Machine-learning assisted prediction of peak wavelengths in four-primary light-emitting diode displays with arbitrary Rec. 2020

Abstract

The color gamut is an important quality evaluation metric for LED (Light-Emitting Diode, LED) displays. Based on GA (Genetic Algorithm, GA) and LM-BP (Levenberg-Marquardt Back-Propagation, LM-BP) neural network model, this paper proposes an effective method for the precise prediction of peak wavelengths in four-primary RYGB-LED displays (Red-R, Yellow-Y, Green-G, and Blue-B LEDs or QDs (Quantum Dots, QDs) with arbitrary values of color gamut. The GA is applied to optimize the CIE (Commission Internationale de l’Eclairage, CIE) chromaticity coordinates in consideration of specific color gamut thresholds, and the LM-BP neural network model is applied to predict the peak wavelengths of four-primary RYGB-LED displays based on the optimal CIE chromaticity coordinates, which enables the spectral design from color gamut to peak wavelengths of LEDs. The spectral power distribution of four-primary RYGB-LED displays is simulated using double-Gaussian functions. A total of 2000 spectral datasets for each color are divided into training sets, validation sets, and test sets in a ratio of 8:1:1. The K-fold cross-validation is used to determine the hyperparameters in LM-BP neural network model, and the optimal model is tested in the test sets, recording mean square error to assess the overfitting. Results indicate that the LM-BP neural network model exhibits excellent performances and generalization capabilities, and this model is applied to CIE chromaticity coordinate sets to design peak wavelengths of LED displays from the color gamut, guiding the manufacturing of high-quality four-primary RYGB-LED displays consisting of LEDs and QDs.