Galaxy Morphology Semantic Segmentation Based on K-Net

At present, modern large-scale astronomical observation equipment can collect a large number of high-resolution images. Astronomy has entered the era of big data. Deep learning has shown great potential in astronomical image analysis due to its powerful feature extraction and segmentation capabilities. In the study of galaxy evolution, image segmentation can separate different parts of galaxies, help identify the types of galaxies and analyze structural details, which is crucial to understanding the formation and evolution of galaxies. Therefore, this paper proposes a galaxy morphology segmentation method based on K-Net. The experimental data comes from the Galaxy Zoo project, and the model is trained using strategies such as transfer learning. The trained model can be used to segment four common galaxy images. The experimental results show that the overall accuracy reaches 97.11%, and the mean intersection over union is also 80.87%, indicating that the model performs well in the segmentation of astronomical images. The mean accuracy and the mean dice coefficient are both higher than 87%, indicating that the model is stable in distinguishing different categories. Therefore, K-Net has great application potential in astronomical image segmentation tasks.