Evaluating CNN and SVM Models in Smart Agriculture: A Case Study on Bell Pepper Leaf Disease Classification

Abstract

<p>This research aims to develop a classification model for <b>bell pepper leaf disease images to support modern agriculture</b> in <b>Smart Green House (SGH)</b>. Diseases on bell pepper leaves caused by pests, fungi, and viruses have caused a decrease in crop yields. To solve these issues, an image classification method called <em>Convolutional Neural Network</em> (CNN) and <em>Support Vector Machine</em> (SVM) is used. The datasets are grouped into four classes: healthy (healthy leaf), pest (leaf disease caused by pest), fungus (leaf disease caused by fungus), and virus (leaf disease caused by virus). The dataset used in this study consists of 173 images, divided into 139 training data (80%) and 34 testing data (20%). The training data were then combined using traditional augmentation methods, increasing the total number of images to 1,390. To obtain robust results, early stopping trials with a patience of 5, followed by 50 epoch and 100 epoch approaches were conducted on the CNN model. On the other hand, SVM using manually extracted features from color (RGB, HSL) and texture (Sobel) components was tested with scenarios such as the use of linear, polynomial, and RBF kernels. The <strong>best evaluation results were delivered by CNN with early stopping with 97% accuracy</strong>. However, SVM also performed quite well with 94% accuracy on the polynomial kernel. These results indicate that CNN excels in classifying leaf diseases, and the results of this study are expected to help develop <strong>a system for detecting diseases in bell pepper plants</strong>.<br></p>