基于3D关联规则深度学习的异构遥感数据检测  

Detection of Heterogeneous Remote Sensing Data Based on Deep Learning of 3D Association Rules

作　　者：覃伟荣 劳燕玲 QIN Wei-rong;LAO Yan-ling(College of Resources and Environment,Beibu Gulf Unversity,Qinzhou Guangxi 535011,China)

机构地区：[1]北部湾大学资源与环境学院,广西钦州535011

出　　处：《计算机仿真》2023年第9期482-486,共5页Computer Simulation

基　　金：国家自然科学基金项目(41966007)。

摘　　要：异构遥感数据中包含大量噪声,影响数据检测效果,为了高效、准确地获取检测结果,提出基于3D关联规则深度学习的异构遥感数据检测方法。通过双树复小波变换多尺度分解遥感数据中的噪声,分解后保留低频分量不变,高频分量使用四阶微分去噪模型去噪,重构高频和低频分量,完成遥感数据去噪。通过3D关联规则深度学习方法对去噪后的异构遥感数据的特征基元进行提取,获取属性特征,通过多尺度词包模型变化检测算法实现异构遥感数据检测。仿真结果表明,所提方法可以快速准确完成异构遥感数据检测。Heterogeneous remote sensing data contains a lot of noise affecting the data detection effect.In order to obtain the results efficiently and accurately,this paper puts forward a method of detecting heterogeneous remote sensing data based on deep learning of 3D association rule.From multiple scales,the noises in remote sensing data were decomposed by dual-tree complex wavelet transform.After decomposition,low-frequency components remained unchanged.Moreover,the fourth-order differential denoising model was used to remove noise from high-frequency components,and then high-frequency and low-frequency components were reconstructed.Thus,the noises were removed from remote sensing data.In addition,homogeneous feature primitives of the denoised heterogeneous remote sensing data were extracted through the deep learning method based on 3D association rule,so that the attribute features could be obtained.Finally,the detection of heterogeneous remote sensing data was realized by the change detection algorithm for the multi-scale bag-of-words model.Simulation results show that the proposed method can detect heterogeneous remote sensing data quickly and accurately.

关 键 词：关联规则 深度学习 异构遥感数据 数据去噪 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]