Research on Rapid Acquisition of Ice Crack Morphology Using a Decision Tree Fusion Model with FP-GPR

Detecting cracks in lake ice to assess ice safety is crucial for ensuring the safety of people's production and other activities. Full-Polarimetric Ground-Penetrating Radar (FP-GPR) can acquire both co-polarization and cross-polarization data, and by applying polarization decomposition methods to these FP data, it is possible to more effectively analyze ice crack morphology. In this study, a FP-GPR was employed to detect typical ice layer cracks and analyze their polarization characteristics. The resulting data were then input into a machine learning fusion model that integrates random forest and gradient boosting decision trees, enabling rapid extraction of the morphological information of ice cracks. The results show that polarization decomposition features facilitate the identification of ice crack morphology, and both forward modeling and real-data tests exhibit small errors. This approach can be used to assess the potential risk of ice cracks and to support preventive measures that reduce the societal and environmental impact of related disasters.