As for aircraft remote sensing images with diversified scales and dense targets,the detection accuracy is relatively low,and the model is complex,which is not easy to deploy.To solve the problems,a remote sensing aircraft detection model based on the improved YOLOv4 is proposed.The K-means++ algorithm is adopted to optimize the anchor frame of target samples to improve size matching between the prior frame and the target,which reduces the missed detection rate.Focal loss is introduced in the loss function to reduce the weight of simple negative samples in the training process.Convolution kernel pruning and inter-layer pruning are fused for sparse training of the convolution kernel and batch normalized BN layer,which simplifies network structure and reduces the amount of parameters.Through experiments,the improved YOLOv4 foreign object detection algorithm has an mAP of 92.23% on the public UCAS-AOD and RSOD remote sensing data sets,and the detection speed is increased to 130.24 frame/s,which is conducive to rapid detection of aircraft targets in remote sensing images in actual industrial scenarios.