大气与环境光学学报 ›› 2020, Vol. 15 ›› Issue (2): 117-124.

• 光学遥感 • 上一篇    下一篇

基于高光谱影像多维特征的植被精细分类

苗宇宏1,2, 杨敏3*, 吴国俊1,2   

  1. 1中国科学院西安光学精密机械研究所陕西省海洋光学重点实验室, 陕西 西安 710119;
    2青岛海洋科学与技术试点国家实验室海洋观测与探测联合实验室, 山东 青岛 266235;
    3 国家海洋局北海海洋技术保障中心, 山东青岛 266033
  • 出版日期:2020-03-28 发布日期:2020-03-24

Sophisticated Vegetation Classification Based on Multi-Dimensional Features of Hyperspectral Image

MIAO Yuhong1,2, YANG Min3*, WU Guojun1,2   

  1. 1 Key Laboratory of Marine Optics (Shaanxi), Xi'an Institute of Optics and Fine Mechanics, Chinese Academy of Sciences, Xi'an 710119, China;
    2Joint Laboratory of Marine Observation and Exploration, Pilot National Laboratory for Marine Science and Technology (Qingdao),Qingdao266235, China;
    3 North China Sea Marine Technical Support Center, State Oceanic Administration, Qingdao 266033, China
  • Published:2020-03-28 Online:2020-03-24

摘要: 目前,高光谱植被精细分类存在三个问题:单纯利用光谱信息得到的分类精度较低;光谱数据存在噪声影响了最终的分类结果;
缺少针对具体应用场景而设计的分类方法。为此,提出了一种基于高光谱影像多维特征的植被精细分类方法,通过光谱
数据降维、纹理特征提取以及植被指数选择三个方面对高光谱影像数据进行分析与利用,依靠前期现场调查得到的地面
植被分布情况,选择训练样本并进行支持向量机(Support vector machine, SVM)监督分类,完成地面植被的精细分类,
对分类结果进行验证,总体精度可达99.6\%。结果表明,基于高光谱影像多维特征的植被分类方法能够有效地减小数据噪声、
提高信息利用率,为植被生态监测提供更为准确的数据支撑。

关键词: 高光谱, 光谱降维, 纹理特征, 植被指数, 支持向量机

Abstract: At present, there are three major challenges in the sophisticated vegetation classification using hyperspectral image. 
The first is that the accuracy of classification obtained simply by using spectral information is low. The second 
is that the presence of noise in the spectral data affects the final classification results, and the third is 
the lack of classification methods designed for specific application scenarios. To this end, a method for 
sophisticated vegetation classification based on multi-dimensional features of hyperspectral images is 
proposed. In this method, hyperspectral image data are analyzed and utilized firstly through three 
aspects of spectral data dimension reduction, texture feature extraction and vegetation index selection. 
And then, based on the distribution of ground vegetation obtained from previous field surveys, training 
samples are selected and Support Vector Machine (SVM) supervised classification is performed, which 
results in the sophisticated classification of ground vegetation at last. To verify the classification 
results, the overall accuracy can reach 99.6\%. The result shows that vegetation classification based on 
multi-dimensional features of hyperspectral image can effectively reduce data noise and improve information 
utilization rate, and can provide more reliable data support for vegetation ecological monitoring work.

Key words: hyperspectral, spectral dimension reduction, texture feature, vegetation index, support vector machine