面向卫星云图云分类的自适应模糊支持向量机

发布时间：2018-05-30 03:01

本文选题：模糊支持向量机 + 隶属度函数　；参考：《武汉大学学报(信息科学版)》2017年04期

【摘要】：云类识别是实现卫星云图自动分析的基础,针对卫星云图易受噪声干扰且不同云系往往相互交叠的特点,构造一种面向云类识别的自适应模糊支持向量机。该方法不仅改进了隶属度函数的表现形式,而且通过定义控制临界隶属度和隶属度衰减趋势的参数,使隶属度能根据不同云系样本的具体分布特性自适应调整,解决了传统模糊支持向量机的隶属度函数难以反映样本分布的问题。在MTSAT卫星云图上的实验结果表明,通过提取云图可见光通道的反照率、红外通道的亮温及三种亮温差作为云图的光谱特征,并结合统计纹理特征,所构造的自适应模糊支持向量机分类器能有效区分晴空区、低云、中云、高云及直展云;云类识别准确率优于标准支持向量机和传统模糊支持向量机,且具有更强的稳定性和自适应性。
[Abstract]:Cloud recognition is the basis of automatic analysis of satellite cloud images. In view of the characteristics that satellite cloud images are susceptible to noise interference and different cloud systems often overlap with each other, an adaptive fuzzy support vector machine for cloud recognition is constructed. This method not only improves the form of membership function, but also adaptively adjusts membership according to the specific distribution characteristics of different cloud samples by defining the parameters controlling the critical membership and the attenuation trend of membership. It solves the problem that the membership function of traditional fuzzy support vector machine can not reflect the distribution of samples. The experimental results on MTSAT satellite cloud images show that by extracting the albedo of visible light channels of cloud images, the brightness temperature of infrared channels and three kinds of bright temperature differences are taken as spectral features of cloud images, and combined with statistical texture features, The adaptive fuzzy support vector machine classifier can effectively distinguish clear sky region, low cloud, middle cloud, Gao Yun and straight cloud, and the accuracy of cloud recognition is better than that of standard support vector machine and traditional fuzzy support vector machine. And has stronger stability and self-adaptability.
【作者单位】：宁波大学信息科学与工程学院;
【基金】：国家自然科学基金(61271399,61471212) 宁波市国际合作项目(2013D10011) 宁波市自然科学基金(2011A610192,2013A610055) 浙江省信息与通信工程重中之重学科项目(XKXL1425,XKXL1306)~~
【分类号】：TP391.41;TP18

【相似文献】