SAR (Synthetic Aperture Radar) image feature information restoration method based on distributed training

The invention relates to an SAR image feature information restoration method based on distributed training, and belongs to the technical field of communication. According to the method, model training is performed on a YOLOv5S network by using an open-source SAR image and a correspondingly labeled data set, a plurality of unmanned aerial vehicle interconnection modes in a scene are used as a distributed scene, a distributed computing interconnection communication network and the trained YOLOv5S network are deployed in an unmanned aerial vehicle intelligent module to obtain the YOLOv5S network meeting requirements, real SAR image data are used as input, and the distributed computing interconnection communication network and the trained YOLOv5S network are used as the distributed scene. Forward propagation is carried out through a YOLOv5S network, and restored SAR image feature information is obtained after output results of all the unmanned aerial vehicle computing nodes are fused. According to the method, the