Distributed optical fiber traffic signal data enhancement method based on physical constraint and gradient domain fusion

Abstract

A distributed optical fiber traffic signal data enhancement method based on physical constraint and gradient domain fusion belongs to the technical field of intelligent traffic and optical fiber sensing data processing. The present application aims at the problem of domain deviation caused by inconsistent background noise distribution between heterogeneous acquisition devices which is not considered in existing data enhancement methods. It includes: obtaining a standardized two-dimensional space-time waterfall graph matrix from the original space-time vibration signal; extracting the effective vehicle vibration signal to obtain independent foreground instances, performing data enhancement to obtain enhanced foreground instances; then collecting pure background noise signals; using the Poisson image editing operator, taking the pure background noise signal as the Dirichlet boundary condition, under the premise of keeping the local gradient field of the enhanced foreground instance unchanged, solving the Poisson equation to fuse the enhanced foreground instance into the corresponding pure background noise signal, and then performing label mapping to obtain a synthetic optical fiber traffic signal with labeled data. The present application realizes the zero-cost expansion of high-quality training samples.

Images