DDSSnet: a fast strain demodulation approach for OFDR-based fiber shape reconstruction.

Abstract

In optical fiber shape sensing technology, enhancing sensing accuracy while simultaneously achieving real-time shape reconstruction presents a notable challenge. This work presents a fast strain demodulation algorithm for the optical frequency domain reflectometry (OFDR) shape sensing system. The fast strain demodulation algorithm comprises deviation calculation and deviation denoising for shape-sensing convolutional neural network (DDSSnet). The initial operating wavelengths of the shape sensor can be effectively calibrated and the phase noise of residual nonlinear tuning in the system can also be compensated. Compared with the cross-correlation algorithm, the fast strain demodulation algorithm has increased the processing speed of demodulating axial strain distribution by 9.691 times and a shape-sensing result by 9.4 times. The shape of one cylinder and one configuration were then reconstructed using the rotation-minimum frame, resulting in maximum relative errors of 0.581% and 1.170%, respectively, and average relative errors of 0.204% and 0.380%, respectively. These errors are all slightly smaller than those obtained using the cross-correlation algorithm. The results from the shape-sensing experiments indicate that this method enables both faster and more accurate shape reconstruction, offering promising potential for practical applications.