Style transfer is a crucial area in image processing, enabling the fusion of content and artistic style from different images to create visually appealing outputs. However, existing approaches often struggle with low semantic consistency and poor stylization quality due to inherent conflicts between content and style, as well as limitations in current style feature extraction techniques. To overcome these challenges, we introduce AMST, an innovative aesthetics-guided multi-scale feature fusion approach for style transfer that captures intricate style patterns while harmonizing content and style. Our method introduces a two-stage aesthetic-aware enhancement (TSAAE) module, which integrates fine-grained style features through an aesthetic discriminator and applies an aesthetic constraint to improve style expression. Additionally, we design a multi-scale feature attention fusion (MSFAF) module to compute a semantic consistency measurement, thus promoting sufficient information exchange between content and style. To further improve stylization quality, we introduce two self-supervised tasks aimed at encouraging the generation of content-clear and vibrant results. Extensive qualitative and quantitative evaluations validate that our approach excels in synthesizing stylized images that are both semantically consistent and aesthetically pleasing.