Early Detection of Potato Leaf Pest and Disease Using EfficientNet and ConvNeXt Architecture

Potato farming has problems in the form of diseases that often attack potato leaves. This disease can affect potato crop production and can even result in crop failure. Early detection is needed to help farmers decide to increase potato production quality. Disease detection models in potato leaf plants have been developed using the Convolutional Neural Network (CNN) algorithm. This research aims to develop a disease detection model on potato leaves using the EfficientNet and ConvNeXt methods and evaluate the effectiveness of these two models in classifying four types of potato plant leaf diseases and 1 type of healthy potato leaf. Augmentation techniques are used to overcome imbalanced data, and Transfer Learning (TL) methods such as EfficientNet and ConvNeXt architectures are used for classification. We evaluated the model resulting from the experimental results using performance measures of accuracy values. Based on our experiments, the final results show that the EfficientNet model for disease detection on potato leaves achieved a validation accuracy of 95.64% and a testing accuracy of 95.92%. Meanwhile, the ConvNeXt model's validation accuracy value was 95.44%, and the testing accuracy value was 94.29%.