The current domain adaptation methods employing feature alignment mainly focus on aligning low-level and high-level features separately. Although they are effective at aligning texture and semantic features, they ignore the scale variations of objects in different data domains. As a result, the model after feature alignment cannot obtain features of objects at different scales on Synthetic Aperture Radar (SAR) remote sensing image. To address these challenges, this paper proposes a domain adaptive SAR image object detection network based on scale aware alignment and designs a multi-scale feature extraction module and a multi-scale feature alignment module. Based on DA Faster RCNN [1], our network uses dilated convolutions to extract and fuse feature representations across different dimensions of the latent space and employs multi-head mixed convolutions for feature alignment, enhancing the domain classification network’s ability to align features of objects at different scales, thereby improving the cross-domain object detection performance of the model. Experimental results demonstrate that the proposed method achieves improvements in Average Precision (AP) compared to other state-of-the-art domain adaptive object detection methods on two cross-domain datasets from optical image to SAR image.