Deep learning method for optical fiber curvature measurements based on time series data

Abstract

Curvature detection is an essential technique for monitoring landslides, which are frequent and destructive disasters. Existing methods for curvature detection using fiber-optic sensors have limitations such as complex fabrication or large data size. We propose a data processing method for high-accuracy curvature detection that employs deep learning. We experimented using different levels of curvature and compared our method with other methods. Our method achieves 99.82% accuracy for classification and root mean square error of 0.042m−1 for regression with a simpler structure and smaller data size. Our approach demonstrates its potential for landslide detection and integration with communication systems.