Optimized Machine Learning Based Multimodal UAV Detection Using Ensemble Stacking

Unmanned Aerial Vehicles (UAVs) are increasingly used across diverse fields such as medical, aerospace, and military applications. However, detecting UAVs remains a challenging task due to the complexity of the classification process, particularly in environments with diverse UAV types and operational conditions. To address the limitations of single-modality detection, the scarcity of large training datasets, and the increased model complexity, we propose an ensemble stacking approach that integrates multiple modalities using transfer learning. Our approach combines image, audio, and radio frequency (RF) data to enhance the accuracy and reliability of UAV detection systems. By leveraging transfer learning, we reduce data requirements and improve feature extraction from diverse modalities, leading to a more cohesive and robust classification model. The ensemble stacking model was specifically developed to enhance adaptability, providing a scalable solution capable of detecting emerging UAV classes even with limited data. The experimental results demonstrate that ensemble stacking model significantly outperforms single-modality classifiers across metrics such as accuracy, precision, recall, and F1 score. Notably, our model exhibited robustness in scenarios where individual modalities were corrupted or unavailable, thereby reducing false negatives. The improved performance and adaptability of the proposed method have practical implications for enhancing real-time UAV detection in dynamic environments. This research contributes to the development of flexible, reliable, and efficient UAV detection systems for diverse applications.