论文部分内容阅读
以山东齐河县为研究区,实地采集土壤样本,在土样高光谱测试并进行一阶导数变换的基础上,先运用离散小波变换(DWT)对土壤光谱去噪降维,然后采用遗传算法(GA)筛选土壤碱解氮定量估测模型的参与变量,最后应用偏最小二乘(PLS)回归构建土壤碱解氮含量的估测模型.结果表明:离散小波变换结合遗传算法和偏最小二乘法(DWT-GA-PLS)用于土壤碱解氮含量定量估测,不仅可压缩光谱变量、减少模型参与变量,而且可改善模型估测准确度;较之于采用土壤全谱,小波离散分解1~2层低频系数构建的模型在参与变量大幅减少的情况下,取得更准确或与之相当的预测结果,其中,基于第2层小波低频系数采用GA筛选变量构建的PLS模型的预测效果表现最好,预测R2达到0.85,RMSE为8.11 mg·kg-1,RPD为2.53.说明DWT-GA-PLS用于土壤碱解氮含量高光谱定量估测的有效性.
Taking Qixian County of Shandong Province as the research area, the soil samples were collected on the spot. Based on the hyperspectral tests of soil samples and the first derivative transformation, the soil spectra were first denoised and reduced by using the discrete wavelet transform (DWT), then the genetic algorithm (GA) was used to select the participatory variables of the quantitative estimation model of soil available nitrogen (NAP). Finally, PLS regression was used to build an estimation model of soil alkali-hydrolyzable nitrogen content.The results showed that the discrete wavelet transform combined with genetic algorithm and partial least- Multiplication (DWT-GA-PLS) can be used to quantitatively estimate soil alkali-hydrolyzable nitrogen content, which can not only compress the spectral variables and reduce the model participation variables, but also improve the accuracy of model estimation. Compared with using soil full spectrum, The models constructed with low-frequency coefficients from 1 to 2 layers obtain more accurate or comparable prediction results with the participation variable greatly reduced. Among them, the prediction results of PLS models constructed with GA-based screening variables based on the second-layer wavelet low-frequency coefficients The best R2R was 0.85, the RMSE was 8.11 mg · kg-1 and the RPD was 2.53, indicating the validity of DWT-GA-PLS for hyperspectral quantitative estimation of soil available nitrogen.