Data-driven High-resolution Spectral Imaging And Highly Sensitive Transient Spectral Measurement Techniques

Spectral measurement can obtain the information of target spectral dimension and reveal more physical characteristics.However,there are still some limitations in the current spectral measurement technology.Therefore,in view of the limitations of Hadamard transform spectral measurement and imaging,this paper introduces neural network into the reconstruction of spectral data,researches from spectral imaging and transient spectral measurement respectively,and achieves good results.The main contents of this paper are as follows:(1)Spectral image super-resolution fusion based on convolutional neural network.Aiming at the contradiction between scanning speed and imaging resolution of Hadamard transform imaging spectrometer,this paper proposes a super-resolution fusion network of spectral image.The network consists of two modules.The image super-resolution reconstruction module is based on the encoder-decoder structure,and achieves high operation efficiency by introducing the non-bottleneck 1D structure and sub-pixel convolution layer;Based on the fusion mathematical model and multi-scales loss function,the spectral image fusion module learns the transformation matrix through the network.Experimental results show that compared with unsupervised traditional algorithms and other network-based supervision algorithms,the network proposed in this paper achieves high reconstruction quality and good generalization ability under the high operation efficiency.(2)Single optical path sub-Hadamard transform transient spectral measurement based on convolutional neural network.In order to solve the problems of duel optical path sub-Hadamard transform transient spectrometer such as low light intensity and poor reliability of the system,this paper presents a new method to reconstruct the light intensity distribution using convolutional neural network.By introducing the spectral dimension convolution kernel and the deconvolution process,and using the double loss function constraints of light intensity and reconstruction spectrum,the network can learn the inverse process of spectral dispersion,and reconstruct the intensity distribution from the spectral dispersion image,realize the single optical path acquisition.Simulation and experimental results show that the proposed scheme has higher spectral signal-to-noise ratio compared with traditional slit spectrometer and dual optical path sub-Hadamard transform transient spectrometer,and has simple optical path and high system reliability.