Clear reasoning and intuitive understanding are essential for making human decision-making interpretable and trustworthy. Our model leverages tree structures and prototypes to guide adaptive learning of local attention patterns, capturing hidden features and providing explanations for both when decisions are made and where the model focuses. We propose a deep fine-grained visual classification framework that integrates spatial attention with soft decision trees, aligning prototypes with inner nodes for improved interpretability. The framework establishes a dynamic coordination mechanism through three components: (1) spatial attention that initially identifies potentially discriminative regions, (2) learnable prototypes aligned with the inner nodes of a soft decision tree that dynamically adapt to local feature distributions, and (3) hierarchical tree decisions driven by prototype activation patterns. End-to-end training enables the attention mechanism to progressively focus on regions that maximally activate discriminative prototypes. Simultaneously, the prototypes evolve to capture salient visual patterns. This co-adaptation results in a self-optimizing system where attention localization and prototype representation mutually enhance each other, with the tree structure providing transparent decision pathways. This approach generates human-interpretable decision pathways grounded in contextually meaningful prototypes, enabling step-by-step verification of model reasoning while achieving both high classification accuracy and clear interpretable explanations. We demonstrate the effectiveness of our approach in classification tasks, achieving strong predictive performance alongside enhanced interpretability.