Semiautomatic acousto-optical tunable filter calibration from spectrometry in the visible range with deep learning

Abstract

Acousto-optical filters are devices in which light dispersion occurs through a crystalline translucent material. Particularly, light interacts with a sound-induced spatially distributed. Post-interaction, diffracted light can be analyzed for different purposes. Although acousto-optics has been studied for decades, practical devices applying its principles are relatively recent. Here, experimental and technical procedures are used to obtain the transfer function of an acousto-optical tunable filter (AOTF) based system used as a hyper-spectral photometer. The reflectance responses at given wavelengths are measured and adjusted from a commercially available color pattern set, while typically, those values are set up manually. We propose a semiautomatic strategy to calibrate as a single black box all components of the system including: the light source, the signal generator power with its frequency-amplitude deviation from the full radio frequency set point, the radio-frequency amplifier, the transmission lines, the piezoelectric impedance, and the filter's own transfer function among others. To achieve that, we explored the capability of neural networks with deep learning. The system's input is reflectance data measured with a spectrophotometer at wavelengths from 400 to 700 nm with a step of 10 nm. Then, the AOTF system was used to gather reflectance data from those color pattern tiles from 400 to 700 nm with a step of 1 nm. Both reflectance datasets were adjusted using the proposed deep learning neural network. Results show that it is possible to calibrate an AOTF system by using ceramic tile color patterns and measuring reference reflectance values with a spectrophotometer in the visible range. Furthermore, a neural network can be trained to learn the compensation values, deriving trustable spectral information with a better wavelength resolution.