Texture Features and Statistical Features for Wood Types Classification System

Wood types research usually regards species of wood images. This research focuses on the classification of wood types in Indonesia, coming from four wood species, with one species divided into two types based on their grade and morphological characteristics. The dataset in this research was also created with variations in scale and position, i.e., horizontal and vertical. The feature extraction method is regarded by its types, known as texture features and statistical features. Texture features were extracted using Gray Level Co-occurrence Matrix (GLCM), and statistical features were obtained using the Histogram method. This research aims to analyze the relationship between classifiers and feature types as their input to classification performance for our unique dataset. Support Vector Machine (SVM) and Random Forest are two classifiers used in this research. Generally, four scenarios of the classification system are considered. The output of this study is an image dataset consisting of 5 types of wood in Indonesia and a wood species classification model. The best model was given by statistical features that were used as input of SVM, with an accuracy of 89%, Weighted Average Precision at 94%, and Weighted Average Recall at 89%. This result leads to an exciting point that statistical features give better classification results than texture features in the case of wood types classification, which contains intra-species. It also found that two statistical features, deviation and smoothness, can be assumed as features that have theoretical implications that make intra-species classification more difficult.