| 引用本文: | 孙宇乐,屈忠义,刘全明.基于地面光谱联合SAR多源数据的农田表土氮磷监测[J].灌溉排水学报,0,():-. |
| Sun Yu-le,Qu Zhong-yi,Liu Quan-ming.基于地面光谱联合SAR多源数据的农田表土氮磷监测[J].灌溉排水学报,0,():-. |
|
| 摘要: |
| 【目的】建立更为精确的高光谱预测模型，以便更加精准地、快速地测定土壤表土氮磷量，进一步推动多源遥感技术在现代化高标准农业生产与管理中的应用和发展。【方法】以内蒙古河套灌区解放闸灌域为试验区，利用地面实测光谱反射率，联合C波段微波雷达SAR（Synthetic Aperture Radar）四极化后向散射数据，通过对土壤养分特征波段的选择，建模评价土壤氮磷量。首先利用光谱反射率，及其对数、一阶与二阶导数4种光谱数据，进行相关性分析而滤选获取了与氮磷相关系数均大于0.4的近红外1 480、2 050、2 314 nm等特征波段，同时利用1~8层小波分析与重构图谱技术去除噪声，排除特异值干扰。小波去噪后找到相关性强的特征波段，结合SAR后向散射系数，与养分做回归及神经网络输入，形成神经网络模型。【结果】通过对高光谱数据的小波分解和重构，能够有效提高反射率及其3种变换形式与土壤养分的相关性，尤其是低频分量的1~3层、高频分量的4~6层的效果更好。反射率一阶导数的神经网络模型为最佳预测模型，其对土壤氮、磷养分含量的预测R2分别为0.749 6、0.759 2，均方差RMSE均为0.110 2，其模型的稳定性和预测精度优于多元线性回归模型。【结论】采用光谱联合SAR可以更好地快速预测土壤全氮、全磷。 |
| 关键词: 土壤养分;遥感;高光谱;模型;小波变换;神经网络 |
| DOI： |
| 分类号:S152. 7；P628+.2 |
| 基金项目:国家重大科技专项(2016YFC0501301)；国家自然科学(51569018)；内蒙古农业大学“双一流”学科创新团队建设人才培育项目(NDSC2018-10) |
|
| Monitoring of Nitrogen and Phosphorus in Farmland Topsoil based on Multi-source Data of Ground Spectrum Combined with SAR |
|
Sun Yu-le, Qu Zhong-yi, Liu Quan-ming
|
|
College of Water Conservancy and Civil Engineering,Inner Mongolia Agricultural University
|
| Abstract: |
| 【Objective】To establish a more accurate hyperspectral prediction model for more accurate and rapid determination of nitrogen and phosphorus content in soil topsoil, and further promote the application and development of multi-source remote sensing technology in modern and high-standard agricultural production and management.【Method】Based on the experimental area of Jiefang gate irrigation in Hetao irrigated area in Inner Mongolia, the measured spectral reflectance on the ground was used in combination with four-polarization backscattering data in C-band microwave Radar SAR (Synthetic Aperture Radar). Based on the selection of characteristic bands of soil nutrients, soil nitrogen and phosphorus quantities were modeled and evaluated.First using the spectral reflectance, and the logarithmic, first and second order derivative 4 kinds of spectral data, correlation analysis and filter selected to obtain the coefficient associated with the nitrogen and phosphorus were greater than 0.4 of near infrared 1, 2, 480, 050, 2 314 nm band at the same time using 1 ~ 8 layer get rid of the noise mapping technology, wavelet analysis and reconstruction to exclude specific value interference.After wavelet denoising, the characteristic bands with strong correlation are found, and a neural network model is formed by combining SAR backscattering coefficient and nutrients as regression and neural network input.【Result】Through the wavelet decomposition and reconstruction of hyperspectral data, the correlation between reflectance and its three transformation forms and soil nutrients can be effectively improved, especially the effect of 1~3 layers of low-frequency component and 4~6 layers of high-frequency component is better.The neural network model with the first derivative of reflectivity was the best prediction model, and its R2 and RMSE of soil nitrogen and phosphorus nutrient contents were 0.749 6 and 0.759 2 respectively, and the mean square error was 0.110 2, respectively. The stability and prediction accuracy of the model were better than that of the multiple linear regression model.【Conclusion】Combined spectral SAR can be used to predict soil total nitrogen and total phosphorus more quickly. |
| Key words: Soil nutrients; remote sensing; hyperspectral; model; wavelet transform; neural networks |
