基于遗传算法的土壤质地高光谱预测模型研究

Hyperspectral Prediction Modeling of Soil Texture Based on Genetic Algorithm

为快速、准确地获取土壤质地信息,利用遗传算法结合偏最小二乘法(GA-PLS)回归建立土壤质地预测模型。采集了丰乐河流域162个表层土样,在实验室内对土样进行质地分析和光谱测量,采用遗传算法(Genetic Algorithm)筛选土壤质地光谱特征波段,在此基础上运用偏最小二乘法(PLS)构建了土壤质地预测模型,并与全谱段PLS模型进行对比分析。结果表明:基于遗传算法结合偏最小二乘的模型验证精度高于全谱段PLS模型,粉粒光谱验证集R^2达到0.841,RPD为2.391,较全谱段PLS模型RPD提高了18.13%,提升效果显著;砂粒光谱验证集的R^2为0.721,RPD为2.142,较全谱段PLS模型RPD提高了10.41%。遗传算法结合偏最小二乘法(GA-PLS)在土壤质地高光谱估测中,压缩了光谱变量,减少了数据冗余,提高了模型预测精度。

In order to obtain soil texture information quickly and accurately, the prediction model of soil texture was established by using the methods of genetic algorithm（GA） and partial least squares（PLS）. Taking 162 surface soil samples in Fengle River Basin as the research objects, soil texture and hyperspectral reflectance of samples were tested in laboratory. GA was used to select spectral bands of soil texture. With the base, PLS regression model of soil texture was established, and comparison and analysis of the model were built with the full spectral band. The results showed that GA-PLS model was more accurate than the full-spectrum model. The R^2 of silt model validation reached0.841 and its RPD reached 2.391, 18.13% higher than that of the full spectrum. The R^2 of sand model validation reached 0.721 and its RPD reached 2.142, 10.41% higher than that of the full spectrum. GA-PLS can be used to predict soil texture, which not only compresses the spectral variables, reduces data redundancy, but also improves the prediction accuracy of the model. Therefore, the method of GA-PLS could be used to effectively estimate the hyperspectral soil texture content.