Intan Karunia Septiani Septiani, Wika Dianita Utami, Nurissaidah Ulinnuha, Dino Ramadhan

Classification of Wood Types Based on Wood Fiber Texture Using GLCM - ANN

Introduction

Classification of wood types based on wood fiber texture using glcm - ann . Classify Indonesian wood types (Teak, Sengon, Mahogany, Gmelina) based on fiber texture using GLCM feature extraction & Artificial Neural Networks (ANN). Explores image processing & accuracy.

Abstract

In Indonesia, various types of wood grow and develop with various characteristics and benefits. Each type of wood has differences in texture and fiber, to classify it must have sufficient knowledge about the texture and fiber of wood. A wood species identification system is needed to help the classification process. The purpose of this research is to classify Teak Wood, Sengon Wood, Mahogany Wood, and Gmelina Wood which are often sold in Indonesia. The classification method used in this research is Artificial Neural Network with Gray Level Co- occurrence Matrix (GLCM) extraction. Pre-processing stages include Histogram Equalization, filtering, converting images into grayscale form, and data augmentation. Feature extraction of pre-processing results using GLCM is taken, namely contrast, correlation, energy, homogeneity, and entropy. From the research results, classification using Artificial Neural Network was obtained with 46% accuracy, 43% precision, 42.5% recall, and 42% F1-Score with a GLCM inclination angle of 90°. So, this method can be used to classify the types of wood, but it is less accurate because there are still deficiencies in the model.

Review

This paper addresses the pertinent challenge of classifying different wood types based on their fiber texture, a task crucial for various applications, particularly in resource-rich regions like Indonesia. The authors aim to differentiate between four common Indonesian wood species: Teak, Sengon, Mahogany, and Gmelina. The chosen approach integrates Gray Level Co-occurrence Matrix (GLCM) for feature extraction from wood images, followed by an TArtificial Neural Network (ANN) for classification. This combination presents a standard and generally robust framework for texture-based image classification, making the research direction relevant and potentially impactful for automating wood identification processes. The methodological pipeline involves several pre-processing steps, including Histogram Equalization, filtering, grayscale conversion, and data augmentation, which are appropriate for preparing image data. Subsequently, GLCM is employed to extract five widely recognized textural features: contrast, correlation, energy, homogeneity, and entropy, using a 90° inclination angle. While these features are fundamental for characterizing texture, the reported classification performance is critically low. An accuracy of 46%, precision of 43%, recall of 42.5%, and an F1-Score of 42% suggest that the model's ability to reliably distinguish between the four wood types is only marginally better than random guessing (which would be 25% for four classes). This indicates a significant deficiency either in the discriminatory power of the extracted GLCM features for these specific wood types, the capacity or architecture of the ANN, or potentially issues with the dataset's quality or size, despite data augmentation. In conclusion, while the research introduces a relevant problem and a conventional methodology for wood classification, the obtained results are largely unsatisfactory. The statement that "this method can be used to classify the types of wood, but it is less accurate" needs to be critically re-evaluated given the low performance metrics, which fall short of practical utility. Future work should focus on several key areas to significantly improve the model's efficacy. This could include exploring a broader range of GLCM angles, investigating other texture descriptors (e.g., LBP, Gabor filters), incorporating shape-based or structural features, experimenting with more complex ANN architectures, or utilizing deep learning models that can learn features automatically. Furthermore, a detailed analysis of the misclassified samples, an expanded and more diverse dataset, and rigorous hyperparameter tuning of the ANN are essential to develop a truly robust and accurate wood classification system.

Full Text

Comments

You need to be logged in to post a comment.