Segmentation of point clouds in all-weather conditions is crucial and constitutes an indispensable task within automated driving perception. The labeling of point clouds during abnormal weather poses a considerable challenge, often restricting the scalability of fully supervised methods. There exists a research gap in transitioning from normal weather datasets to abnormal weather datasets across different domains. To address this, We propose a method to randomly extract a limited set of point clouds for each abnormal weather type. These extracted point clouds are then integrated into the normal weather point clouds of the source domain for training. Additionally, we introduce a comprehensive pipeline for semi-supervised domain migration applicable to all-weather scenarios. This approach achieves nearly unsupervised point cloud labeling in all-weather datasets, achieves 72.6% of the fully supervised method’s performance, even with less than 1% of the target domain data.