摘要
云雾分离是浓雾遥感监测的难点,地物光谱信息和图像纹理信息的综合利用,分形理论和BP神经网络技术的应用,使夜间云雾分离结果更为可信,基于灰度连通域的图像纹理提取提高了云雾边界的识别能力,灰度加权拉伸后的分数维增强了云雾的可分性,与传统最大似然法比较,本文所用方法对晴空地表、雾区、云区的识别精度均有提升,特别是云区的识别精度提高了10%,三类地表的总体识别率提高了7%,达到93%以上,文章最后对类的归并作了讨论。
The nodus of remote sensing monitoring for fog is the separation of cloud ahd fog. Synthetically using the ground-object spectral information and the image-textural information, together with the applications of fractal theory and BP neural network, have increased the reliability of the cloud and fog separation result. The image-textural extraction basing on gray scale connected region has improved the recognising ability of the cloud and fog boundary, while the fractional dimension weighted by gray scale has improved the separability of cloud and fog. Compared with the traditional Maximum Likelihood Classification, the identified precision of clear sky ground, fog areas, cloud areas was increased, especially that of the cloud areas was increased by 10% , and that of the three kind objects in total was increased by 7% , so it was more than 93%. The merging of these kinds was discussed also.
出处
《遥感学报》
EI
CSCD
北大核心
2006年第4期497-501,共5页
NATIONAL REMOTE SENSING BULLETIN
基金
四川省计委项目"四川省农业气象决策咨询服务平台建设"资助
关键词
BP神经网络
分形理论
云雾分离
分类后处理
BP neural network
fractal theory
cloud and fog separate
post-class processing
