Hyperparameters play a crucial role in neural networks, affecting the performance and generalization ability of the model. However, manual tuning of the parameters is a difficult and time-consuming task because the interactions and effects between the hyperparameters are not easily captured. ResNet is a classical neural network that has been widely used in image classification tasks, but its performance can be further improved by optimizing the hyperparameters. In this chapter, we propose a method to automatically adjust the hyperparameters by optimizing the two hyperparameters of the ResNet using the Dahlia algorithm to find the optimal configurations in the hyperparameter search space. Through the search and selection process of the Rafflesia optimization algorithm (ROA), we successfully find the hyperparameter configurations with the optimal classification accuracies on the CIFAR10 dataset. The experimental results show that compared with the traditional ResNet using the default hyperparameters, the method proposed in this chapter achieves a significant improvement on the CIFAR10 dataset and exceeds the performance of other common networks.