FCA-YOLO: A small object detection method based on feature attention fusion for UAV remote sensing images

Small object detection remains a significant challenge in computer vision, especially in scenarios such as remote sensing and unmanned aerial vehicles (UAVs) application, where there is considerable room for improvement in detection accuracy. The difficulty primarily arises from factors such as low resolution of the images, complex backgrounds in the images, and insufficient feature representation. To address these challenges, we propose FCA-YOLO, a novel detection framework built upon YOLOv11, specifically optimized for small object detection from UAV perspectives. First, a down-sampling structure was designed to better preserve small object features, which integrated a redundant compression feature transformation strategy with an inverted bottleneck residual block to enhance feature flow and representation capacity. Second, we propose a cross-scale feature fusion module that integrates spatial and channel attention mechanisms to effectively align and optimize multi-scale features, thereby enhancing the model's focus on small objects. Finally, a specialized detection structure was designed to enhance sensitivity to small targets, combining a dedicated detection head with skip connections that fused deep semantic features and shallow details, thereby improving the model's ability to capture fine-grained small information. Experimental results on the VisDrone2019 dataset demonstrated that FCA-YOLO outperforms the baseline model, achieving improvements of 3.1% in precision, 4.7% in recall, and 5% in mAP@0.5, while reducing the number of parameters by 30%. Compared with other YOLO variants and state-of-the-art algorithms, the proposed method achieves superior performance in terms of detection accuracy. Further evaluations on the DOTAv1.0 and VEDAI datasets validated the robustness and consistent detection performance of the proposed model across aerial imaging scenarios.