Multi-source threatening event recognition scheme targeting drone intrusion in the fiber optic DAS system

Although fiber-optic distributed acoustic sensing (DAS) has been widely applied for intrusion recognition in the perimeter security field, challenges still remain for high-accuracy recognition of new types of intrusion events. Drone, as a non-contact, stealthy intrusion event, has not been reported to achieve effective monitoring and recognition by employing the fiber-optic DAS technology. By introducing drone intrusion, the scope of the threatening intrusion in the perimeter security monitoring field is extended. This study proposes a multi-source threatening event recognition scheme targeting drone intrusion in the fiber optic DAS system. To achieve this objective, a dual-stage recognition method is proposed. Besides, the wavelet denoising method is applied to extract the effective signal from weak disturbances introduced by drone flight. The variational mode decomposition (VMD)-based hybrid feature vector is formed to remove the noise and further extract the effective signal characteristics. Next, we demonstrate a hybrid model framework based on convolutional neural network (CNN)+long short term memory (LSTM)+self-attention mechanism to achieve the effective recognition of multi-source threatening event targeting drone intrusion. Particularly, we thoroughly discuss the distinguishing ability between drone intrusion and non-threatening wind blowing, and the recognition ability for the simultaneous occurrence of both human-contact intrusion and non-contact drone intrusion. The experimental results show that, the proposed recognition scheme can effectively distinguish the drone intrusion from the non-threatening wind blowing event with a high accuracy of 100%. In addition, when drone intrusion, environmental disturbances and human-contact intrusion occur simultaneously, a high recognition accuracy of 96.25% is achieved with a fast response time of 0.733 s.