Feature extraction and recognition based on distributed optical fiber network vibration signals

Phase-sensitive optical time-domain reflectometry (Φ-OTDR) offers advantages such as a simple structure, multi-point vibration localization, and long-distance disturbance detection in optical fiber networks. However, accurately distinguishing diverse environmental vibration events remains challenging. In this study, experiments were conducted using a distributed optical fiber vibration sensing system. Seven common types of environmental vibration events were simulated, and the corresponding signals were collected. Ten time-domain and frequency-domain features were extracted to construct a 10-dimensional feature vector for model training and classification. An improved decision tree optimal ensemble (IDTOE) method was introduced to optimize the random forest model, resulting in the proposed IDTOE-RF classifier. Model performance was evaluated using confusion matrices, accuracy, recall, and F1-score. The IDTOE-RF model outperformed the conventional random forest and support vector machine (SVM) models, achieving an average recognition accuracy of 93.46%, which was 3.84 percentage points higher than that of the conventional random forest model. The proposed method demonstrates good statistical stability and practical applicability for perimeter security monitoring.