Support Vector Machine and Multilayer Perceptron Optimization for Banana Leaf Disease Classification Using Hue Saturation Value and Gray Level Co-occurrence Matrix

Abstract

Banana leaf disease poses a significant threat to the quality and productivity of banana plants. Conventional disease identification relies heavily on expert knowledge and is time-consuming, highlighting the need for an automated and efficient solution. This study presents a classification system for banana leaf diseases by comparing the performance of Support Vector Machine (SVM) with a Radial Basis Function (RBF) kernel and Multilayer Perceptron (MLP). The dataset employed was the Banana Leaf Spot Diseases (BananaLSD) dataset, comprising four disease categories: healthy, cordana, pestalotiopsis, and sigatoka. Preprocessing steps included image resizing, data cleaning, feature extraction using Hue Saturation Value (HSV) color features and Gray Level Co-occurrence Matrix (GLCM) texture features, and data normalization via StandardScaler. Experimental results demonstrate that SVM achieved an accuracy of 94.34%, outperforming MLP which reached 93.40%. These findings confirm that the integration of HSV and GLCM features with SVM constitutes an effective approach for automated banana leaf disease classification, offering a promising foundation for intelligent plant health monitoring systems.