High-Resolution Spectrum Reconstruction of Cascaded FBG-FPI Sensor in Resource-Constrained Environments

Monitoring external physical variables with fiber optic sensors requires highly efficient demodulation instruments, particularly for strategies based on spectrum analysis. However, balancing performance with cost is a significant challenge in resource-limited settings. To address these challenges, this article introduces a new framework that uses a simple yet effective multimodal fusion strategy based on deep learning. It achieves high-resolution and precise demodulation by reconstructing the original spectrum from a limited number of ultralow resolution samples. Specifically, these ultralow resolution spectra samples are transformed into images. A pretrained model, which contains extensive prior knowledge, encodes them as potential multilayer spectral features. Additionally, spectral drift is captured as projected intensity information, taking advantage of the wave decomposition multiplexing feature of arrayed waveguide gratings (AWGs). A simple neural network then fuses and decodes the multilayer spectral features and projected intensity information to produce high-resolution sensor spectra. We demonstrate the effectiveness of our approach with a cascade structure sensor that includes a fiber Bragg grating-Fabry–Perot interferometer (FBG-FPI) for strain monitoring. The results confirm that our framework can efficiently reconstruct spectra within a 75 nm wavelength range, achieving high resolution and precision.