A Novel Image-Based Disease Detection Model for Potato Leaves Using CNN

Potatoes are a vital agricultural commodity world wide, with a production volume nearing 356.95 million metric tons and a global market worth approximately 92.7 billion annually. Their nutritional and economic significance necessitates reliable cultivation practices with minimal disease interference. Potato crops are especially vulnerable to severe diseases such as early and late blight, which can significantly l ower yields, increase input costs, and place added demands on labor. Traditional disease detection methods, whichrelymainlyonvisualassessment, tend to be time-consuming, unreliable, and ineffective in identifying infections at an early stage. Consequently, there is growing adoption of automated image based approaches. Among these, Convolutional Neural Networks (CNNs) have demonstrated strong capabilities in classifying and identifying diseased potato foliage. Advances in image classification, combined with the availability of robust datasets, have improved model reliability and accuracy in disease detection. Modern architectures like VGG-19, DenseNet121, ResNet50, along with efficient custom CNNs, are frequently employed for agricultural deployments. In this study, the pro posed model was trained on the open-access PlantVillage dataset, comprising 2,152 labeled images. To enhance data diversity and address class imbalance, augmentation techniques were applied, which significantly boosted classification accuracy. T he finnalized mo del at tained a test accuracy of 98.98%, demonstrating strong performance and applicability in real-world disease detection for potato crops.