基于差分隐私的矢量地理数据脱密方法  

Decryption method for vector geographic data based on differential privacy

作　　者：徐雅鑫 徐彦彦[1] 欧阳雪 徐正全[1] XU Yaxin;XU Yanyan;OUYANG Xue;XU Zhengquan(State Key Laboratory of Information Engineering in Surveying,Mapping and Remote Sensing,Wuhan University,Wuhan 430079,China)

机构地区：[1]武汉大学测绘遥感信息工程国家重点实验室,湖北武汉430079

出　　处：《测绘学报》2025年第2期356-370,共15页Acta Geodaetica et Cartographica Sinica

基　　金：国家重点研发计划项目(2021YFB2501103);国家自然科学基金(42271431);湖北省技术创新计划科技重大项目(2024BAA011)。

摘　　要：矢量地理数据必须采用脱密方法降低几何位置精度后才能安全共享和使用,现有脱密方法均无法对方法的安全性和脱密数据的可用性进行定量分析,难以实现安全性和可用性的最优均衡。本文首次将差分隐私技术应用于矢量地理数据脱密,创新性提出一种基于差分隐私的矢量地理数据脱密方法(DP-VGS),将现有非线性变换的脱密模型和差分隐私技术结合,通过敏感区域的划分和聚合、脱密安全预算的分配,使得敏感性高的区域脱密后的安全性更高;设计一种基于函数扰动和截断拉普拉斯机制的脱密模型加噪保护方法(FM-TL),提高脱密数据可用性。理论证明DP-VGS满足ε-差分隐私,即给定脱密安全预算ε的值,能够确定脱密模型的安全性并得到脱密模型的误差上界;并且这种基于差分隐私的脱密方法能跟现有脱密模型兼容。在4个真实数据集上的试验结果表明,本文方法达到了使脱密数据安全性和可用性最优的目的。Vector geographic data can be shared and used only after the geometric position accuracy is reduced by decryption methods,and none of the existing decryption methods are able to quantitatively analyze the security of the methods and the utility of the decrypted data.This paper is the first to combine differential privacy technology to decryption vector geographic data,and innovatively proposes a differential privacy-based method for vector geographic data decryption(DP-VGS),which combines the existing decryption model of nonlinear transformation and differential privacy.Firstly,through the division and aggregation of sensitive regions and the allocation of the decryption security budget,the regions with high sensitivity are made more secure after decryption.Secondly,a decryption model noise protection method based on function perturbation and TrunLap mechanism(FM-TL)is designed to improve the utility of decrypted data.Theoretical analysis demonstrates that DP-VGS satisfies differential privacy,which means that the security and error upper bound can be obtained by giving the decryption security budget,and DP-VGS is compatible with most of the existing decryption models.Experimental results on four real datasets show that the security of DP-VGS achieves the goal of optimizing the security and availability of the decrypted data.

关 键 词：矢量地理数据 差分隐私 脱密模型 函数扰动 截断拉普拉斯机制 

分 类 号：P208[天文地球—地图制图学与地理信息工程]