A high-quality imaging method of a spectral imaging system based on a convolutional neural network

Abstract

The invention discloses a high-quality imaging method of a spectral imaging system based on a convolutional neural network, and belongs to the field of computational photography. According to the invention, the hyperspectral image imaging process and the reconstruction process are considered together; in the reconstruction process, the spatial correlation and the spectral correlation between images are respectively considered, the residual error learning is used to accelerate the training speed and convergence speed of the network, the coding network is optimized while the network is reconstructed, and a GPU is used to complete the optimization solution of the whole network. A cuDNN library is used to accelerate the operation speed of the network. The method comprises updating network parameters by using a random gradient descent method; and performing block-by-block processing to complete reconstruction of the hyperspectral image. According to the method, the hyperspectral image reconstruction of the CASI spectral imaging system can be completed with high quality, the high spatial resolution and high spectral fidelity of a reconstruction result are ensured, meanwhile, the efficiency of hyperspectral image reconstruction is improved, and the application range of hyperspectral images is expanded. The method can be used in the fields of manned spaceflight, geological survey, agricultural production, biomedicine and the like.

Description

technical field [0001] The patent of the present invention relates to a high-quality hyperspectral image imaging method for spectral imaging, in particular to a method capable of quickly acquiring high-quality hyperspectral images, which belongs to the field of computational photography. Background technique [0002] Hyperspectral imaging technology is a combination of spatial imaging technology and spectral imaging technology, which can intensively collect the spectral signal of each point in the scene. The collected hyperspectral image is called a data cube, which contains a large amount of lighting and material information of the scene. This technology has been applied in many fields such as remote sensing, computer vision, and medical diagnosis. Currently existing two-dimensional imaging sensors cannot simply obtain three-dimensional hyperspectral images through one exposure. The traditional hyperspectral imaging technology chooses to sacrifice the time dimension and s...

AI Technical Summary

Problems solved by technology

[0009] Aiming at the problems existing in existing algorithms, such as not considering the imaging process and low quality of reconstructed images, etc.

Images