Embodied visual exploration is a fundamental challenge for autonomous agents navigating unknown environments. Traditional strategies often rely on complex geometric reasoning or handcrafted heuristics, which struggle with high-dimensional states in complex scenarios. In this paper, we propose an improved global policy for the Active Neural SLAM (ANS) framework by integrating a parameter-free attention mechanism and self-supervised auxiliary tasks. Specifically, the introduced SimAM module is for 3D attention weighting to prioritize salient features without adding parameters. Furthermore, rotation-based self-supervised tasks are introduced to improve generalization across unseen environments. Results on Gibson and MP3D datasets demonstrate that our approach significantly enhances exploration efficiency and coverage, particularly in large-scale scenes, outperforming current state-of-the-art learning-based and heuristic strategies.