Artificial neural network-assisted optical fiber sensor for accurately measuring salinity and temperature

The observation of seawater salinity and temperature is indispensable for sustainable development and utilization of marine resources. In this research, a simple and low-cost single-mode fiber (SMF) - no-core fiber (NCF) SMF structure based on Mach-Zehnder interference (MZI) is proposed to measure two variables with the assistance of an artificial neural network (ANN). In contrast to traditional wavelength linear fitting, direct matching of the entire spectrum to the variables through machine learning analysis effectively improves the accuracy of the output predictions. Specifically, mean absolute errors of salinity and temperature reduce from 1.540%o to 0.808%o and from 1.061 degrees C to 0.154 degrees C, respectively. The relaxation of the light source and the optical spectrum analyzer (OSA) requirements, the fluctuation of salinity or temperature, the disturbance of environmental parameters will all not weaken the demodulation performance of the new method. Furthermore, the accuracy of the predicted results on the new probes demonstrates the adaptability of the demodulation approach. Importantly, ANN can simultaneously demodulate two parameters from a single spectrum, avoiding the cross-sensitivity problem in traditional methods. The strategy is highly generalizable and promising to be extended to any other parameters measured by optical fiber sensors.