A Lightweight Remote Sensing Object Detection Network Based on YOLO11

Remote sensing object detection is a critical research direction in computer vision field for its broad application prospects in both military and civilian fields. In remote sensing object detection tasks, the images often have notable characteristics of high resolution, complex scenes, small target sizes, and dense distributions. These characteristics make the extraction of fine-grained information at shallow layers particularly important. Thus, this paper proposes a lightweight object detection network with a novel feature detection layer (C3P2) to improve the accuracy and robustness of small object detection in remote sensing images. C3P2 employs a structural control mechanism which is responsible for switching between SplitC3K (SC3K) and SplitC2f (SC2f), thereby flexibly balancing the model performance and efficiency. The SplitBottleneck in SC3K and SC2f processes only part of channel features, where the splitting-and-concatenation strategy is utilized to preserve the feature information density. This allows the model to better perceive small objects while reducing its model complexity. The experimental results on RSOD and NWPU VHR-10 datasets show that our method has significant advantages compared with other SOTA models.