Image aesthetic assessment (IAA) quantitatively measures the resonance level between human aesthetic perception and images by analyzing visual content and artistic aesthetic value. Human aesthetic perception follows a hierarchical process. It begins with a coarse observation of the overall image and then proceeds to fine-grained judgments of the harmony of details. To address the impact of complex relationships between features at different levels on aesthetic assessment, this paper proposes a multi-level hierarchical adaptive fusion model for IAA, named MHAF. First, multi-hierarchy feature extraction is performed using 2D selective scan mechanism of VMamba to capture aesthetic features at different levels. Subsequently, an adaptive feature propagation module enhances the expressive power of hierarchical features and filters redundant information. Finally, cross-hierarchy attention fusion is conducted in a top-down manner, enabling the fusion of multi-level aesthetic information. Experimental results demonstrate that the proposed model consistently achieves superior competitive performance across all IAA tasks compared to the state-of-the-art methods.