<![CDATA[Diseases affecting pepper leaves can significantly reduce crop productivity and quality, while manual disease identification remains subjective, time-consuming, and prone to error. Therefore, an accurate automated classification system is required to support early disease detection. This study aims to evaluate and compare the performance of a conventional Convolutional Neural Network (CNN) with two transfer learning–based architectures, VGG16 and EfficientNetB3, for classifying pepper leaf images into healthy and bacterial spot classes, as well as to analyze the impact of applying a soft voting ensemble method on classification performance. The dataset was obtained from Kaggle and divided into training, validation, and test sets. Image preprocessing included resizing all images to 224×224 pixels and applying data augmentation to improve model generalization. Model performance was evaluated using accuracy, precision, recall, and F1-score metrics. The experimental results indicate that EfficientNetB3 outperforms the conventional CNN and VGG16 models. Furthermore, the application of the soft voting ensemble enhances prediction stability, achieving an accuracy of 99.68% on the test dataset with balanced precision and recall across both classes. These findings demonstrate that the integration of transfer learning and soft voting ensemble methods is an effective approach for image-based pepper leaf disease classification under the experimental conditions, and provides a basis for further validation using more diverse datasets.]]>
Copyrights © 2026
