基于FDCT与ELM的遥感影像湿地类型分类——以黄家湖国家湿地公园为例  

Method for automatic identification of wetland types using FDCT and ELM ——Taking Huangjiahu National Wetland Park as an example

作　　者：辛动军[1] 袁梦 陈建安 钟旭 臧艺元 胡平 钟国忠 唐鼐[1] 王传立[1] XIN Dongjun;YUAN Meng;CHEN Jianan;ZHONG Xu;ZANG Yiyuan;HU Ping;ZHONG Guozhong;TANG Nai;WANG Chuanli(Central South University of Forestry and Technology, Changsha 410004, Hunan, China;Management Office of Huangjiahu National Wetland Park, Yiyang 413000, Hunan, China)

机构地区：[1]中南林业科技大学,湖南长沙410004 [2]黄家湖国家湿地公园管理处,湖南益阳413000

出　　处：《中南林业科技大学学报》2018年第6期30-35,共6页Journal of Central South University of Forestry & Technology

基　　金：国家湿地保护补助资金项目(林湿发[2011]273号);长沙市科技计划项目(k1508007-11);国家自然科学基金项目(31570627);湖南省教育厅优秀青年项目(14B193)

摘　　要：以黄家湖国家湿地公园为例,利用高分辨率遥感影像数据,研究提出了一种基于FDCT与ELM的湿地类型分类方法。该方法利用湿地遥感影像的光谱特征、纹理特征、空间特征作为ELM的输入数据,将遥感影像进行FDCT变换分解,获得影像的高频曲波变换系数和低频曲波变换系数,选取高频部分曲波变换系数组合作为地物的纹理特征;采用连续可微的Sigmod函数作为学习函数。湿地类型识别实验结果表明,该方法能够快速实现湿地类型的自动分类,总体分类精度达到了86.7%,高于传统SVM方法 76.5%的分类精度,Kappa系数超过0.83,为湿地景观遥感动态监测奠定了较好的研究基础。Taking Huangjia lake national wetland park as an example, we used data of high resolution remote sensing images, and presented a type of wetland classification method based on FDCT and ELM. The method used the spectral feature, texture feature and spatial feature of the wetland remote sensing image as the input data of ELM and then used FDCT to transform and decompose the remote sensing image so that we can obtain the low-frequency curvelet transform coefficients and the high-frequency curvelet transform coefficients of the images, high frequency portion curvelet coefficient combination were selected as a characteristic feature of the texture and used continuously differentiable Sigmoid function as a learning function. The experimental results showed that the method can quickly realize the automatic classification of wetland types. The overall classification accuracy is 86.7%, higher than the traditional SVM method 76.5% classification accuracy and the Kappa coefficient reached greater than 0.83. This method can lay a foundation for the study of automatic identification of wetland types.

关 键 词：快速离散曲波变换 湿地类型 极限学习机 自动分类 

分 类 号：S762.3[农业科学—森林保护学]