The proliferation of Unmanned Aerial Vehicles (UAVs) has brought about transformative advancements across industries, accompanied by new security challenges. This study introduces a novel approach to detecting intruder signals in UAV s using sparsity-based machine-learning techniques. Utilizing the Masked Sparse Bottleneck Centroid-Encoder (MSBCE), an artificial neural network-based feature selection model, we conducted feature pruning on manually derived features from raw WiFi traffic signals. Remarkably, leveraging MSBCE enabled us to achieve exceptional prediction accuracy of 100% across five diverse UAV datasets, even with a sparse feature subset of merely one to two features. This breakthrough promises faster runtime predictions and reduced resource usage (CPU time and memory) on smaller devices, owing to the compact UAV feature set determined by MSBCE. Building on MSBCE's proven effectiveness in high-dimensional biological and big datasets, this study highlights its potential as a state-of-the-art feature selection model in UAV security applications. Through this work, we aim to drive advancements in UAV security, fostering safer and more secure deployment of UAV technology across various domains.
Enhanced UAV Security: Optimizing Accuracy and Efficiency with MSBCE Feature Selection
2024-04-29
1416604 byte
Conference paper
Electronic Resource
English
Optimizing Feature Selection for Industrial Casting Defect Detection Using QLESCA Optimizer
Springer Verlag | 2024
|