Research Progress in Task-Driven Intelligent Data Processing for Distributed Fiber-Optic Sensing
TL;DR
This review aims to reveal, from a task-driven perspective, the intrinsic connections among DFOS architectures, data representations, and intelligent methods, thereby providing references for related research and applications.
Abstract
Distributed fiber optic sensing (DFOS) enables the continuous acquisition of physical information such as vibration, temperature, and strain along an optical fiber. This technology offers significant value in fields including infrastructure monitoring, energy security, transportation sensing, and geophysical exploration. As measurement range, spatiotemporal resolution, and sampling rates continue to improve, DFOS data increasingly exhibits high throughput, strong spatiotemporal coupling, complex noise, and high scene dependency. Consequently, data processing and information extraction have become critical bottlenecks affecting system performance and practical deployment. This article reviews the progress of intelligent data processing in DFOS from a task-driven perspective. The discussion begins with an overview of the primary sensing architectures and summarizes typical data formats along with features relevant to intelligent modeling. Focusing on three core tasks which include signal enhancement and preprocessing, event detection and pattern recognition, and localization and parameter estimation, the work systematically categorizes the research progress of traditional signal processing and intelligent methods such as machine learning and deep learning. The analysis further evaluates the suitability of these methods across different data forms and application scenarios. This review aims to reveal, from a task-driven perspective, the intrinsic connections among DFOS architectures, data representations, and intelligent methods, thereby providing references for related research and applications.