Abstract: To address the challenges of detecting small objects in remote sensing images, such as difficulties with small object detection, large scale variations, arbitrary orientations, and insufficient detection accuracy on lightweight devices, this paper proposes a lightweight remote sensing image rotation object detection algorithm, LR-YOLO (Lightweight Rotating-YOLO), based on an improved YOLOv8 framework. First, the feature extraction module C2f-RA is designed using structural re-parameterization techniques and incorporating the EMA attention mechanism to enhance the network's feature extraction and representation capabilities, effectively filtering out complex background noise. Secondly, ADown sampling is employed in the network architecture to reduce feature map dimensions while minimizing the loss of contextual information. Finally, to address the issue of inconsistent predictions caused by the lack of interaction between the classification and localization branches of the detection head, a task-dynamically aligned detection head (ETDA-Head) is proposed. This promotes synergy between the classification and localization tasks, thereby improving the accuracy of the network model.Experimental results demonstrate that the improved algorithm achieves a 1.4% and 1.6% increase in detection accuracy on the DIOR-R and HRSC2016 datasets, respectively, compared to the baseline model, while reducing the number of parameters by 56.25%. Additionally, compared to the representative RoI Transformer algorithm, the detection accuracy is improved by 16.9%, indicating that the LR-YOLO algorithm significantly enhances detection accuracy while reducing model complexity, thereby validating its effectiveness.