Deep Learning-based approach for detection and classification of Micro/Mini drones

Abstract

In the recent years, the micro/mini drones' industry has witnessed an explosive growth, making these flying objects become highly accessible to Terrorist groups. This phenomenon has caused specific security concerns due to the fact that these suspicious flying gadgets can cause serious hazards. To protect the sensitive locations and restricted areas, we suggest, in this paper, a drone detection method that integrates deeplearning-based classification and localization tasks. Specially, we selected a family of fast and accurate one-stage object detector: YOLOv3. So, we use and improve YOLOv3 deep learning neural network, by upgrading its architecture and fine-tuning its parameters to better accommodate small object detection such as micro/mini drone. Furthermore, to train our algorithm to classify the detected drone, we have constructed a multi-class drone dataset consisting of drones' images that may fly in Tunisian airspace and among which some may be a possible threat.