基于全卷积网络的遥感影像噪声点云监测仿真  

Research on New Algorithm for Remote Sensing Image Noise Point Cloud Monitoring Based on Full Convolution Network

作　　者：梁晓梅 孙荣 LIANG Xiao-mei;SUN Rong(Guilin Institute of Information Technology,Guangxi Guilin 541004,China;Guilin University of Electronic Technology,Guangxi Guilin 541004,China)

机构地区：[1]桂林信息科技学院,广西桂林541004 [2]桂林电子科技大学,广西桂林541004

出　　处：《计算机仿真》2023年第7期205-209,219,共6页Computer Simulation

基　　金：2021年度广西高校中青年教师科研基础能力提升项目(2021KY1656)。

摘　　要：云检测作为当前遥感影像数据处理的主要手段,其在地球气候分析、自然灾害分析、国土资源利用等方面越来越重要。在目前众多云检测技术研究过程中,大都采用基于阈值等传统经典的分析方法来研究。但是方法存在依赖先验知识、同时受主观的影响比较大等缺点。基于此提出了基于全卷积神经网络的遥感影像噪声点云检测的分析方法。方法首先将全幅遥感影像进行分割,然后为了真实还原遥感影像数据引入干扰噪声,同时利用全卷积神经网络可自动提取深层隐含特征的网络特性,来保留整个遥感影像的特征信息,在云检测特征提取完成之后通过随机场模型完成对云图的平滑处理得等到更为准确的云图检测。最后通过采用风云气象卫星云图,完成对云图影像的数据分析,结果表明上述算法可很好的完成对云图遥感影像的数据分析,特别是针对边缘云图具备很好的应用前景。As the main means of remote sensing image data processing,cloud detection is becoming more and more important in Earth climate analysis,natural disaster analysis,land and resources utilization and so on.In the current research process of many cloud detection technologies,most of them use traditional classical analysis methods such as threshold.However,this method has some disadvantages,such as relying on prior knowledge and being greatly influenced by subjectivity.Based on this,this paper proposes an analysis method of remote sensing image noise point cloud detection based on fully convolution neural network.Firstly,the full remote sensing image was segmented,and then in order to truly restore the remote sensing image data,the interference noise was introduced.At the same time,the fully convolution neural network can automatically extract the network characteristics of deep hidden features to retain the characteristic information of the whole remote sensing image.After the cloud detection feature extraction is completed,the cloud image smoothing was completed through the random field model,so we have to wait for more accurate cloud image detection.Finally,the paper completed the data analysis of cloud image by using the cloud image of Fengyun meteorological satellite.The results show that the algorithm can well complete the data analysis of cloud image remote sensing image,especially for edge cloud image,which has a good application prospect.

关 键 词：云监测 全卷积网络 遥感影像 精细化检测 云图影像 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]