Small object detection is a common and challenging task in UAV image analysis, where numerous small objects and complex backgrounds often lead to detection difficulties. Traditional detection methods typically struggle with small object detection due to insufficient feature extraction, resulting in suboptimal accuracy. To address these challenges, we propose an improved model based on RT-DETR, named STD-DETR. To overcome the issue of insufficient feature extraction for small objects, we introduce a Fast Fourier Transform Model (FFTM), which effectively extracts features from different layers, providing high-quality data for subsequent feature fusion. Additionally, we design a Bidirectional Concatenation Feature Pyramid Network (BCFPN) module to enhance the network’s ability to process multi-scale features. We also combine the internal IoU with Focaler-Shape-IoU to focus on difficult samples, accelerating convergence and improving small object detection performance. Our model demonstrates strong performance across various datasets during validation.