Consequently, a neural network-based algorithm is suitable for detecting Bragg wavelength for CCD-based and tunable filter-based interrogators in real-time/time critical applications. In [8], Timur Agliullin et al. proposed a neural network-based algorithm for determining the Bragg wavelength of an FBG sensor for a CCD arraybased interrogator. In that work, the model is trained by the relation ’pixel amplitude (i.e., optical signal intensity captured on each pixel) of CCD sensor vs. absolute wavelength data of FBG spectrum.’ This training is done for some fixed wavelength range (3 nm in their work). However, this method has two significant disadvantages. First, there are cases where the FBG spectrum, due to some precondition (like prestain etc.) shifts significantly and goes out of the wavelength range window used for training (3 nm in the discussed case). The ANN model developed for the FBG sensor that determines the central wavelength from the pixel data can no longer be used in such a thing happens. Second, if the wavelength range window for training increases, the number of training samples will increase proportionately. Therefore, the training window cannot be increased arbitrarily; otherwise, the implementation becomes very expensive.
to our article [8] T. Agliullin, V. Anfinogentov, R. Misbakhov, O. Morozov, A. Nasybullin, A. Sakhabutdinov, B. Valeev, Application of Neural Network Algorithms for Central Wavelength Determination of Fiber Optic Sensors, Applied Sciences 13 (2023) 5338. https://doi.org/10.3390/app13095338.