3 and the WSN-DS dataset and it identified adversarial nodes with high level of accuracy of 99.75% accuracy, 99.30% precision, 99.02% F1-Score, and 98.8% recall, which is better than traditional methods