The Research on Galaxy Image Classification Model Based on ResNet18 and Transfer Learning

Traditional galaxy image classification relies on manual identification and categorization, which is time-consuming and costly. In response to this, this paper proposes a galaxy image classification method based on the combination of ResNet18 and transfer learning. The ResNet18 residual model is pretrained on the ImageNet dataset and the galaxy image training set. During training, the transfer learning (TF) method is employed, and a new fully connected layer is established to classify galaxy image categories into five classes. Experimental results show that this method achieves an accuracy of 93.16%, a precision of 90.40%, a recall of 86.71%, and an F1 score of 88.46%, outperforming methods such as AlexNet, LeNet, and ViT. The experiments demonstrate that transfer learning can accelerate convergence, reduce training time, and ResNet18 can effectively extract the feature information of galaxy images. This study establishes an automated galaxy image classification method, capable of accurately classifying galaxy images.