Mass production-enabled computational spectrometers based on multilayer thin films.

Multilayer thin film (MTF) filter arrays for computational spectroscopy are fabricated using stencil lithography. The MTF filter array is a 6 × 6 square grid, and 169 identical arrays are fabricated on a single wafer. A computational spectrometer is formed by attaching the MTF filter array on a complementary metal-oxide-semiconductor (CMOS) image sensor. With a single exposure, 36 unique intensities of incident light are collected. The spectrum of the incident light is recovered using collected intensities and numerical optimization techniques. Varied light sources in the wavelength range of 500 to 849 nm are recovered with a spacing of 1 nm. The reconstructed spectra are a good match with the reference spectra, measured by a grating-based spectrometer. We also demonstrate computational pinhole spectral imaging using the MTF filter array. Adapting a spectral scanning method, we collect 36 monochromatic filtered images and reconstructed 350 monochromatic images in the wavelength range of 500 to 849 nm, with a spacing of 1 nm. These computational spectrometers could be useful for various applications that require compact size, high resolution, and wide working range.

Compressive Sensing Spectroscopy Using a Residual Convolutional Neural Network.

Compressive sensing (CS) spectroscopy is well known for developing a compact spectrometer which consists of two parts: compressively measuring an input spectrum and recovering the spectrum using reconstruction techniques. Our goal here is to propose a novel residual convolutional neural network (ResCNN) for reconstructing the spectrum from the compressed measurements. The proposed ResCNN comprises learnable layers and a residual connection between the input and the output of these learnable layers. The ResCNN is trained using both synthetic and measured spectral datasets. The results demonstrate that ResCNN shows better spectral recovery performance in terms of average root mean squared errors (RMSEs) and peak signal to noise ratios (PSNRs) than existing approaches such as the sparse recovery methods and the spectral recovery using CNN. Unlike sparse recovery methods, ResCNN does not require a priori knowledge of a sparsifying basis nor prior information on the spectral features of the dataset. Moreover, ResCNN produces stable reconstructions under noisy conditions. Finally, ResCNN is converged faster than CNN.