In recent years, significant advancements have been made in optical remote sensing technology, enabling the acquisition of vast quantities of high-resolution imagery that has facilitated progress across multiple domains. Among various computer vision tasks in remote sensing, aircraft detection holds substantial importance for both civilian and defense applications. However, the detection of small targets remains challenging due to limited feature representation and complex environmental interference. To address these issues, this study presents a novel multi-scale feature fusion framework specifically designed for robust small target detection in aerial imagery. The proposed methodology introduces several key innovations: First, a lightweight feature extraction module incorporating dynamic selection mechanisms is developed, which autonomously adjusts receptive field dimensions and sampling frequencies based on target characteristics. Second, an enhanced feature pyramid network (FPN) with adaptive channel-wise weighting is implemented, utilizing grouped convolutions to optimize feature representation. Additionally, a specialized dataset for aerial vehicle detection is constructed, supplemented by processed samples from the DOTA benchmark. Experimental evaluations demonstrate the effectiveness of the proposed approach. The comprehensive strategy (Group D) achieves a mean average precision (mAP) of 52.3%, representing a 12.7% improvement over conventional training methods (Group A, 39.6%). Data augmentation techniques (Group B) enhance sample diversity, enabling the model to capture more discriminative features and yielding a 45.1% mAP. These results validate the framework's capability to improve detection accuracy for small targets in complex remote sensing scenarios.