Aiming at the problem  of intra class  gap  in underwater semantic segmentation wi th few lab els, a feed shot un- der water santic segmentation wi th few lab els, a feed shot un- der water seman tic se gmentation ( F USS) model wit h informa- tion i nteractio n was prop ose d. By int eractin g the ext racted global feature info rm ation of the qu ery im age with th e local feat ure infor mat ion of the su pportin g image obj ect, the effec- tive attention to t he com mon o bjec t inf ormation was re alized. T he experi me ntal resul ts sh ow th at the pro posed FUSS model is sup eri or to the three less label semantics egmentation algorithms of CaNet、 PGNet 和 PaNet either in un derwa ter environment or on the ground image in the case of using only 50 supporting data sets.