Banana Leaf Disease Classification Using HSV and LBP Feature Extraction with Support Vector Machine

Abstract

Banana leaf diseases are one of the primary factors contributing to the decline in the quality and productivity of banana plants, as manual identification remains time-consuming, subjective, and prone to misclassification due to the similarity of symptoms among different diseases. This raises the question of whether a combination of Hue, Saturation, Value (HSV) and Local Binary Pattern (LBP) feature extraction can provide an effective alternative for banana leaf disease classification using a Support Vector Machine (SVM). While previous studies have relied on deep learning methods or combined HSV with Histogram of Oriented Gradients (HOG) features for this task, the combination of HSV and LBP for banana leaf disease classification remains largely unexplored. Using the Banana Leaf Spot Diseases (BananaLSD) dataset, comprising four classes (Cordana, Healthy, Pestalotiopsis, and Sigatoka), the original images were first divided into training and testing sets using an 80:20 ratio to prevent data leakage, after which data augmentation was applied exclusively to the training set. All images were center cropped before HSV and LBP features were extracted, combined, and classified using an SVM with a Radial Basis Function (RBF) kernel. On an independent test set of original, non-augmented images, the proposed model achieved an accuracy of 87.30%, precision of 89.06%, recall of 87.30%, and F1-score of 87.80%, with consistent results confirmed through Stratified Group 5-Fold Cross-Validation and an ablation study showing that the HSV and LBP combination outperformed either feature type alone. These findings indicate that combining HSV and LBP features offers a reliable, feature based alternative to deep learning for automated banana leaf disease identification.