An increase in global population growth has necessitated an increase in food production. One of the main factors influencing annual agricultural production is the abnormal physiological functioning of plants or plant diseases, which directly affects the vegetation, leading to a reduction in plant yields and in the worst case, may even destroy the entire plantation. A majority of the diseases can be identified through the plant’s leaves, which a plant pathologist traditionally does. This, however, is a time-consuming task and the accuracy of diagnosis depends a lot on the expertise of the pathologist. Convolution neural networks (CNNs) have shown immense potential in image identification tasks. However, optimizing its hyperparameters and layouts is a challenging task. We proposed a genetic algorithm to enhance the performance of CNNs for plant disease identification by assessing the most effective hyperparameters and architecture for the fully connected layers of four cutting-edge CNNs: VGG16, Xception, DenseNet201, and ResNet152V2. The results show that genetic algorithms possess the potential to enhance the performance of CNN architecture in the agriculture domain, especially when used for plant disease identification.