Spectral Demodulation of Tapered Microfiber Grating Using MobileNet

Tapered microfiber Bragg grating (<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mu $ </tex-math></inline-formula>FBG) has significant applications in biosensing and medical diagnostics, yet its complexed chirp spectrum still required manual demodulation. Herein, we present a spectral demodulation method for the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mu $ </tex-math></inline-formula>FBG based on the MobileNet convolutional neural network. Compared to five classical convolutional neural network models, MobileNet achieves 100% testing accuracy with a 20%–40% reduction in testing time. Due to its relatively simple network structure, the memory requirement of MobileNet is only 50% of the other models, while the storage size is one to two orders of magnitude lower. Moreover, in resource-constrained environments using only CPU, MobileNet’s time cost of training and testing is about 40%–60% of other models. The findings indicate that MobileNet is well suited for the spectral demodulation of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mu $ </tex-math></inline-formula>FBG, providing high accuracy, rapid demodulation, and low resource consumption. Its minimal memory requirements and compact storage size make it easily deployed in small embedded and mobile devices, demonstrating significant potential for miniature sensors based on real-time spectral demodulation and applications in biosensing.