BEMD-Based Color–Texture Feature Fusion with Image Sharpening for Multiclass Huanglongbing Severity Classification
Abstract
Huanglongbing (HLB), or citrus greening disease, is one of the most destructive citrus diseases worldwide. Its symptoms often resemble nutrient deficiencies, making severity assessment difficult and subjective. This study proposes a color-texture feature fusion method to classify citrus leaf images into four HLB severity classes: healthy, mild, moderate, and severe. A dataset of 1,000 citrus leaf images from Sumanto et al. (2023), comprising 250 images per class, was used. HLB infection was validated using PCR/qPCR analysis, while severity annotations followed the original labeling protocol. Images were sharpened using a standard deviation of 0.1 and decomposed using Bidimensional Empirical Mode Decomposition (BEMD) into IMF1, IMF2, IMF3, and residue components to capture multiscale texture information. Color and texture features were fused and classified using Support Vector Machine (SVM), k-Nearest Neighbor (KNN), Random Forest (RF), and Neural Network (NN). Performance was evaluated using stratified k-fold cross-validation with out-of-fold predictions. BEMD with Color Fusion achieved the best SVM performance, with accuracy, precision, recall, and F1-score of 0.93 and computation time of 1.24 seconds. It outperformed Basic BEMD and BEMD Fusion Baseline, with accuracies of 0.77 and 0.89. These results demonstrate improved multiclass HLB severity classification, although independent field validation is required before practical deployment.
Related
- Information coding strategies in biological reservoir computing for multiclass time-series classification
- Texture-aware diffusion modelling for severity-conditioned synthetic progression-like visualization of diabetic foot ulcers from single images
- A Computer Vision-Based Framework for Multilevel Egg Fertility Classification Using L*a*b* Color Segmentation, Wavelet Texture Features, and Support Vector Machine
- Hyperparameter-Tuned K-Nearest Neighbor for Herbal Leaf Classification Using Shape, Color, and Texture Feature Fusion in Smart Agriculture
- SeverYX: An ensemble framework for robust variant severity classification
BibTeX
@article{purnama2026bemdbased,
title = {{BEMD-Based Color–Texture Feature Fusion with Image Sharpening for Multiclass Huanglongbing Severity Classification}},
author = {Rachmat Adi Purnama and Dwiza Riana and Mochamad Wahyudi and Sumanto},
journal = {Journal of Information Systems and Informatics},
year = {2026},
month = oct,
doi = {10.63158/journalisi.v8i5.1867},
url = {https://doi.org/10.63158/journalisi.v8i5.1867},
note = {CC BY},
}
RIS
TY - JOUR
TI - BEMD-Based Color–Texture Feature Fusion with Image Sharpening for Multiclass Huanglongbing Severity Classification
AU - Rachmat Adi Purnama
AU - Dwiza Riana
AU - Mochamad Wahyudi
AU - Sumanto
PY - 2026
DA - 2026/10/24
JO - Journal of Information Systems and Informatics
DO - 10.63158/journalisi.v8i5.1867
UR - https://doi.org/10.63158/journalisi.v8i5.1867
AB - Huanglongbing (HLB), or citrus greening disease, is one of the most destructive citrus diseases worldwide. Its symptoms often resemble nutrient deficiencies, making severity assessment difficult and subjective. This study proposes a color-texture feature fusion method to classify citrus leaf images into four HLB severity classes: healthy, mild, moderate, and severe. A dataset of 1,000 citrus leaf images from Sumanto et al. (2023), comprising 250 images per class, was used. HLB infection was validated using PCR/qPCR analysis, while severity annotations followed the original labeling protocol. Images were sharpened using a standard deviation of 0.1 and decomposed using Bidimensional Empirical Mode Decomposition (BEMD) into IMF1, IMF2, IMF3, and residue components to capture multiscale texture information. Color and texture features were fused and classified using Support Vector Machine (SVM), k-Nearest Neighbor (KNN), Random Forest (RF), and Neural Network (NN). Performance was evaluated using stratified k-fold cross-validation with out-of-fold predictions. BEMD with Color Fusion achieved the best SVM performance, with accuracy, precision, recall, and F1-score of 0.93 and computation time of 1.24 seconds. It outperformed Basic BEMD and BEMD Fusion Baseline, with accuracies of 0.77 and 0.89. These results demonstrate improved multiclass HLB severity classification, although independent field validation is required before practical deployment.
SN - 2656-5935
SN - 2656-4882
ER -