| 基于多尺度纹理分类及矿物识别的ASTER地质填图 |
| |
| 引用本文： | 唐淑兰.基于多尺度纹理分类及矿物识别的ASTER地质填图[J].地质科技通报,2022,41(3):311-320. |
| |
| 作者姓名： | 唐淑兰 |
| |
| 基金项目： | 中国地质调查局项目DD20190364中国地质调查局项目202009000000180703西安财经大学科学研究扶持计划项目21FCJH008 |
| |
| 摘 要： | 遥感技术已经成为基础地质调查必不可少的手段。为提高地质填图效率及精度, 本研究提出了基于ASTER的岩性自动分类加主要和典型造岩矿物识别的填图方法。首先, 对ASTER数据进行主成分变换, 对选取的第一主分量采用Haar小波进行多尺度小波分解, 将小波系数的统计特征作为纹理特征, 构建纹理及光谱多维特征空间; 接着, 运用支持向量机(SVM)进行岩性分类; 同时, 根据光谱特征提取主要造岩矿物; 最后将主要造岩矿物叠加在分类结果上, 结合野外调查背景进行岩性填图。混淆矩阵结果显示光谱+小波纹理分类精度可以达到83.496 2%, 较光谱+灰度共生矩阵纹理分类精度提高了2.675 6%, 较光谱特征分类精度提高了6.318 9%。与最大似然法(MLC)分类相比, SVM分类精度提高了6.623 7%。矿物提取结果表明, 构造的提取指数可有效提取白云母、黑云母、方解石、角闪石等矿物。野外工作证明, 填图结果与野外调查结果的相关系数为0.7, 可见, 基于ASTER数据利用图像处理技术、机器学习算法及波段运算可作为植被覆盖较少地区有效的地质填图手段。
|
| 关 键 词： | 地质填图 小波变换 多尺度纹理 支持向量机 矿物识别 ASTER |
| 收稿时间： | 2021-12-11 |
Aster geological mapping based on multi-scale texture classification and mineral recognition |
| |
| Abstract: | Remote sensing technology has become an indispensable means in geological survey. In order to improve the efficiency and accuracy of geological mapping, a method based on Aster automatic lithology classification combined with the identification of main rock forming minerals is proposed in this study. Firstly, the principal component transform of ASTER data is carried out, the first principal component is selected for multi-scale Haar wavelet decomposition, and the statistical characteristics of wavelet coefficients are taken as texture features to construct multi-dimensional feature space of texture and spectrum; Then, support vector machine is adopted to classify lithology; At the same time, the main rock forming minerals are extracted according to the spectral characteristics; Finally, the main rock forming minerals are superimposed on the classification results, and the lithology mapping is completed in combination with the field investigation background. The confusion matrix results show that the classification accuracy of spectrum- wavelet texture can reach 83.496 2%, which is 2.675 6% higher than that of spectrum-gray level co-occurrence matrix texture classification and 6.3189% higher than that of spectral feature classification. Compared with the maximum likelihood classification method, the classification accuracy of SVM is improved by 6.623 7%. The mineral extraction results indicate that the extraction index of structure can effectively extract muscovite, biotite, calcite, amphibole and other minerals. It can be seen that image processing technology, machine learning algorithm and band operation can be used as effective means of remote sensing mapping in areas with less vegetation coverage. |
| |
| Keywords: | |
|
| 点击此处可从《地质科技通报》浏览原始摘要信息 |
|
点击此处可从《地质科技通报》下载全文 |
|
