A Dual Semantic-Aware Recurrent Global-Adaptive Network For Vision-and-Language Navigation