MFVL-YOLO:Small Object Detection Algorithm for Aerial Remote Sensing Images

Abstract Small object detection in aerial remote sensing images remains a challenging task due to low resolution, dense object distribution, and complex backgrounds. In this paper, we enhance the YOLOv10 architecture by introducing a lightweight framework that combines multi-scale feature extraction in the spatial domain with high-frequency enhancement in the frequency domain to improve the extraction of fine details. The approach further incorporates an entropy-guided mechanism to strengthen foreground discrimination, a statistically constrained loss function to suppress background interference, and a shared detection head to reduce parameter redundancy and maintain scale consistency. Experiments on the VisDrone dataset show that the proposed method achieves improvements of 3.1% in mAP@0.5 and 3.5% in mAP@0.5:0.95 over strong baselines, while keeping computational overhead low. Evaluation on the NWPU VHR-10 dataset confirms the model’s robustness and generalization across varied remote sensing scenarios. These results demonstrate the effectiveness of the proposed method for accurate and real-time small object detection in complex aerial imagery.