中文摘要：

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8．11mg·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．

英文摘要：

Taking the Qihe County in Shandong Province of East China as the study area, soil sam- ples were collected from the field, and based on the hyperspectral reflectance measurement of the soil samples and the transformation with the first deviation, the spectra were denoised and com- pressed by discrete wavelet transform （DWT） , the variables for the soil alkali hydrolysable nitrogen quantitative estimation models were selected by genetic algorithms （GA）, and the estimation mod- els for the soil alkali hydrolysable nitrogen content were built by using partial least squares （PLS） regression. The discrete wavelet transform and genetic algorithm in combining with partial least squares （DWT-GA-PLS） could not only compress the spectrum variables and reduce the model var- iables, but also improve the quantitative estimation accuracy of soil alkali hydrolysable nitrogen con- tent. Based on the 1-2 levels low frequency coefficients of discrete wavelet transform, and under the condition of large scale decrement of spectrumvariables, the calibration models could achieve the higher or the same prediction accuracy as the soil full spectra. The model based on the second level low frequency coefficients had the highest precision, with the model predicting R2being 0.85,the RMSE being 8. 11 mg. kg-1 , and RPD being 2.53, indicating the effectiveness of DWT-GA- PLS method in estimating soil alkali hydrolysable nitrogen content.