Multi-point time series classification is a critical task in phase-sensitive optical time-domain reflectometry (Φ-OTDR) for vibration signal processing, with broad applications in security monitoring. While most existing classifiers rely on convolutional operations, they struggle to capture global dependencies in long spatiotemporal sequences, limiting their effectiveness. To address this, we propose STFM (Spatio-Temporal Feature Fusion Mamba), a novel hybrid architecture based on the Mamba framework, which significantly enhances spatiotemporal feature extraction. The Temporal Feature Fusion (TFF) module employs a cross-time gating mechanism to selectively amplify target variations while suppressing irrelevant noise, improving temporal modeling efficiency. Additionally, separable convolutions are introduced to reduce computational overhead. Meanwhile, the Spatial Feature Fusion (SFF) module utilizes cross-scale attention to capture fine-grained spatial details. Experimental results demonstrate that STFM effectively extracts long-range spatiotemporal features, outperforming existing methods in both accuracy and efficiency. This work provides a solution for Φ-OTDR-based classification tasks.