AMM-RF: Adaptive Multi-Modal Meta-Ensemble Learning for Robust RF-Based Drone Detection in Noisy Environments

In this paper, we propose the adaptive multi-model (AMM)- radio frequency (RF) classification model, that significantly enhances drone classification accuracy compared to existing solutions. The model integrates a dual-stream ensemble approach, combining temporal features from long short term model (LSTM)-convolutional neural network (CNN) models and frequency-domain features from Wavelet Packet Transforms (WPT). Simulation results show notable improvements in accuracy, recall, and F1-score, especially under low-SNR conditions. For example, the recall for AR Drone improves from 58.8% to 9 5.2%, and for Phantom, it improves from 8 7.6% to 93.1%. These improvements are attributed to the meta-learner's ability to dynamically adjust the contributions of each base model, enhancing the system's adaptability to varying environmental conditions, including noise, as compared to the baseline algorithm.