| 基于t-SNE降维算法的区域化探数据中地质体空间分布信息可视化:以英格兰西南部为例 |
| |
| 引用本文： | 陈军林,闫岩,彭润民.基于t-SNE降维算法的区域化探数据中地质体空间分布信息可视化:以英格兰西南部为例[J].地质科技通报,2021,40(2):186-196. |
| |
| 作者姓名： | 陈军林 闫岩 彭润民 |
| |
| 基金项目： | 国家重点研发计划“深地资源勘查开发”重点专项2016YFC0600502 |
| |
| 摘 要： | 区域化探数据中包含了丰富的地质信息，提取出蕴含在这些数据中的地质体空间分布信息，对于区域地质研究和找矿勘查具有重要意义。区域化探数据通常包括数十个元素，属于高维数据，隐藏在这些高维数据中的地质体空间分布信息无法直接从数据中观察到。针对这个问题，构建了一个基于t分布随机近邻嵌入(t-distributed stochastic neighbor embedding，简称t-SNE)算法的高维区域化探数据降维可视化模型。t-SNE算法是一种非线性降维方法，特别适用于高维数据集的降维和可视化。选择对岩性鉴定比较稳定的元素，通过t-SNE算法将高维化探数据降维到人眼可观察的一、二、三维，把降维之后的变量表达为栅格图，通过三原色混合等方法进行可视化，从而把隐藏在高维化探数据中的地质体空间分布信息可视化表达出来。以英格兰西南部某地区水系沉积物区域化探数据为例进行研究来检验t-SNE算法在高维化探数据可视化上的实际应用效果。结果显示:①通过t-SNE算法对高维化探数据进行可视化得到的结果能够很好地反映研究区的地质体空间分布情况；②可视化的效果与t-SNE算法的目标维度和复杂度两个参数密切相关。在t-SNE算法中设定要降维到的目标维度越高，所显示的地质体信息越详细。③基于t-SNE算法的化探数据降维可视化效果比基于主成分分析(PCA)的化探数据降维可视化效果更好。本文研究表明基于t-SNE算法的化探数据降维可视化方法能够很好地将地质体空间分布信息可视化表达出来，对于推断地质体的空间分布有一定的指导意义。
|
| 关 键 词： | 区域化探 t-SNE算法 降维 可视化 地质空间信息 |
| 收稿时间： | 2020-06-04 |
Visualization of geological spatial distributing information in regional geochemical exploration data based on t-SNE algorithm: A case study of SW England |
| |
| Abstract: | Regional geochemical prospecting data contains a lot of geological information. Extracting the geological spatial distributing information contained in these data is of great significance for regional geological research and mineral prospecting. Regional geochemical data usually includes dozens of elements, which belong to high-dimensional data. Geological spatial distributing information hidden in these high-dimensional data cannot be observed directly from the data. In order to solve this problem, we constructed a dimensionality reduction and visualization model of high-dimensional regional geochemical exploration data based on the t-Distributed Stochastic Neighbor Embedding (t-SNE) algorithm. t-SNE algorithm is a nonlinear dimensionality reduction method, which is especially suitable for dimensionality reduction and visualization of high-dimensional data. Select the elements that are sufficiently stable in lithology identification. Reduce the dimension of high-dimensional geochemical exploration data to 1D, 2D or 3D through the t-SNE algorithm, because the low-dimensional data less than 3D can be observed by human eyes easily. Express the output variables of dimension reduction algorithm as raster files, and visualize them by RGB color mixing and other methods, thus the spatial distribution information of geological bodies hidden in high-dimensional geochemical exploration data can be observed directly. The regional geochemical exploration data of stream sediments in a region of southwest England are taken as an example to evaluate the t-SNE algorithm in visualization of high-dimensional geochemical exploration data. The case study shows that: (1) The high-dimensional geochemical exploration data visualization results through t-SNE algorithm can represent the spatial distribution of geological bodies in the study area very well; (2) The visualization results are tightly related to two parameters: target dimension and perplexity of the t-SNE algorithm. The higher the target dimension was be reduced in the t-SNE algorithm, the more detailed the geological spatial information displayed. (3) The results of dimension reduction and visualization of geochemical exploration data based on the t-SNE algorithm are better than those based on principal component analysis (PCA). The research in this paper shows that the high-dimensional geochemical exploration data visualization method based on the t-SNE algorithm can display the spatial distribution information of geological bodies, which has certain guiding significance for inferring the spatial distribution of geological bodies. |
| |
| 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》下载免费的PDF全文 |
