Reconstructing computational spectra using deep learning's self-attention method

Miniaturized computational spectrometers have become a new research hotspot due to their portability and miniaturization. However, there are several issues, like low precision and poor stability. Because the problem of spectrum reconstruction accuracy is very evident, we suggested a novel approach to raise the reconstruction accuracy. A library of optical filtering functions was acquired using the time-domain finite-difference (FDTD) method. A cross-correlation algorithm was then used to choose 100 sparse filter functions, which were then built as an encoding matrix and then, based on the encoding matrix, a self-attention mechanism algorithm to improve the accuracy. The reconstructed spectrum's mean square error (MSE) is 0.0019, and its similarity coefficient (R2) is 0.9780. This self-attention mechanism spectral reconstruction technique will open up new possibilities for high-accuracy reconstruction for various computational spectrometer types.