摘要
为提高采用遥感影像监测开采沉陷演化的准确性,探讨了基于多维纹理特征的影像分类方法。首先提取影像的多维纹理特征:局部方差、局部平均梯度、局部能量和局部信息熵,然后将其与地物光谱值一并作为人工免疫算法中样本的特征向量,利用免疫算法的选择、克隆、变异算子进行自学习得到全局最优聚类中心,从而提高影像分类精度。对淮南煤田进行开采沉陷遥感监测,结果表明,该方法分类总精度为88.26%,Kappa系数为0.853,优于传统的Parallelepiped和Maximum likelihood分类方法。
To improve the accuracy of monitoring mining subsidence by remote sensing image, the image classification based on multi-dimensions texture features was proposed. In this classification process, the multi- dimensions texture features including local square difference, local average grades, local energy and local information entropy were extracted, and then along with spectrum were used to compose eigenvector in the artificial immune algorithm. Through the selection operator, clone operator and mutation operator, the global optimum cluster center was obtained, so the accuracy of image classification was improved. This method was applied to monitor mining subsidence in Huainan based on TM image classification. The results show that this method is superior to the Parallelepiped and Maximum likelihood methods, and its overall accuracy and Kappa coefficient reaches to 88.26% and 0.853 respectively.
出处
《煤田地质与勘探》
CAS
CSCD
北大核心
2008年第6期29-34,共6页
Coal Geology & Exploration
基金
安徽省2003年度地勘基金项目(2003-38)
关键词
遥感影像
开采沉陷
影像分类
多维纹理特征
remote sensing image
mining subsidence
image classification
multi-dimensions texture features
