| 加权空-谱联合保持嵌入的高光谱遥感影像降维方法 |
| |
| 引用本文： | 黄鸿,石光耀,段宇乐,张丽梅.加权空-谱联合保持嵌入的高光谱遥感影像降维方法[J].测绘学报,2019,48(8):1014-1024. |
| |
| 作者姓名： | 黄鸿 石光耀 段宇乐 张丽梅 |
| |
| 作者单位： | 重庆大学光电技术与系统教育部重点实验室,重庆,400044;重庆大学光电技术与系统教育部重点实验室,重庆,400044;重庆大学光电技术与系统教育部重点实验室,重庆,400044;重庆大学光电技术与系统教育部重点实验室,重庆,400044 |
| |
| 基金项目： | 国家自然科学基金（41371338）；重庆市基础研究与前沿探索项目（cstc2018jcyjAX0093）；重庆市研究生科研创新项目（CYB18048） |
| |
| 摘 要： | 高光谱遥感影像数据量大、波段数多,容易导致“维数灾难”。传统流形学习方法一般仅考虑其光谱特征,忽略了空间信息。为此提出一种非监督的基于加权空-谱联合保持嵌入(WSCPE)的维数约简算法。首先采用加权均值滤波(WMF)方法对高光谱影像进行滤波,以消除噪点和背景点的干扰。然后根据遥感影像地物分布的空间一致性,通过采用加权空-谱联合距离(WSCD)来融合像素点的光谱信息和空间信息,有效选取各像素点的空-谱近邻,并根据像素点与其空-谱近邻点之间的坐标距离来有区别的利用其近邻点进行流形重构,提取低维鉴别特征进行地物分类。在PaviaU和Indian Pines数据集上的分类结果表明,总体分类精度分别达到了98.89%和95.47%。该方法在反映影像内部流形结构的同时,有效融合了影像的空间-光谱信息,故能提高影像特征的鉴别性,并提升分类性能。
|
| 关 键 词： | 高光谱遥感影像 流形学习 维数约简 空-谱近邻 鉴别特征 |
| 收稿时间： | 2018-05-15 |
| 修稿时间： | 2018-08-30 |
Dimensionality reduction method for hyperspectral images based on weighted spatial-spectral combined preserving embedding |
| |
| HUANG Hong,SHI Guangyao,DUAN Yule,ZHANG Limei.Dimensionality reduction method for hyperspectral images based on weighted spatial-spectral combined preserving embedding[J].Acta Geodaetica et Cartographica Sinica,2019,48(8):1014-1024. |
| |
| Authors: | HUANG Hong SHI Guangyao DUAN Yule ZHANG Limei |
| |
| Institution: | Key Laboratory of Optoelectronic Technique and System of Ministry of Education, Chongqing University, Chongqing 400044, China |
| |
| Abstract: | Hyperspectral image (HSI) contains a large number of spectral bands, which easily leads to the curse of dimensionality. However, the traditional manifold learning methods generally only consider the spectral features, while the spatial information of HSI is ignored. To overcome this shortcoming, it is proposed that an unsupervised dimensionality reduction algorithm called weighted spatial-spectral combined preserving embedding (WSCPE) for HSI classification. Firstly, the proposed algorithm uses a weighted mean filter (WMF) to filter the image, which can reduce the influence of background noise. Then, according to the spatial consistency property of HSI, it adopts the weighted spatial-spectral combined distance (WSCD) to fuse the spectral and spatial information of pixels to effectively select the spatial-spectral neighbors of each pixel. Finally, the proposed method explores the coordinate distances between pixels and their spatial-spectral neighbors to perform manifold reconstruction, and the low-dimensional discriminative features are extracted for HSI classification. The experimental results on PaviaU and Indian Pines datasets indicate that the overall classification accuracies of the proposed method reached 98.89% and 95.47%, respectively. The WSCPE method not only discovers the intrinsic manifold structure of HSI data, but also effectively integrates the spatial-spectral combined information, which enhances the classification performance. |
| |
| Keywords: | hyperspectral remote sensing image manifold learning dimensionality reduction spatial-spectral neighbors discriminant features |
| 本文献已被 维普 万方数据 等数据库收录！ |
| 点击此处可从《测绘学报》浏览原始摘要信息 |
| 点击此处可从《测绘学报》下载免费的PDF全文 |
|
