Optimizer Performance in Deep Learning for Leaf Disease Classification

Global food security is at risk due to the spread of plant diseases, particularly in regions where agriculture is a significant economic activity. Early and precise identification of plant diseases helps avoid crop loss and maintain agricultural output. Here, we investigate different deep learning optimizers for the classification of leaf diseases. We employ EfficientNet-B0, a cutting-edge model built on high accuracy and efficiency in image classification tasks intended for agricultural settings with limited resources. The PlantVillage and PlantDoc databases are used to determine the best optimizer. We evaluate the results of five popular optimizers on EfficientNet-B0: Adam, Nadam, Adagrad, RMSprop, and SGD. Empirical findings indicate that Adam produces training, validation, and testing outcomes that outperform other optimizers. Future real-time agricultural application implementations are anticipated to be fueled by this insight.