Plants account for over 80% of the human diet, being essential for food security and worldwide feeding. In this context, it is important to maintain
high productivity, avoid crop losses and preserve the environment. Plant leaf diseases are a drawback in this reality, where they are responsible for
depreciation in food quality, directly causing economic losses in agricultural production and great environmental impact if pesticides are used
indiscriminately. Therefore, the classification of diseases is of fundamental importance and is a great challenge since many leaf diseases present
similarities, inducing misidentification. Also, manual classification is an exhaustive task and gives subjective results, causing misidentification in
addition to being economically unfeasible. To solve this problem, a new arrangement of deep convolutional neural networks, activation
functions and optimizers for plant leaf disease classification is proposed. In the proposed method, we tested different EfficientNet Convolutional Neural Network models, scaling the model size and number of parameters alongside with the Mish activation function and Ranger Optimizer in the task of plant leaf disease classification. Compared with previous work applied into the same dataset, the
proposed arrangement achieved better performance with 94,0% accuracy for EfficientNetB0, achieving state-of-the art results for the tested
dataset. The model with best performance also has less parameters, therefore being effective and demonstrating potential for portability.