Open-Set Recognition Model for Distributed Fiber-Optic Vibration Sensor Based on Hierarchical Attention Network With Openmax

Recent advancements in deep learning have improved the signal recognition capabilities of distributed fiber-optic vibration sensors. However, most existing research focuses on closed-set scenarios, assuming that all potential events are known and accounted for during training. In practical applications, the environmental complexity surrounding sensing fibers makes it impractical to anticipate and train on every intrusion event. To address this, we propose a novel hierarchical attention network with Openmax (HANO) model that integrates a hierarchical attention mechanism with the Openmax algorithm. This model effectively extracts and emphasizes discriminative features at multiple levels, enhancing the recognition of both known and unknown events. Incorporating the Openmax layer enhances the classifier’s ability to manage unknown classes, thereby facilitating open-set recognition. To validate the effectiveness of the HANO model, we conducted experiments involving human intrusion events within a fenced area. The network parameters were optimized accordingly, and HANO’s performance was compared against the Swin transformer, as well as models utilizing softmax and sigmoid layers. Additionally, we benchmarked it against other open-set recognition algorithms. The experimental results demonstrate that the HANO model achieves an overall recognition accuracy exceeding 95%, outperforming the comparative methods. These results demonstrate that the HANO model enhances the recognition of open-set events, indicating its strong potential for practical applications, particularly in scenarios requiring precise intrusion detection, such as security monitoring and vibration detection.