| 结合小波包变换和随机森林的ASTER蚀变信息提取 |
| |
| 引用本文： | 唐淑兰,曹建农,王国强,卜涛.结合小波包变换和随机森林的ASTER蚀变信息提取[J].地质学报,2021,95(3):924-933. |
| |
| 作者姓名： | 唐淑兰 曹建农 王国强 卜涛 |
| |
| 作者单位： | 长安大学地球科学与资源学院,西安,710054;西安财经大学管理学院,西安,710100;长安大学地球科学与资源学院,西安,710054;中国地质调查局西安地质调查中心,西安,710054 |
| |
| 基金项目： | 本文为中国地质调查局项目（编号DD20190812）、国家自然科学基金项目（编号41571346）、陕西省教育厅项目（编号18JK0317）、陕西省自然科学基础研究计划项目(编号2020JM- 585)、西安财经大学项目（编号16FCJH05）资助的成果。 |
| |
| 摘 要： | 为了更准确地提取蚀变信息,本文选择新疆、甘肃和内蒙古三省交界部位为研究区,结合小波包变换和随机森林提取ASTER蚀变信息。首先,选择主要蚀变类型的诊断性波段进行特征向量主成分分析,得到主分量影像;接着,对主分量影像进行小波包变换,使用代价函数选择最优小波包树,并提取高低频信息构造分类向量;然后,经过特征筛选构造随机森林分类模型,并提取矿化蚀变信息;最后,通过野外采样、薄片鉴定对提取结果进行精度评价。铁染、Al-OH及Mg-OH蚀变信息的主成分分析波段组合分别选择Band 1、2、3、4,Band 1、3、4、6及Band 1、5、8、9。结果表明,本文方法提取铁染、Al-OH基团及Mg-OH基团蚀变信息的总体精度可达到88.7443、85.5469及91.7594,Kappa分别为0.7767、0.6732及0.8362,与成矿区带及已有的该区域的成矿特征相关性较好。本研究采用的最优小波包树能充分利用矿物光谱的能量特征,随机森林可削弱矿物组分的噪声干扰,研究结果可为遥感蚀变信息提取提供技术参考。
|
| 关 键 词： | ASTER 蚀变信息 小波包变换 随机森林 主成分分析 |
| 收稿时间： | 2020/8/6 0:00:00 |
| 修稿时间： | 2020/9/25 0:00:00 |
Aster alteration information extraction based on wavelet packet transform and random forest |
| |
| TANG Shulan,CAO Jiannong,WANG Guoqiang,BU Tao.Aster alteration information extraction based on wavelet packet transform and random forest[J].Acta Geologica Sinica,2021,95(3):924-933. |
| |
| Authors: | TANG Shulan CAO Jiannong WANG Guoqiang BU Tao |
| |
| Institution: | (School of Earth Science and Resources,Chang'an University,Xi'an 710054;School of Management,Xi'an University of Finance and Economics,Xi'an 710100;Xian Center of Geological Survey,CGS,Xi'an 710054) |
| |
| Abstract: | In order to extract alteration information more accurately, this paper selects Xinjiang, Gansu and Inner Mongolia as the research area, and extracts aster alteration information by combining wavelet packet transform and random forest. Firstly, the diagnostic wavebands of main alteration types are selected for feature vector principal component analysis to obtain principal component images. Then, the principal component image is transformed by wavelet packet, and the optimal wavelet packet tree is selected by using the cost function, and the high and low frequency information is extracted to construct the classification vector. Then, the random forest classification model is constructed through feature screening, and the mineralization and alteration information is extracted. Finally, the accuracy of the extraction results is evaluated by field sampling and thin section identification. The principal component analysis band combinations of iron stain, Al- OH and Mg- OH alteration information are Band 1, 2, 3, 4, Band 1, 3, 4, 6 and Band 1, 5, 8 and 9, respectively. The results show that the overall accuracy of the method is 88.7443, 85.5469 and 91.7594, and kappa is 0.7767, 0.6732 and 0.8362, respectively. It has a good correlation with the metallogenic characteristics of the metallogenic belt and the existing area. The optimal wavelet packet tree used in this study can make full use of the energy characteristics of mineral spectrum, and random forest can weaken the noise interference of mineral components. The research results can provide technical reference for remote sensing alteration information extraction. |
| |
| Keywords: | aster alteration information wavelet packet transform random forest principal component analysis |
| 本文献已被 维普 万方数据 等数据库收录！ |
| 点击此处可从《地质学报》浏览原始摘要信息 |
| 点击此处可从《地质学报》下载免费的PDF全文 |
|
