SPA 与 GA 土壤 EC 检测模型波长选择能力对比

摘要：土壤电导率（（ EC）是评价土壤盐分含量的重要指标，光谱检测法是其中一种有效的方法，本研究旨在比较连续投 影算法（（ SPA）和遗传算法（（ GA）对敏感波长的选取能力并建立高光谱反演模型。首先，将140个土壤样品带回实验室， 测定EC（（ 土：水=1：5，EC1：5）和光谱反射率（（ 900～1700nm）。其次，利用SPA和GA选取敏感波长以降低模型的复 杂度，然后采用PLSR方法建立检测模型。最后通过判定系数（（ R2）、均方根误差（（ RMSE）、预测离差比（（ RPD）和选取 的波长个数对各类方法进行评价。试验结果表明，采用SPA方法建立的预测模型优于GA方法，原因是SPA具有波段数 和时间较少、R2和RPD较高、RMSE较小的特点（（ 波段数=5，R2 =0.9707，RPD=8.4445，RMSE=0.31178），验证了SPA 方法可用于筛选敏感波长，从而提高土壤EC模型精度。

期刊：Materials Science and Engineering

DOI：10.1088/1757-899X/392/4/042037

设备：GaiaSky

The Contrast of Selecting Wavelength Capability of SPA and GA for Soil EC Detecting Model

Abstract：The electrical conductivity(EC) of the soil was an important index to evaluate soil salinity. Spectral detecting method was an effective method. The objectives of this study were to compare choosing ability of the successive projections algorithm (SPA) and Genetic Algorithm (GA) methods for sensitive wavelengths and establish hyper spectral inversion model. Firstly, 140 soil samples were taken back to the laboratory and measured the EC (soil: water=1:5, EC1:5) and spectral reflectance(900-1700nm). Secondly, the SPA and GA were used to select sensitive wavelengths for reducing the complexity of the model, then PLSR method was used to establish detecting model. Finally, the determination coefficient (R2 ), root mean-square error (RMSE), ratio prediction to deviation (RPD) and the number of wavelengths selected were used to evaluate different kinds of methods. The experimental results showed the prediction models after using SPA were better than GA method because SPA had lesser band number and time, higher R2 and RPD, smaller RMSE(band number=5, R2 =0.9707,RPD=8.4445 and RMSE=0.31178). Thus, it was validated the SPA method can be used to select sensitive wavelengths to improve the model accuracy of soil EC.