Spectral reflectance recovery using convolutional neural network

A procedure for spectral reflectance recovery from CIE tristimulus values is proposed using the convolutional neural network method. Unlike the common spectral recovery methods in a linear way, the nonlinear transformation from the CIE tristimulus values to spectral reflectance is to achieve in this paper. In consideration of the computation time and accuracy of spectral recovery, the internal parameters of convolutional neural network are adjusted by the number of neurons and the interval between neurons. The effectiveness of the proposed method and the previous methods are analyzed by calculating the spectral recovery accuracy under different spectral datasets and different error metrics. The results show that the proposed method is superior to traditional algorithms.