Deep-Learning-Based Hyperspectral Imaging Through An RGB Camera

Hyperspectral image(HSI)contains both spatial pattern and spectral information which has been widely used in food safety,remote sensing,and medical detection.However,the acquisition of hyperspectral images is usually costly due to the complicated apparatus for the acquisition of optical spectrum.Recently,it has been reported that HSI can be reconstructed from single RGB image using convolution neural network(CNN)algorithms.Compared with the traditional hyperspectral cameras,the method based on CNN algorithms is simple,portable and low cost.In this study,we have achieved the acquisition of hyperspectral images in a single shot using an RGB camera based on a purely computational imaging approach.The main research elements are as follows.1.We compared and summarized advantages and drawbacks of domestic and international hyperspectral imaging technology roadmaps.In this study,the spectral response of the RGB camera was first accurately calibrated and a set of hyperspectral images was collected.We analyzed the possibility of hyperspectral reconstruction using principal component analysis according to the sparsity of the natural spectrum and the imaging principle of RGB cameras.2.Based on the sparsity of the natural spectral image,a super-resolution task of natural spectra was implemented using interpolation algorithms.The interpolation effects of several interpolation algorithms were compared and the best one was picked out.We also proposed the parallel transpose network according to the physical characteristics of the hyperspectral image.It has a creative transpose structure which allows the network to extract features in both spatial and spectral dimensions of the data cubes thoroughly.The ultimate goal of hyperspectral imaging using an RGB camera was achieved.3.We analyzed the impact of the camera spectral response on the hyperspectral reconstruction task.Thus,we confirmed that the spectral response of each RGB camera needs to be calibrated for computational hyperspectral imaging.In addition,we have investigated the effect of spectral response intensity characteristics on the quality of hyperspectral reconstructions.These conclusions inform the camera selection for hyperspectral reconstruction and helps to enhance the generalizability of this scheme.