Overcoming Overfitting in CNN Models for Potato Disease Classification Using Data Augmentation

Classification of diseases in potato plants is crucial for agriculture to ensure quality and yield. Potatoes, being staple foods worldwide, are vulnerable to diseases that cause significant production losses. Early and accurate disease identification is essential. This study evaluates the impact of data augmentation on reducing overfitting in deep learning models for potato disease classification. Various CNN architectures, including VGG16, VGG19, Xception, and InceptionV3, were compared in transfer learning and fine-tuning phases. The "Potato Disease Dataset", consisting of 451 images across seven classes, was used. The dataset was split into training, validation, and test sets, and augmentation increased the training set from 360 to 2160 images. The results indicate that models trained with augmented data exhibited improved performance in terms of accuracy, precision, recall, and F1-scores compared to those trained without augmentation. The learning curves show that data augmentation helps in reducing overfitting and enhancing model stability. Data augmentation is crucial for developing robust deep learning models for potato disease classification. Future work will explore advanced augmentation techniques and other architectures to enhance model performance.