Acoustic Source Drone Detection System Using Tetrahedral Microphone Array and Deep Neural Networks

The rapid integration of Unmanned Aerial Vehicles (UAVs) into civilian airspace has introduced complex security challenges, particularly regarding the protection of critical infrastructure and personal privacy. While conventional detection mechanisms such as radar and optical sensors are widely deployed, they are frequently limited by line-of-sight obstructions and the small radar cross-section of modern commercial drones. Acoustic analysis presents a viable passive alternative; however, accurate three-dimensional localization remains a computationally demanding task, further complicated by the use of directional sensors with non-uniform sensitivity patterns. In this paper, a deep learning framework is proposed to address these ambiguities. The method involves the fusion of raw acoustic data with explicit sensor geometry metadata within a neural network architecture. To enhance localization precision, a composite loss function is introduced, which independently optimizes planar and altitude coordinates while penalizing outlier predictions. Experimental validation was conducted using a custom dataset of real-world drone flights, utilizing a distributed array of directional microphones. The results demonstrate that the proposed system effectively mitigates the spatial irregularities of ad hoc sensor deployment, achieving robust localization performance in complex acoustic environments.