Abstract: Aimed at the problem that the traditional attention model is not effective in complex object detection scenarios, based on the CenterNet algorithm, an anchor-free object detection algorithm based on multi-scale feature aggregation attention is proposed. Multi-scale feature aggregation attention network (MFANet) was used in this algorithm. This attention network not only combined the features of different receptive fields between different blocks of the backbone network, but also paid attention to the important feature regions on the feature map. In addition to this, the keypoint prediction branch of CenterNet was redesigned to use a deeper convolutional network for better predictions. The experimental results on the UAVDT dataset and the COCO dataset show that, on the premise of ensuring real-time detection, the MFANet can effectively suppress the influence of background noise on the detector and improve the object detection accuracy in complex scenes.