Neural network for classification of Chinese zither panel wood via near-infrared spectroscopy

Abstract

The wood grade used for Chinese zither panels is primarily determined through an artificial experience method, and the number of related practitioners is decreasing annually. In this study, a method using an improved BP neural network is proposed to assess the wood grade for Chinese zither panels. Abnormal spectral samples were first removed based on the Mahalanobis distance method. Normalization and Savitzky Golay second derivatization were applied to the remaining data set. According to the spectral peak, the spectral data were divided into three bands, which were applied to the model proposed in this paper, and the most critical spectral region for judging the wood grade of Chinese zither panels was obtained. Through principal component analysis, the appropriate feature variables were selected and applied to the experimental model for an analysis to reduce the calculated amount in the experiment. When the number of principal components was 6, the classification accuracy of unknown samples was 96.7%. Compared with the PLS model, the proposed model is more robust and accurate and has fewer losses. The experimental results indicated that the proposed method effectively identifies the wood grade used in Chinese zither panels.