Identification of Multiple Events Based on Two-Dimensional Images and Isotropic Network in Optical Fiber Perimeter Security System

The identification of multiple events within perimeter security systems is crucial for bolstering safety measures and realizing cost savings. Despite the development of numerous data-driven strategies grounded in innovative intelligent methods, there remains a need for deeper exploration into intelligent identification methods leveraging original, acquired sensing data. In this article, an enhanced intelligent identification scheme, utilizing 2-D images and an isotropic network, is investigated to further enhance the performance of perimeter security system. First, a signal conversion method, leveraging a Constant-Q Transform (CQT), is employed to encode original time-series data signals into 2-D spectrum images. Notably, these images accurately capture specific time-frequency distribution features, thus outperforming traditional signal conversion methods. Second, an isotropic network, grounded in a ConvMixer architecture, is proposed for identifying sensing patterns using the captured 2-D images. In particular, this network can effectively balance input size and accuracy. Third, a distributed optical fiber perimeter security system, based on a dual Mach-Zehnder interferometer (DMZI), is constructed to collect original sensing data and verify the validity of the proposed intelligent identification scheme. The experimental results demonstrate that, in comparison with previously reported intelligent methods, the proposed scheme effectively identifies sensing patterns in complex operating environments, significantly enhancing the average recognition accuracy. Specifically, the intelligent scheme successfully classifies ten distinct types of sensing patterns, achieving a mean accuracy of 99.61%, and its mean processing time of 0.486 s sufficiently meets the real-time requirements of the sensing system in practical applications.