基于改进型PSPNet模型的高分辨率遥感影像建筑物变化检测方法研究
Research on building change detection method of high resolution remote sensing image based on improved PSPNet model
刘海红 1李延龙 1李亚刚 1刘鑫2
作者信息
- 1. 青海省基础测绘院,青海 西宁 810000
- 2. 某部队,北京 100042
- 折叠
摘要
针对复杂场景下小地物漏检,裸地、道路等干扰因素引起建筑物变化样本不足导致的精度低等问题,提出一种改进的金字塔场景解析网络(Pyramid Scene Parsing Network,PSPNet).搭建孪生PSPNet网络作为基础模型,通过融合不同尺度的特征层信息,提高小尺度地物的检测精度;其次针对变化样本不足问题,结合多任务思想,使语义任务和变化检测任务在同一个网络中进行,从而解决变化样本不足的问题.结果表明:改进后的PSPNet模型在建筑物变化检测中的精确率为92.35%,召回率为85.61%,F1分数为0.888 5.相比原始的PSPNet模型精度提升6.2%、12.03%和0.09.本研究可为复杂场景下建筑物变化检测提供技术支持.
Abstract
Aiming at the low precision problems caused by the missing detection of small ground objects in complex scenes and the insufficient sample of building changes caused by interference factors such as bare ground and roads.An improved Pyramid Scene Parsing Network(PSPNet)was proposed.First,twin PSPNet networks were built as the basic model,and the detection accuracy of small-scale ground objects was improved by integrating the feature layer information of different scales.Second,to solve the problem of insufficient change samples,the semantic task and change detection task are carried out in the same network by combining the multi-task idea,so as to alleviate the problem of insufficient change samples.The results show that the accuracy rate of the improved PSPNet model is 92.35%,the recall rate is 85.61%,and the F1 score is 0.8885.Compared with the original PSPNet model,the accuracy is improved by 6.2%,12.03%and 0.09.This study may provide technical support for building change detection in complex scenes.
关键词
多尺度特征融合/建筑物变化检测/检测精度Key words
multi-scale feature fusion/building change detection/detection accuracy引用本文复制引用
出版年
2024
NETL
NSTL
万方数据