An Improved YOLOv11-Based Algorithm for Impact Crater Detection

Crater contains lots of geological information about celestial bodies. Then, analyzing celestial history to detect lunar impacts is an important project. In this paper, an enhanced convolutional neural network is proposed. This is achieved by utilizing an improved Local Importance-based Attention (LIA) mechanism. Then, the model integrated a Channel of Local Importance-based Attention (CLA) improved from LIA, named YOLOv11_CLA. In the experiment using datasets with about 190000 labeled targets for training and 50000 for validation, the result shows that the performance of the YOLOv11_CLA mode in precision is 79.30%, surpassing that the default model YOLOv11, which is 78.60%.