Abstract
Recently, drones have become widely used in almost everyday life and at many crowded events due to their versatility. There were incidents involving them, both intentional and unintentional. One of the most essential studies of preventing drone occurrences could involve evaluating the recognition of their distance to people and protected areas. The potential performance of deep learning methods in the field of recognition has led to motivation to explore the distances of UAVs based on their sounds. Therefore, this study examined a deep learning method based on Gated Reccurent Unit (GRU) to study the classification capability of drone sounds, particularly the classification capability of drone sounds flying at different distances. So, we tried to study whether the sound classification method has the ability to recognize drone flights at different distances. During our experimental tests, drones were launched at intervals of every five meters from ground level. According to experimental studies, drone sounds can be recognized from a distance of every 10 meters or more, and the average recognition accuracy in such tasks can be 94-98 percent. Such a system can be offered into a complex recognition system, which can be inserted as a functional part of bimodal and multi-modal systems.