UAV-Assisted Victim Localization for Disaster Scenarios Using Channel-Aware Deep Learning and Swarm Optimization

Accurate indoor localization in post-disaster and complex environments remains a challenging task due to severe multipath propagation, signal blockage, dynamic line-of-sight (LoS), and non-line-of-sight (NLoS) conditions. To address these challenges, this paper proposes an Unmanned Aerial Vehicle (UAV)-assisted Received Signal Strength Indicator (RSSI)-based localization framework that jointly exploits deep feature learning, channel awareness, and swarm-based optimization. An enhanced feature combinational with bidirectional long short-term memory (EFC-BiLSTM) network is proposed for robust temporal feature extraction, which classifies into Line-of-Sight (LoS) or Non-Light-of-Sight (NLoS) signal. The extracted channel features, along with measured distance and LoS/NLoS information, are fed into a multilayer perceptron (MLP) to refine RSSI-based distance estimates. Finally, the corrected distances are then integrated into an adaptive particle swarm optimization (APSO) algorithm to estimate the target position, which improves the convergence stability. The proposed framework is evaluated in three different scenarios that show varied propagation characteristics. The proposed method attains root-mean-square-error values of 0.7024 m, 0.6332 m, and 0.9124 m in multi-layout scenario, semi-open corridor, and laboratory environments, respectively. Furthermore, the proposed method evaluates node localization efficiency (NLE) across varying error thresholds, resulting in enhanced reliability and node coverage.