Multi-focus image fusion is a crucial technique for improving image clarity by combining images taken at different focal depths. However, existing deep learning-based methods tend to overlook crucial details in the feature extraction and fusion phases, substantially affecting the quality of the fused images. To address this, in the feature extraction phase, we crafted an effective feature extraction block that boosts feature recognition by modeling channel interdependencies and captures global information through global average pooling. This block excels in gathering positional details from feature maps, significantly boosting clarity and detail retention. In the feature fusion phase, we introduce a novel feature fusion module (SFLE) that integrates the Local Energy (LE) operator to uncover intensity and detail distribution in source images, with Spatial Frequency (SF) measuring pixel value changes. This fusion technique ensures vital details are preserved with greater fidelity, facilitating a seamless integration of features from the input images. Accordingly, we present a novel dual-stream deep feature aware network (DFANet). Extensive experiments show that our model outperforms existing methods both qualitatively and quantitatively.