The proposed algorithm utilizes an improved YOLOv8L model to detect rotating objects (such as ships and aircraft) in complex remote sensing images with arbitrary orientation, large scale variation, and dense array of objects. By incorporating a rotating frame with angle, the algorithm achieves more accurate target localization. Firstly, the decoupling angle prediction head is incorporated into the network's head section to accurately forecast the angular information of the target object. Secondly, by integrating a coordinate attention mechanism module, the model's capability to suppress noise is significantly enhanced. Lastly, an adaptive spatial feature fusion module is introduced in the neck section to effectively address inconsistencies in feature information fusion across different scales and retain valuable information for optimal fusion. The experimental results demonstrate that the proposed algorithm achieves a detection accuracy of 73.85% on the DOTA dataset, surpassing the original YOLOv8L model by 3.53%.