Full-colour imaging behind a glass diffuser through a diffractive neural network with multi-wavelength channels

It is difficult for general imaging systems to see an object hidden by scattering media such as fog and biotissues. A diffractive neural network (DNN) has been demonstrated to reconstruct an object behind an unknown scattering medium at the speed of light through repeated light modulation. However, the architectures of most conventional DNNs have been designed for specific wavelengths. Such architectures are insufficient for the reconstruction of an object behind a scattering medium across multiple wavelengths. Here, we introduce a wavelength-multiplexed DNN composed of volume holographic optical elements for the multi-spectral imaging of an object behind a scattering medium. Our DNN, which was completely independent and parallel for multiple wavelengths, reconstructs spectral images of a hidden object with performance superior to that of a conventional DNN. Our proposed DNN can be expanded to broader and more numerous wavelengths creating opportunities for various applications including machine vision in autonomous systems, underwater imaging and bioimaging.