Self-Learning Filtering Method Based on Classification Error in Distributed Fiber Optic System

Abstract

The distributed fiber optic system based on phase-sensitive optical-time-domain reflectometry ( $\Phi $ -OTDR) has attracted extensive interest in the field of perimeter security. However, the noise caused by environmental factors in fiber optic signal processing is a great difficulty due to the complexity and variability of the natural environment. In this paper, an improved wavelet de-noising method based on classification error of training signal data is proposed to solve the difficulty of filtering thresholds selection. By maximizing the true positive rates under selected false positive rates, initial filtering thresholds based on experience are constantly adjusted and optimized. Experimental results prove that improved wavelet filtering thresholds get better performance than traditional wavelet thresholds in indoor and outdoor environments. In addition, the improved method can automatically select and optimize thresholds to adapt to different situations.