Lightweight-based Intrusion Event Recognition Method for Distributed Fiber Optic Vibration Sensing

Abstract

When Distributed Fiber Optic Vibration Sensing (DVS) systems are used for perimeter security intrusion detection in complex industrial environments, existing recognition models suffer from insufficient adaptability, high computational overhead, and high latency. This paper proposes the MNV4-LDA end-to-end recognition method, which integrates the lightweight feature extraction capability of MobileNet-V4 (MNV4) with the advantages of supervised dimensionality reduction of Linear Discriminant Analysis (LDA), constructing a framework from feature enhancement to dimensionality reduction and then to classification. The experiment targets three types of events: background noise, mechanical operations, and human walking. Tests based on 6000 groups of datasets show that the number of parameters of this method is reduced from 632.22M to 2.38M, with a reduction rate exceeding 99.6%. The recognition accuracy reaches 98.28%, 4.67% higher than that of MNV4, and the single-sample inference time on the CPU is 127.0ms. Moreover, no denoising preprocessing is required. This method effectively solves the edge deployment challenges of traditional models, balances efficiency and stability, and provides a practical solution for the edge application of DVS systems in perimeter security.