遥感影像浅水河道提取二维经验模态分解方法

The Application of Shallow Water Channel Extraction in Remote Sensing Using Two-dimensional EMD Method

遥感影像中浅水河道与居民地具有相似的光谱特性,在浅水河道自动提取过程中噪声较多,经验模态分解(EMD)可获取原始信号不同尺度上的细节信息,有效地提高遥感影像浅水河道自动提取的精度。利用环境与减灾小卫星数据,以彰武县柳河为研究区,对该区2012年10个时期NDVI时间序列分别EMD分解,选取每个时相信息量较大的前三个固有模态函数(IMF),结合2012年9月18号影像共34波段作为研究数据,利用极大似然法、BP神经网络传统的分类方法进行分类。结果表明结合EMD的分类方法可有效地去除居民地噪声,Kappa系数最高达到0.9690,总体分类精度最高达到93.35%,从而有效地解决了遥感影像中浅水河道提取正确率低的难题。

Remote-sensing images between the shallow water channel and residents have similar spectral pattern, there are serious fault classifications in the process of automatically extracting, the method named empirical mode decomposition （EMD） can be used to details to obtain different scales of the original signal. EMD can effectively improve the accuracy of shallow water channel automatic extraction. Based on environment and disaster Mitigation Small Satellite datas of Liu River in Zhangwu. First of all, the 10 periods NDVI time series of 2012 with the EMD method should be decomposed, and then select the first three intrinsic mode function （IMF） of each phase information which include more information, coupled with the images on September 18, 2012, a total of 34 bands as the research data, and finally using Maximum Likelihood method, BP Neural Network method to classify. Results show that the classification of the EMD method can effectively remove the noise of residents, the Kappa coefficient up to 0. 969 0, the overall classification accuracy up to 93.35%, so as to effectively solve the shallow water channel in the remote sensing image to extract the problem of low accuracy.