军事知识服务平台
基于分数阶微分光谱指数和潜在变量的土壤盐分定量估算
Quantitative estimation of soil salinity based on fractional order differentiation spectral indices and latent variables
摘要 为解决土壤盐渍化制约农业生产与可持续发展的问题,高效、精准地获取土壤盐分信息是保障农田土壤可持续利用的关键环节之一.试验土壤采集自果园,通过实验室制备不同含盐量梯度土壤样本,以电导率表征土壤盐分状况,结合高光谱信息定量估算土壤含盐量.原始高光谱数据经0~2.0 阶(步长为0.1)精细化分数阶微分(FOD)变换,基于微分变换后的高光谱数据完成光谱指数构建和潜在变量提取,并进一步构建光谱信息数据集,应用随机森林(RF)、极限学习机(ELM)和偏最小二乘回归(PLSR)方法构建土壤含盐量估算模型.结果表明,FOD 技术能够在一定程度上增强光谱数据与电导率之间的相关性,微分变换后的各阶光谱与土壤电导率的相关系数呈先上升后下降趋势;基于 FOD 变换后数据构建的光谱指数与电导率的相关系数均低于基于原始光谱(0 阶)构建的光谱指数,其中相关系数最大值为 0.790;在自变量输入相同的情况下,ELM方法取得了最佳建模效果;以最优光谱指数和潜在变量组成的数据集建模效果最佳.当微分阶数为0.2 时,ELM的建模效果最优,训练集 R2 为 0.828,RMSE 为 437 μS/cm;验证集 R2 为0.822,RMSE为 511 μS/cm,RPD 为 2.034.研究结果可为提高土壤含盐量反演精度和发展精准农业提供理论参考.
Abstract:
To address the problem of soil salinization hindering agricultural production and sustainable development,obtaining soil salinity information efficiently and accurately is one of the key steps in en-suring the sustainable utilization of farmland soil.Soil samples were collected from orchards,and soil samples with different salt concentrations were prepared in the laboratory.The soil salinity status was characterized by electrical conductivity,and soil salt content was quantitatively estimated by combining the hyperspectral information.The original hyperspectral data were refined by fractional order differen-tiation(FOD)transformation from 0 to 2.0 orders(with a step size of 0.1).Based on the hyperspectral data after the differential transformation,spectral indices and latent variables were constructed and ex-tracted.Furthermore,a spectral information dataset was constructed,and soil salinity estimation models were built using random forest(RF),extreme learning machine(ELM),and partial least squares regression(PLSR)methods.The results indicate that FOD technology can enhance the corre-lation between spectral data and soil electrical conductivity to a certain extent.The correlation coeffi-cients between the processed spectra of each order after differential transformation and soil electrical conductivity first increase and then decreased.The correlation coefficients between the spectral indices and conductivity constructed based on the FOD-transformed data are all lower than those of the spectral indices constructed based on the original spectra(order 0),with the highest correlation coefficient be-ing 0.790.Among the evaluated models,the ELMmodel demonstrates superior performance under the same input variable conditions.The dataset consisting of the optimal spectral indices and latent varia-bles yields the optimal predictive accuracy.Specifically,when the differentiation order is set to 0.2,the ELMmodel achieves the best modeling performance,achieving an R2 of 0.828 and an RMSE of 437 μS/cm on the training set,as well as an R2 of 0.822,an RMSE of 511 μS/cm,and an RPD of 2.034 for the validation set.The research results can provide a theoretical reference for improving the accuracy of soil salt content inversion and for precision agriculture.
刊 名 排灌机械工程学报 ISTICPKU
年,卷(期): 2026, 44(10)
分类号: S127
机标分类号: S127 TP79 S562
在线出版日期: 2026年11月6日
基金项目: 52569008:国家自然科学基金52209056:国家自然科学基金202501AT070378:云南省基础研究计划项目202401AT070325:云南省基础研究计划项目KKXX202523082:云南省兴滇英才支持计划,