基于PLSR和LSSVM模型的土壤水分高光谱反演  

Hyperspectral inversion of soil moisture based on PLSR and LSSVM models

作　　者：刘英[1,2] 范凯旋 裴为豪 沈文静 葛建华 LIU Ying;FAN Kaixuan;PEI Weihao;SHEN Wenjing;GE Jianhua(School of Earth and Environment,Anhui University of Science and Technology,Huainan 232001,China;The Anhui Province Engineering Laboratory of Water and Soil Resources Comprehensive Utilization and Ecological Protection in High Groundwater Mining Area,Huainan 232001,China)

机构地区：[1]安徽理工大学地球与环境学院,安徽淮南232001 [2]安徽省高潜水位矿区水土资源综合利用与生态保护工程实验室,安徽淮南232001

出　　处：《矿业安全与环保》2024年第5期147-153,共7页Mining Safety & Environmental Protection

基　　金：国家自然科学基金青年项目(52204181);安徽理工大学校级项目(xjzd2020-04);安徽理工大学青年教师科学研究基金项目(XCZX2021-02)。

摘　　要：为对地下采矿扰动区表层土壤水分进行反演,以大柳塔煤矿52501工作面为例,利用无人机搭载成像光谱仪获取高光谱影像,对获取的光谱数据进行对数、倒数对数、一阶和包络线去除变换,结合地面采集的128个土壤水分数据,基于偏最小二乘回归(PLSR)和最小二乘支持向量机(LSSVM)构建土壤水分预测模型并验证其预测精度。结果表明,基于一阶变换的PLSR模型和LSSVM模型预测精度相对较好,一阶变换的PLSR模型建模集R^(2)_(c)和预测集R^(2)_(p)分别为0.7021和0.6405,均方根误差RMSE_(c)和RMSE_(p)分别为1.6384%和1.1034%,相对分析误差RPD_(p)为1.7263;一阶变换的LSSVM模型建模集R^(2)_(c)和预测集R^(2)_(p)分别为0.8125和0.5979,均方根误差RMSE_(c)和RMSE_(p)分别为1.2755%和1.3459%,相对分析误差RPD_(P)为1.6323。最终基于PLSR和LSSVM模型完成了土壤水分的制图,实现了土壤水分的空间预测,为该研究区植被引导修复中土壤水分精准提升提供了空间数据支持。In order to invert the surface soil moisture in the underground mining disturbance area,taking the 52501 working face of Daliuta Coal Mine as an example.An unmanned aerial vehicle(UAV)equipped with an imaging spectrometer was used to acquire hyperspectral images.The obtained spectral data underwent logarithmic,reciprocal logarithmic,first-order,and envelope removal transformations.Combining these data with 128 ground-collected soil moisture samples,partial least squares regression(PLSR)and least squares support vector machine(LSSVM)models were constructed to predict soil moisture content and validate their predictive accuracy.The results indicate that the PLSR model and LSSVM model based on the first-order transformation exhibit relatively good predictive accuracy.For the first-order transformation PLSR model,the coefficient of determination for the modeling set R^(2)_(c)is 0.7021,and for the prediction set R^(2)_(p)is 0.6405,with root mean square errors of calibration(RMSE_(c))and prediction(RMSE_(p))being 1.6384%and 1.1034%,respectively,and relative prediction error(RPD_(p))of 1.7263.For the first-order transformation LSSVM model,the modeling set R^(2)_(c)is 0.8125,and the prediction set R^(2)_(p)is 0.5979,with RMSE_(c)and RMSE_(p)being 1.2755%and 1.3459%,respectively,and RPD_(p)of 1.6323.Ultimately,based on the PLSR and LSSVM models,soil moisture mapping was completed,achieving spatial prediction of soil moisture.This provides spatial data support for the precise enhancement of soil moisture in vegetation-guided restoration efforts in the study area.

关 键 词：土壤含水量 高光谱 偏最小二乘回归 最小二乘支持向量机 无人机 干旱阈值 引导修复 

分 类 号：TD88[矿业工程—矿山开采] X14[环境科学与工程—环境科学] TP79[自动化与计算机技术—检测技术与自动化装置]