Remote sensing images object detection is used in mineral exploration, transportation, national defense and military, emergency rescue and disaster relief and other fields. However, the common rotation object detection algorithm model applicable to remote sensing images is too large, difficult to deploy and can not meet the requirements of real-time detection, ignoring the balance between accuracy and speed. To solve the above problems, a rotation object detection algorithm YOLOV5s-R based on YOLOv5s is proposed. Firstly, angle parameters were added on the basis of YOLOv5s, then the horizontal bounding box was modified into oriented bounding box to adapt to the angular diversity of remote sensing image objects. Secondly, the Circular Smooth Label was introduced to avoid the angle mutation problem caused by the periodicity of angle regression. Then, the Efficient Channel Attention module was introduced to improve the ability of the model to extract important features. Finally, the Adaptively Spatial Feature Fusion module was introduced to solve the inconsistencies between different feature scales inside the feature pyramid. On the dataset DOTA, the experimental results show that the mAP₅₀ reaches 75.6%, the mAP_50∶95 reaches 46.7%, and the FPS reaches 81.9. Compared with the base model, mAP₅₀ and mAP_50∶95 increased by 1% and 3.1% respectively, and FPS increased by 85.9%. Therefore, YOLOv5s-R achieves more accurate and high-speed remote sensing images detection, achieving a good balance between accuracy and speed.