“Experts propose a cross modal prior driven semantic enhancement active learning method to improve sample selection quality, reduce data annotation costs, and enhance model learning efficiency.”