NIR Inversion Model of Larch Wood Density at Different Moisture Contents Based on MVO-BPNN

Abstract

Owing to variation of moisture content, the larch wood basic density near-infrared (NIR) prediction model shows reduced accuracy and robustness, or even model failure. To solve this technical problem, a multi-verse-algorithm-optimized BP neural network (MVO-BPNN) prediction model is proposed to improve the accuracy of the model. The preprocessing effects of the Savitzky–Golay smoothing, detrending, and 15-point moving average smoothing method were compared. The synergy interval partial least squares was used to extract the feature bands of the NIR spectra. Results showed that the prediction model based on MVO-BPNN was better than those based on BPNN and the genetic algorithm-optimized BPNN. It indicated that the NIR model based on the MVO-BPNN could effectively predict the basic density of wood with different moisture contents.