A hyperspectral imaging method based on deep learning to solve dispersion blur

Abstract

The invention discloses a hyperspectral imaging method for solving dispersion blur by deep learning. The steps of the method are: S1, collecting high-precision spectral data and dispersion RGB data; S2, preparing a convolutional neural network training data set; S3, constructing a convolutional neural network for unambiguous dispersion, which uses a three-dimensional convolution kernel, and Convolution kernels of multiple scales are juxtaposed; the input of the convolutional neural network is dispersion RGB data, and the output is reconstructed hyperspectral data; S4, using the convolutional neural network constructed by S3, trains the network through the back propagation algorithm to reconstruct from dispersion RGB data The ability to produce high-precision spectral data; S5, after multiple iterations, select the optimal model obtained through training to directly deblur the dispersion RGB data in the test set to obtain hyperspectral data. The present invention utilizes the deep learning convolutional neural network to solve the dispersion fuzziness, realizes hyperspectral reconstruction work, and greatly improves the spectral reconstruction speed under the premise of ensuring the reconstruction accuracy.

Description

technical field [0001] The invention belongs to the field of spectral imaging, in particular to a hyperspectral imaging method for solving dispersion blur by deep learning. Background technique [0002] Compared with the traditional RGB three-color imaging system, the spectral imaging system can obtain richer color dimensional information, and has important applications in military, medical, biological, and agricultural fields. Traditional spectral imaging methods have many limitations: scanning imaging uses different filters to scan the entire light band to obtain spectral images, which is related to the number of filters used, and imaging is slow; the spatial resolution of computed tomography is low; compressed imaging The coded spectral signal is encoded by the coded aperture, and its dispersion is compressed to reconstruct the spectrum, but the system design is complicated and the cost is high. [0003] In recent years, with the development of computer science, software...

AI Technical Summary

Problems solved by technology

However, the solution to dispersion is an underdetermined problem, and the relevant optimization algorithm needs a certain iteration time to reconstruct the spectrum, so it is difficult to achieve real-time reconstruction

In addition, to solve the underdetermined problem, it is necessary to independently design the feature prior, and the quality of the selected prior information will directly affect the accuracy of the reconstruction.

Images