Classification of Fir Seeds Based on Feature Selection and Near-infrared Spectroscopy

Fir trees account for about a quarter of the country's building materials. Fir trees grow quickly, and have certain economic, social, ecological and other benefits. Different fir trees require different growth environments, so it is necessary to accurately identify the fir seeds to cultivate them more specifically, so as to maximize the various benefits of the fir trees. In this paper, six kinds of fir seeds, namely Abies fabri, Cryptomeria japonica, Metasequoia glyptostroboides, Cathaya argyrophylla, Keteleeria fortune, Picea asperata, were used as experimental samples. The spectral data of six fir seeds were measured by Fourier transform near-infrared spectrometer. To reduce the high dimensions of spectra data, four different feature selection methods (InfoGain, Symmetric Uncert, Gainratio and ReliefF) were used to select effective feature subset for random forest classification. Experiments compared performances of the original data and the selected features cascade data with standard normal variate (SNV) method to preprocessing. The results show that the accuracy rate of the selected feature subset through SNV processing reaches 98.90%, 1.1% higher than that of the original data and the rate of dimensionality reduction reaches 97.7%. The study provides an effective classification method for fir seeds of NIRS data.