Classification of Leguminous Wood Species Based on Small Sample Hyperspectral Images

Abstract

Leguminous wood occupies an important position in the market of cultural and high-end wood. Accurate identification and classification of its species is crucial for the development of the industry. However, existing studies are still deficient in classification methods under small sample conditions. This paper uses hyperspectral image data and combines models such as support vector machine (SVM), random forest (RF), logistic regression (LR), and one-dimensional convolutional neural network (1-CNN). The synthetic minority oversampling technique (SMOTE) data enhancement technology was introduced to classify and recognize 18 common legume woods. After data processing, the classification accuracy of the traditional models was improved by about 5% on average, with the SVM model reaching 98.86%; the accuracy of the 1-CNN model was increased to 97.67% after adding the first-order derivative transform and Savitzky-Golay filtering, it reached 98.89% after further adding the SMOTE.