Due to the imaging Angle and height of remote sensing images, there are scale differences even for the same object in remote sensing images. However, traditional multi-scale methods often stack a large number of convolutions to ensure the extraction of multi-scale features, which leads to a sharp increase parameters of the model. Therefore, in order to solve the above problems, we propose a algorithm based on multi-scale feature enhancement based on YOLOv8 framework. Firstly, a lightweight multi-scale feature extraction module is proposed, which extracts rich multi-scale information through convolution kernels of different sizes, and introduces modules such as spatial channel reconstruction convolution to reduce the increase of model parameters and improve the detection efficiency of the model. Secondly, for the SPPF module of multi-scale feature fusion, we proposed SPPF-AA, which fused richer multi-scale features, on the other hand alleviated the problem of small object feature loss caused by the original module, and enhanced the multi-scale detection effect. Finally, for the feature decoupling requirements in multi-scale object detection, we design a scale-spatial weight fusion detection head. For the difference of feature requirements between classification and localization, we design a scale-weight fusion module and a spatial weight fusion module to realize the feature optimization of classification and localization tasks.