To overcome the training-data dependence and enhance the model transferability, we propose an unsupervised GPR target clustering method based on the contrastive language-image pretraining (UGTC-CLIP) ...