针对相位敏感光时域反射系统中传统人工特征提取和模式识别方法实时性差、准确率低的问题，提出一种波网络模式识别方法。该方法通过因果空洞卷积结构充分分析光纤振动信号的时序因果性，通过残差块结构使模型更快收敛，以实现更高的识别准确率和效率。实验结果表明，在对手拍、脚踩、棒击三种信号识别时，与一维卷积神经网络结构和长短期记忆网络结构方法相比，该方法识别准确率高达99.85%；且训练耗时最少，低至96 s，测试耗时也仅为30 ms，满足应用实时性的要求。该模式识别方法既具有高准确率又具有高实时性，对于φ-OTDR系统在周界安防中的应用推广具有重要意义。

An all-optical light–control–light functionality with the structure of a microfiber knot resonator (MKR) coated with tin disulfide (SnS2) nanosheets is experimentally demonstrated. The evanescent light in the MKR [with a resonance Q of ～59,000 and an extinction ratio (ER) of ～26dB] is exploited to enhance light–matter interaction by coating a two-dimensional material SnS2 nanosheet onto it. Thanks to the enhanced light–matter interaction and the strong absorption property of SnS2, the transmitted optical power can be tuned quasi-linearly with an external violet pump light power, where a transmitted optical power variation rate ΔT with respect to the violet light power of ～0.22dB/mW is obtained. In addition, the MKR structure possessing multiple resonances enables a direct experimental demonstration of the relationship between resonance properties (such as Q and ER), and the obtained ΔT variation rate with respect to the violet light power. It verifies experimentally that a higher resonance Q and a larger ER can lead to a higher ΔT variation rate. In terms of the operating speed, this device runs as fast as ～3.2ms. This kind of all-optical light–control–light functional structure may find applications in future all-optical circuitry, handheld fiber sensors, etc.