Georgia Southern University
Machine Learning for Automated Classification of Tree Species Based on Color-Texture Fusion
Abstract
dc:description.abstractThis study aimed to develop robust machine learning tools for the classification of southeastern United States tree species based on bark images, addressing challenges posed by significant intraspecies variation in bark texture and color. The primary objectives included constructing specialized machine learning libraries and classification models, identifying optimal algorithms and parameters to maximize accuracy, and integrating the best-performing texture and color classifiers into a unified system to enhance the overall classification performance. The methodology involved collecting a dataset of 360 field images, extracting hue and saturation histograms for color features, and employing Histogram of Gradients (HOG) and Histogram of Binary Patterns (HBP) for texture features. These features were classified using Nearest Neighbor and Support Vector Machine (SVM) algorithms, with extensive parameter sensitivity analysis across multiple pipelines to optimize performance. Results demonstrated that the highest accuracy for color classification (75.4%) was achieved with an SVM utilizing a 1 x 1 grid and 256 histogram bins, while texture classification reached 63.7% accuracy with an SVM configured with similar grid and 180 bins for HOG and HBP features. A late fusion approach combining the top color and texture classifiers enhanced the overall accuracy to 84.4%. This work highlights the effectiveness of machine learning for tree species identification in challenging intraspecies variation contexts.
Degree
thesis:*- Name thesis:degree_name
- Master of Science in Applied Engineering (M.S.A.E.)
- Level thesis:degree_level
- Thesis (open access)
- Discipline thesis:degree_discipline
- Department of Manufacturing Engineering
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Jones, Brandon C
- Contributors dc:contributor
-
- Hossein Taheri
- Doyun Lee
Subjects
dc:subject × 1Rights
- Licence dc:rights.license
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://digitalcommons.georgiasouthern.edu/etd/3039