Small object detection in remote sensing images faces challenges such as small target size, indistinct features, and complex backgrounds, which limit the effectiveness of traditional detection methods. In recent years, deep learning-based object detection algorithms, especially single-stage methods like the YOLO series, have made significant progress in remote sensing due to their efficiency and real-time performance. However, existing methods still exhibit limitations in feature extraction, fusion, and downsampling, which affect their performance in small object detection tasks. To address these issues, this paper proposes a feature enhancement model, FE-YOLO. The model introduces a Hybrid Convolution Module (HCM), which combines standard convolution with RFAConv to significantly improve feature extraction capabilities. Additionally, a Hierarchical Feature Fusion Module (HFFM) is proposed to efficiently integrate multi-level features, enhancing the representation of small objects. Furthermore, a Haar Wavelet Downsampling Module (HWD) is employed to replace traditional strided convolution, reducing information loss during downsampling. Experimental results demonstrate that FE-YOLO achieves outstanding performance in small object detection tasks in remote sensing images.