Multi-Model Comparative Study for Bark-Texture Based Tree Species Classification Using Custom Indian Tree Species Dataset

Abstract

Accurate wood species identification is crucial for biodiversity preservation and forest management. Because traditional identification methods are time-consuming and heavily rely on expert knowledge, automated image-based solutions have become more and more important. This research suggests a hierarchical framework for identifying wood species that makes use of both machine learning (ML) and deep learning (DL) approaches. The dataset of bark images utilized in the testing includes 22 distinct wood species. The ML-based approach evaluates classifiers such as Random Forest, Support Vector Machine, XGBoost, and ensemble models following extensive preprocessing and manually created feature extraction using statistical, color, and texture descriptors. The DL-based technique uses a specially designed CNN architecture in combination with convolutional neural networks that apply transfer learning models, such as MobileNetV2, DenseNet121, and ResNet50. The short dataset size is addressed using data augmentation and fine-tuning techniques. The experimental results demonstrate that Deep Learning models achieve greater classification performance and robustness when compared to Machine Learning models. This study provides a detailed comparison of complex DL algorithms with conventional ML to show the effectiveness of transfer learning for automated wood species identification.