Results and Discussions The DMZI-UAV-based combination security system achieves high identification accuracy of five typical sensing events. Although the corresponding time-domain waveforms of the five typical sensing signals are highly similar to each other, their corresponding STFT spectral distributions show significantly different features (Fig. 6). In this experimental test, the image of the original data consists of two parts: one is the two-dimensional STFT spectrogram corresponding to the one-dimensional vibration sensing signal, and the other is the real-time image captured by the camera mounted on the UAV (Fig. 7). Specifically, 5800 samples are collected and processed during the field test to verify the feasibility and the effectiveness of the proposed model. The size of the data collected by the DOFVS system is 3000, and the size of the data captured by the UAV is 2800 (Table 1). During training, the labeled images in the training set are trained for 150 epochs in the YOLOv5s network model after parameter adjustment. After about 100 epochs, the classification loss and localization loss tend to be stable, and the mAP@0.5:0.95 remains above 90% (Fig. 8). Then, the comparison of the identification and classification results of three different datasets shows that the overall identification performance of the original one-dimensional data is significantly lower than that of the Mel spectrum identification scheme and the STFT identification scheme (Table 3). The results reveal that the dataset of the third scheme can obtain better event identification performance, which greatly enhances the analysis and processing ability of vibration signals. Finally, in the testing phase, the detection time of the proposed model is at the level of milliseconds, which can fully meet the real-time detection requirements of practical engineering applications (Table 4).

The distributed optical fiber vibration sensing (DOFVS) system is a pre-alarm system based on security monitoring technology, which can realize continuous distributed detection and measurement of vibration events along single optical fiber links. The DOFVS system has many advantages such as high positioning accuracy, a large monitoring range, simple structure, and easy installation, and it has been widely and successfully used in many vibration sensing fields, such as long-distance oil and gas pipeline leak detection, security monitoring of transmission line networks, and perimeter security monitoring. However, due to the complexity and diversity of its application environment, the DOFVS system still faces problems such as low reliability and poor stability in practical applications. In our research, we propose an intelligent sensing detection scheme, which combines the DOFVS system and artificial intelligence (AI). This scheme can significantly improve the practical reliability and stability of the DOFVS system in engineering applications.

This paper proposes an accurate detection scheme for multiple optical fiber vibration sensing events based on the You Only Look Once version 5s (YOLOv5s) model by integrating the dual Mach-Zehnder interferometer (DMZI) system and the quadrotor unmanned aerial vehicle (UAV) monitoring system. When an intrusion event occurs, the DMZI system transmits the location of the disturbance point to the UAV via Qgroundcontrol. After the UAV flies to the disturbance point, the camera on it can automatically capture and photograph the surrounding environment of the vibration position in real time and then transmit the real-time image information back to the ground station through the first-person view (FPV). First, the DMZI system and the UAV system are controlled by the ground station Qgroundcontrol. Second, the short-time Fourier transform (STFT) is performed to obtain the corresponding two-dimensional spectrum from the one-dimensional time-series signal. Third, the spectrum of the two-dimensional vibration signal and the corresponding original images captured by the UAV are jointly sent into the YOLOv5s-based convolutional neural network (CNN) model for identification and classification. Fourth, massive experiments are carried out to verify the effectiveness and feasibility of the proposed scheme. The mean average precision (mAP) and identification times of the five sensing events are measured to demonstrate the performance of the proposed scheme.

According to the application requirements, this paper proposes and designs a vibration identification scheme based on the DMZI-UAV-fused security system, which is realized by the combination of STFT and the YOLOv5s algorithm. By the DMZI-UAV-based combination security monitoring system, the features of the optical path signal from the perspective of time and frequency can be effectively extracted. Moreover, the proposed scheme can also discriminate and classify the intrusion events in the actual space with high efficiency. The method based on the YOLOv5s algorithm can automatically extract features, which avoids the low robustness problem in manual feature extraction. The effectiveness of the method is verified by the detection of five common sensing events, namely, no intrusion, waggling, knocking, crashing, and fence kicking. The training results show that the mAP for the five sensing events is all above 95%. Furthermore, the field test results demonstrate that the proposed scheme can accurately identify and classify five typical sensing events, with mAP of 96.6%. Meanwhile, compared with traditional machine learning and other deep learning schemes, the proposed scheme has a significantly shorter response time that can be controlled within 5 ms. Therefore, we believe that the proposed scheme can improve the reliability and stability of the DMZI DOFVS system in practical engineering applications.