| 傅里叶谱纹理和光谱信息结合的高分辨率遥感影像地表覆盖分类 |
| |
| 引用本文： | 佃袁勇,杨光,方圣辉.傅里叶谱纹理和光谱信息结合的高分辨率遥感影像地表覆盖分类[J].武汉大学学报(信息科学版),2017,42(3):362-368. |
| |
| 作者姓名： | 佃袁勇 杨光 方圣辉 |
| |
| 作者单位： | 1.华中农业大学园艺林学学院, 湖北 武汉, 430070 |
| |
| 基金项目： | 国家863计划2012AA12A304国家863计划2013AA102401中央高校基本科研业务费专项基金2014QC018地理国情监测国家测绘地理信息局重点实验室开放研究基金2013NGC05 |
| |
| 摘 要： | 以高空间分辨率遥感影像为研究对象，将纹理特征与影像的光谱特征结合起来，用于地表覆盖类型分类。设计了一种基于傅里叶谱纹理的分类策略，对主成分分析后的第1、2主分量特征影像，利用径向谱（r-spectrum）提取纹理特征，并将纹理与光谱特征结合起来，构建了不同的分类特征用于支持向量机分类模型。以Salinas数据集和QuickBird影像为例，验证该算法。结果表明，纹理与光谱信息的结合可以明显提高高分辨率遥感影像的分类精度；由傅里叶径向谱提取的纹理特征可以很好的应用到高分辨率遥感影像的分类问题中，分类精度高于基于傅里叶总能量谱和灰度共生矩阵的分类精度；利用该算法对PCA变换后的第1和第2分量提取的纹理特征具有一定的互补性，并且结合多特征图像的纹理特征提取优于单特征图像的纹理特征提取。
|
| 关 键 词： | 高空间分辨率 径向谱 纹理特征 主成分分析 |
| 收稿时间： | 2015-10-23 |
Combining Fourier Spectrum Texture and Spectral Information for Land Cover Classification with High Resolution Remote Sensing Images |
| |
| Affiliation: | 1.Institute of Garden Forestry, Huazhong Agriculture University, Wuhan 430070, China2.School of Remote Sensing and Information Engineering, Wuhan University, Wuhan 430079, China |
| |
| Abstract: | The extraction of textural features gives more information in many pattern recognition issues. The textural features are widely exploited in the classification problems. In this study, the high resolution images were employed as the research object.Textural features and spectral features were combined to solve the problems of land cover classification. The paper designed a classification strategy based on Fourier spectrum texture. The spectral image was submitted to the principal components analysis (PCA) and r-spectrum was extracted from the first two principal components as the textural features. The common power spectrum might be insufficient because of the lack of the feature number. However, it would bring redundancy when taking into account each sample spectrum associated with each frequency. In fact, it could be divided and quantified flexibly. Consequently, new features were yielded through the method. In this study, different scenarios associated with different input features were designed. Support vector machine (SVM) was employed as the classifier. The algorithm was tested on hyperspectral dataset acquired in Salinas area and QuickBird images acquired in Jiufeng area. Results showed that the combination of textural features and spectral features could obviously improve the accuracy; the textural features extracted by Fourier r-spectrum could be applied well to the classification problems of high resolution remote sensing images; the classification accuracy was higher than the one that is based on the whole sample spectrum and the commongray-level co-occurrence matrix (GLCM); it was suitable to move the window pixel by pixel when extracting the textural features; adaptive weight would better deal with the problems of multiple features; the textural features extracted from the first and second principle components had complementary properties. In addition, the extraction of textural information from multi-feature pictures was superior to single feature picture. |
| |
| Keywords: | |
| 本文献已被 CNKI 等数据库收录！ |
| 点击此处可从《武汉大学学报(信息科学版)》浏览原始摘要信息 |
|
点击此处可从《武汉大学学报(信息科学版)》下载全文 |
