UVDA网络在建筑物损毁检测中的应用  

Application of UVDA Network in Building Damage Detection

作　　者：孙彦春 李云栋[1] 

机构地区：[1]北方工业大学信息学院,北京100144

出　　处：《工业控制计算机》2025年第2期100-101,104,共3页Industrial Control Computer

摘　　要：由于无人机技术和遥感图像技术的发展,灾后灾区图像更容易获取。由于新获取的图像受到地区建筑风格和灾害类型的影响,已经训练好的模型无法对新的受灾情况做出准确的判断。为了解决这个问题,使用了一种无监督变分域适应网络,分别通过于跨域概率生成模型(CPGM)和对抗性训练来适应类别级别和全局特征。在分别以三种飓风造成的自然灾害采集到的图像为源域数据集,以一种地震造成的自然灾害采集到的图像为目标域数据集上取得了较好的检测效果。The development of drone technology and remote sensing image technology has greatly facilitated the acquisition of post disaster images in disaster areas.Due to the influence of regional architectural styles and disaster types on the newly acquired images,the trained models are unable to make accurate judgments on the new disaster situation.To address this issue,this paper employs an unsupervised variational domain adaptation network that adapts to category level and global features through cross domain probability generation model(CPGM)and adversarial training,respectively.Good detection performance was achieved on the source domain dataset using images collected from three types of natural disasters caused by hurricanes,and on the target domain dataset using images collected from one type of natural disaster caused by an earthquake.

关 键 词：神经网络 域适应 建筑物损毁检测 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]