Early Detection and Classification of Potato Leaf Disease Using Convolutional Neural Networks

Potato farming faces significant challenges due to the prevalence of leaf diseases, which lead to substantial economic losses and reduced yields. Traditional detection methods relying on expert observations and laboratory examination of leaves are impractical, particularly in most remote areas of Ethiopia. This study introduces a deep learning‐based solution for automating potato leaf disease detection and classification, addressing gaps in existing systems, such as limited datasets and narrow disease detection scopes. A custom convolutional neural network (CNN) model was developed and trained with the Plant Village dataset (16,000 images) and tested using a curated dataset (1000 manually collected images). The study classified potato leaf conditions into five categories: Potato Early Blight, Potato Late Blight, Potato Virus Diseases, Potato Insects, and Healthy Leaves. Techniques such as data augmentation and hyperparameter optimization were employed to enhance the model’s performance and reduce overfitting. Experimental results revealed that the proposed CNN model achieved superior accuracy (85%), precision (87%), recall (85%), and F1‐score (86%) compared with pretrained models (VGG19 and ResNet50) when tested with manually collected data. To demonstrate real‐world applicability, a prototype system integrating the trained CNN model was developed, offering farmers an efficient, low‐cost tool for early disease detection through a user‐friendly interface. This research bridges the gap in precision agriculture by providing a scalable and robust automated solution. Future directions include expanding the dataset, exploring alternative architectures, and enabling real‐time detection to support sustainable crop management.