Graph Attention-Guided Three-Branch Network for Feature Fusion and Recognition of Distributed Optical Fiber Vibration Signals

Accurately recognizing vibration signals remains a significant challenge in the field of distributed optical fiber sensing. To address this issue, this paper proposes a novel graph attention-guided three-branch network. Firstly, the three-branch network is used to extract multimodal features from various signal forms. The first branch employs Transformer to extract temporal features from the signal in the form of time series. The second branch utilizes the enhanced gated recurrent unit (GRU) to capture spectral features from the spectral sequences. The third branch adopts the improved convolutional neural network (CNN) to extract time-frequency features from the time-frequency images. Secondly, to effectively integrate the extracted features, the branch attention mechanism is introduced. This mechanism dynamically assigns weights to the temporal, spectral, and time-frequency features, realizing the efficient fusion of multimodal features. Finally, considering the potential correlations between different samples, graph attention network (GAT) is incorporated into the three-branch network to further optimize the extracted multimodal features. Experimental results demonstrate that the proposed three-branch network can extract more informative and discriminative features than other models. Moreover, the guidance provided by GAT further enhances recognition accuracy, validating the effectiveness of the proposed method.