Deep Learning Framework Enhances Spectral Prediction for MXene-Based Metasurfaces

A novel deep learning framework has been developed to enhance and expedite the prediction of electromagnetic spectra for MXene-based solar absorbers, a process that is typically resource-intensive and managed by full-wave solvers. Central to this study are MXenes, which are a group of two-dimensional transition metal carbides and nitrides. The new architecture incorporates transfer learning, multi-channel spectral refinement, and Savitzky-Golay smoothing techniques. It utilizes a fine-tuned version of the pretrained MobileNet model 2 to forecast 102-point absorption spectra from 64x64 metasurface designs. The multi-channel spectral refinement enhances feature extraction by processing feature maps through various convolutional channels, while Savitzky-Golay smoothing reduces high-frequency noise. Experimental results show that this model significantly surpasses baseline configurations. This research can be found on arXiv with the identifier 2602.08406, categorized as replace-cross, and may influence the design and optimization of solar absorbers and other photonic devices, potentially lowering computational expenses and facilitating quicker prototyping.