| 基于双边滤波和空间邻域信息的高光谱图像分类方法 |
| |
| 引用本文： | 廖建尚,王立国,郝思媛.基于双边滤波和空间邻域信息的高光谱图像分类方法[J].农业机械学报,2017,48(8):140-146，211. |
| |
| 作者姓名： | 廖建尚 王立国 郝思媛 |
| |
| 作者单位： | 广东交通职业技术学院,哈尔滨工程大学,青岛理工大学 |
| |
| 基金项目： | 国家自然科学基金项目（61275010、 61675051）、国家星火计划项目（2014GA780056）和广东交通职业技术学院校改重点科研课题（2017-1-001） |
| |
| 摘 要： | 提出了一种基于双边滤波和像元邻域信息的高光谱图像分类(BS-SVM)算法。该方法首先利用双边滤波器提取经主成分分析降维后的高光谱图像空间纹理信息,然后通过设计一种高光谱像元邻域信息来构建高光谱的空间相关信息,最后将2种空间信息融合后与光谱信息结合,形成空谱信息(空间信息和光谱信息)后交由支持向量机完成分类。实验结果表明,相比单纯使用光谱信息的支持向量机的分类方法以及基于Gabor滤波的空谱信息结合分类方法,所提出的BS-SVM方法分类精度有较大幅度提高,充分证明了该方法的有效性。
|
| 关 键 词： | 高光谱图像 分类 双边滤波 像元邻域信息 空间相关信息 |
| 收稿时间： | 2016/12/18 0:00:00 |
Hyperspectral Image Classification Method Combined with Bilateral Filtering and Pixel Neighborhood Information |
| |
| LIAO Jianshang,WANG Liguo and HAO Siyuan.Hyperspectral Image Classification Method Combined with Bilateral Filtering and Pixel Neighborhood Information[J].Transactions of the Chinese Society of Agricultural Machinery,2017,48(8):140-146，211. |
| |
| Authors: | LIAO Jianshang WANG Liguo and HAO Siyuan |
| |
| Institution: | Guangdong Communication Polytechnic,Harbin Engineering University and Qingdao University of Technology |
| |
| Abstract: | Supplementing spectral information with spatial information to improve the classification of hyperspectral image is becoming a hot research in recent years. An improved scheme was put forward according to existing methods. An algorithm of supervised classification was proposed which was combined with bilateral filter and pixel neighborhood information (BS-SVM). Firstly, the spatial texture information of hyperspectral image was extracted whose dimensionality was reduced by PCA. Secondly, spatial correlation information was formed by building pixel neighborhood information of hyperspectral image. Finally, spatial-spectral information was merged by the two kinds of spatial information and the spectral information, which was classified by SVM. The BS-SVM classification method was implemented on the hyperspectral data of Indian Pines and Pavia. The results indicated that in the first place, the OA (Overall accuracy) of G-SVM for Indian Pines and Pavia were 3%~4% and 2%~3% higher than those of SVM, the same index for B-SVM were 3%~4% higher than that of G-SVM, and the classification performance can be improved effectively by the spatial texture information of hyperspectral image extracted by bilateral filter. Furthermore, the salt and pepper can be removed effectively by BS-SVM, showing very good performance in hyperspectral classification. In the second place, the classification of some methods for Pavia was better than the Indian. The reason was that the types and distribution of grounds for Indian were more complicated than Pavia. The classification for the less ground were bad, especially the Oats (only 20) was the worst. Therefore, it directly led to the AA (Average accuracy) generally lower than OA. However, the standard deviation of the classification for BS-SVM was much smaller than those of other methods, and the effectiveness of the method was verified with good stability. The experiments showed that the BS-SVM algorithm was better than original SVM with the pure spectrum information, the spatial-spectral information-based methods with Gabor. With the spatial correlation information extracted by the bilateral filter and the pixels neighborhood information, the performance of the classification with BS-SVM algorithm was greatly improved, and the effectiveness of BS-SVM was fully verified in the classification of hyperspectral image.The method can be applied to the field of crop growing, accurate classification and identification. |
| |
| Keywords: | hyperspectral image classification bilateral filter pixel neighborhood information spatial correlation information |
| 本文献已被 CNKI 等数据库收录！ |
| 点击此处可从《农业机械学报》浏览原始摘要信息 |
| 点击此处可从《农业机械学报》下载免费的PDF全文 |
|
