摘要
滑坡区域遥感检测与识别在灾情提取、救援决策和防灾减灾等方面都有着巨大的应用前景。针对滑坡遥感检测中目标颜色特征化模型不准确,对滑坡区域检测识别效果不够理想等问题,提出一种基于滑坡区域颜色特征模型的支持向量机(support vector machine,SVM)遥感检测方法。根据光谱学和色度学的基本理论,建立滑坡区域红绿蓝特征获取方法,以多光谱图像为基础,通过典型样本分析,确立目标/背景颜色特征化数字模型和有效边界。将该模型生成训练样本用于滑坡区域SVM检测模型训练,再将训练好的分类器用于滑坡区域的检测识别,在此基础上根据滑坡基础形状模型的轴向长宽比、面积参数和不变矩等典型形状特征指标对滑坡区域进行目标精确分类与识别。利用九寨沟地震后获取多光谱遥感图像进行了滑坡区域检测识别效果对比试验,试验结果表明,该方法能有效识别遥感图像中的滑坡样本点,对滑坡区域的识别精度由传统方法的90%左右提高到97.03%。
Landslide remote sensing detection has huge potential in the areas such as extracting disaster information,rescue decision making,disaster prevention and reduction.To solve the problem that target color characterization model is not accurate in landslide remote sensing detection,and rate of detection and recognition of the landslide surface is poor,a detection method for landslide based on color feature model and SVM in remote sensing imagery is proposed in this paper.The RGB feature acquisition method for landslide is established based on the basic theories of spectroscopy and chromaticity.Multi spectral remote sensing image of landslide area is used for color feature modeling and target detection experiment.The target/background color characteristic digital model and effective boundary are established through typical sample analysis.Training samples generated by the model are used to SVM,and then the trained SVM is used to landslide recognition.The landslide targets can be further identified and recognized with high accuracy by classification based on the axial aspect ratio,area and invariant moment of fundamental shape model.Landslide detection experiment based on multi spectral remote sensing images after Jiuzhaigou earthquake shows that the proposed method can effectively identify landslide sample points in remote sensing imagery,and the recognition accuracy for landslide is promoted to 97.03%from the about 90%of the traditional methods.
作者
陈善静
康青
沈志强
周若冲
CHEN Shanjing;KANG Qing;SHEN Zhiqiang;ZHOU Ruochong(Department of Military Facilities,Army Logistics University,Chongqing 401311,China)
出处
《航天返回与遥感》
CSCD
2019年第6期89-98,共10页
Spacecraft Recovery & Remote Sensing
基金
重庆市教委科学技术研究项目(KJQN201912905)
关键词
目标检测
滑坡颜色特征建模
滑坡灾害信息提取
支持向量机
遥感图像
target detection
color feature modeling on landslide
information extraction of landslide
support vector machine
remote sensing image
