Performance Comparison of EfficientNetB0 in Potato Leaf Disease Classification with Adam and SGD

Abstract

Potatoes (Solanum tuberosum L.) are an important food commodity for global food security, but they are highly susceptible to leaf diseases that reduce yield and tuber quality. This study aims to classify potato leaf diseases using the EfficientNetB0 architecture with two optimizers, Adam and SGD, and applying data augmentation techniques such as rotation, flipping, and cropping. The dataset consists of 3076 images divided into seven categories: Bacteria, Fungi, Healthy, Nematodes, Pests, Phytophthora, and Viruses. The results show that the Adam optimizer with a learning rate of 0.001, a batch size of 16, and 100 epochs provides the best performance. The training accuracy reached 92.10%, validation 81.49%, and testing 78.14%. The model precision was 0.7982, recall was 0.7536, and the F1 score was 0.7671. Meanwhile, the SGD optimizer produced a test accuracy of 79.55%, with precision of 0.7752, recall of 0.7781, and an F1 score of 0.7715. Although Adam's accuracy is higher, SGD shows better stability in preventing overfitting. This study confirms that data augmentation plays an important role in improving model performance, although the challenge of overfitting still needs to be addressed. Further studies are expected to optimize hyperparameters and explore other model architectures to improve the accuracy and efficiency of potato leaf disease classification.