摘要
针对大尺度遥感中重要特征破损所导致的检测精度低、耗时长等问题,提出了一种基于YOLOv5模型的残缺目标检测与评估方法,弥补了深度网络在残缺目标识别与评估领域的空缺。围绕机场典型区域,构建了以残缺目标为核心的机场目标数据集(Airport Target Dataset,ATD);在YOLOv5模型的基础上进行了目标推理的适配性改进,构建了一种针对大尺度遥感图像的切片增强型辅助处理框架(Slice Enhancement Aided Inference Framework,SEAIF)。实验结果表明:残缺目标识别精度达到90%以上,单张图像平均处理时间小于40 s,在精度与速度上远超专业判读员。该方法有助于及时准确地评估机场基础设施,帮助灾难响应和维护操作,有着重要的应用前景。
Aiming at the problem of low detection accuracy and long time due to the damage of important features in largescale remote sensing,this paper proposes a detection and evaluation method of incomplete targets based on YOLOv5 model to make up for the gap in the field of recognition and evaluation of incomplete targets in deep networks.Firstly,the Airport Target Dataset(ATD)with incomplete target as the core is constructed around the typical area of the airport,and the adaptation of target reasoning is improved based on the YOLOv5 model,the Slice Enhancement Aided Inference Framework(SEAIF)for large-scale remote sensing images is constructed.The experimental results show that the accuracy of incomplete target recognition is more than 90%,and the average processing time of single image is less than 40 s,which is far higher than that of professional interpreters in accuracy and speed.This method is helpful to timely and accurate assessment of airport infrastructure,aid disaster response and maintenance operations,and has important application prospects.
作者
于淼
宋政伟
张元淳
孙莉
张国和
刘达
YU Miao;SONG Zhengwei;ZHANG Yuanchun;SUN Li;ZHANG Guohe;LIU Da(School of Microelectronics,Department of Electronics and Information,Xi′an Jiaotong University,Xi′an 710077,China;School of Aeronautical Engineering,Air Force Engineering University,Xi′an 710038,China)
出处
《微电子学与计算机》
2025年第4期16-27,共12页
Microelectronics & Computer
基金
陕西省重点产业创新链(群)-工业领域(2022ZDLGY06-02)。
