AquaticCLIP: A Vision-Language Foundation Model for Underwater Scene Analysis