多源遥感影像湿地检测概率潜在语义分析

发布时间：2018-03-31 02:14

  本文选题：概率潜在语义分析　切入点：湿地检测　出处：《测绘学报》2017年08期

【摘要】：提出了一种基于概率潜在语义分析的多源遥感影像湿地检测方法。首先提取高分辨率影像的光谱、纹理和湿地场景的地物组成成分,并结合由多光谱遥感数据提取的湿地地表温度、土壤含水量,组成湿地场景的特征空间;然后利用概率潜在语义分析将湿地场景表示成多个潜在语义的组合,并用潜在语义的权值向量来描述湿地场景的特征空间;最后利用SVM分类器实现湿地场景的检测。试验表明,概率潜在语义分析能够将湿地的高维特征空间映射到低维的潜在语义空间中,地物组成成分和定量环境特征的加入能更加有效地表征湿地特征空间,提高湿地检测精度。
[Abstract]:A multi-source remote sensing image wetland detection method based on probabilistic latent semantic analysis is proposed. Combined with the wetland surface temperature and soil water content extracted from multispectral remote sensing data, the wetland scene is represented as a combination of multiple potential semantics by using probabilistic latent semantic analysis. The potential semantic weight vector is used to describe the feature space of the wetland scene. Finally, the SVM classifier is used to detect the wetland scene. Probabilistic latent semantic analysis can map the high-dimensional feature space of wetland to low-dimensional potential semantic space. The addition of the composition of ground objects and quantitative environmental features can more effectively represent the wetland feature space and improve the accuracy of wetland detection.
【作者单位】： 中国地质大学(武汉)信息工程学院;武汉大学遥感信息工程学院;
【基金】：国家重点研发计划(2016YFB0502603) 地理国情监测国家测绘地理信息局重点实验室开放基金(2016NGCM09)~~
【分类号】：P237
，

本文编号：1688701