一种间接从高光谱数据中提取黑土硒含量的新方法
|
摘要
我国东北黑土富含养分,随着土壤数字制图、精确农业和土壤资源调查等研究的深入,引入航空高光谱数据并提供科学的预测结果成为研究热点。硒元素相对于黑土土壤的主要成分属于微量元素,但其对作物的正常生长的作用与大量元素是同等重要的,亦是人体健康所必要的营养元素。针对硒含量反演,建立了一个基于主要成分的间接提取模型,该模型能够显著提升硒含量回归系数,降低实测值与预测值的误差。数据源自CASI-1500航空高光谱成像系统,光谱范围380～1 050 nm,空间分辨率1.5 m。在黑龙江建三江地区采集60个土壤样本,化验获得硒、有机质、全铁、 pH和氧化钙含量数据,选择BP神经网络,建立光谱与含量的反演模型。分析不同含量的黑土成分在可见-近红波段范围内光谱变换规律,掌握了硒元素随着含量升高,光谱反射率会逐步升高的规律。但当硒含量较低时,在其他成分的干扰下,这一规律会逐渐减弱,直至不显著。有机质的光谱特征与硒元素相反,总体上随着含量的增高,反射率整体下降,这与有机质的光谱特性紧密相关。全铁光谱呈现出与有机质光谱类似的规律,说明二者具有较高的相关性。不同pH值和氧化钙含量的光谱特征与检测值没有呈现出明显的特征规律,反射规律不明显。对60个采样点不同养分含量进行逐波段求反射率对养分的相关系数。结果表明, pH值各个波段相关系数最高,均值达到0.63;其次是全铁的相关系数,为0.54;有机质和氧化钙的相关系数接近,分别为0.42和0.47;而硒元素含量与逐波段的平均相关系数最低,为0.38。选取相关系数较高的前5个波段,作为建模波段。硒特征波段为447, 437, 456, 466和475 nm;有机质特征波段为447, 456, 466, 437和475 nm;全铁特征波段为752, 695, 800, 762和733 nm; pH特征波段为905, 752, 800, 943和695 nm;氧化钙特征波段为752, 695, 800, 523和762 nm。通过计算样本点硒含量与其他成分的相关系数,硒与有机质呈正相关,相关系数为0.79;与全铁、 pH、氧化钙呈负相关,相关系数分别为-0.80,-0.94和-0.69。针对有机质、全铁、 pH和氧化钙反演精度较高,而硒元素含量较低,直接反演精度不足的问题,设计了一种先提取4种成分含量,再根据其提取结果建立硒元素函数关系,间接反演硒元素含量的方法。首先将五种成分与特征光谱进行神经网络分析,计算每种成分的回归系数R~2和RMSE。显示全铁和pH具有较高的反演精度,有机质和氧化钙归系数虽低于0.8,但也显著高于硒元素的反演精度。建立硒元素与其他4种成分含量的回归模型,得出Se=0.522 9+0.041 8Som-0.016 6Fe_2O_3-0.035 6pH-0.005CaO,进行硒元素间接提取,回归系数从0.516增长到0.724,均方根误差从0.182降低到0.136,显著改进了反演硒含量的精度,为硒元素大范围精确制图提供了一种新技术。
In the field of soil digital mapping, precision agriculture and soil resource investigation, the study of aerial hyperspectral data to provide scientific prediction results by aerial hyperspectral have become the focus of research, especially in the case of black soil rich in nutrients in Northeast China. Compared with the main components of soil in the black soil, selenium is a trace element, whose effect on the normal growth of crops is as important as a large number of elements, and it is also a necessary nutrient element for human. In this paper, an indirect extraction model based on the main component is created for the retrieval of selenium content. This model can significantly increase the regression coefficient of selenium content and reduce the error between the measured value and the predicted value. The data source is CASI-1500 aerial hyperspectral imaging system with a spectral range of 380～1 050 nm, and a spatial resolution of 1.5 m. 60 soil samples were collected from the Jiansanjiang area of Heilongjiang. The data of selenium, organic matter, total iron, pH and calcium oxide content were obtained. The BP neural network was selected to establish the inversion model of spectrum and content. In addition, the law of spectral change in the visible and near infrared range of different content of black soil composition was analyzed, and the rule that the spectral reflectance would increase gradually as the content of selenium increased. However, when the selenium content was low, the law would gradually weaken until the other components are disturbed. The spectral characteristics of organic matter were opposite to that of selenium. In general, the reflectance decreases as the content increases, which is closely related to the spectral properties of organic matter. The spectra of the total iron showed similar laws with the organic matter spectrum, indicating that the two have high correlation. The spectral characteristics and detection values of different pH values and calcium oxide contents did not show obvious characteristics, and the law of reflection was not obvious. The correlation coefficients of nutrient contents in different nutrient contents of 60 sampling points were obtained by bands. The results show that the correlation coefficient of each band of pH is the highest, the mean value is 0.63, the second is the correlation coefficient of total iron, 0.54, the correlation coefficient of organic matter and calcium oxide is close to 0.42 and 0.47, while the average correlation coefficient of selenium element content and bands is the lowest, which is 0.38. The first 5 bands with higher correlation coefficients are selected as modeling bands. The characteristics of selenium are 447, 437, 456, 466 and 475 nm; the characteristic bands of organic matter are 447, 456, 466, 437 and 475 nm; the characteristic bands of the whole iron are 752, 695, 800, 762 and 733 nm, and the characteristics of pH are 905, 752 and 695 nm. By calculating the correlation coefficient of sample point selenium content and other components, selenium has a positive correlation with organic matter, and the correlation coefficient is 0.79. The correlation coefficient is negatively correlated with total iron, pH and calcium oxide, and the correlation coefficients are-0.80,-0.94 and-0.69, respectively. In view of the high precision of the inversion of organic matter, total iron, pH and calcium oxide, while the content of selenium is low and the accuracy of direct inversion is insufficient, a method of extracting the functional relationship of selenium elements by extracting the content of four components is designed, and the content of selenium elements is indirectly retrieved. First, the five components and characteristic spectra are analyzed by using neural network, and the regression coefficients R~2 and RMSE of each component are calculated. It is concluded that total iron and pH have higher inversion accuracy, while organic matter and calcium oxide coefficient are lower than 0.8, but they are also significantly higher than those of selenium. A regression model for the content of selenium and other four components was obtained, and Se=0.522 9+0.041 8 Som-0.016 6 Fe_2O_3-0.035 6 pH-0.005 CaO. The selenium element was extracted indirectly, the regression coefficient increased from 0.516 to 0.724, the root mean square error was reduced from 0.182 to 0.136 based on this model, which improved the accuracy of the selenium content inversion, and provided a new technique for the precise mapping of selenium elements in a large scale.
In the field of soil digital mapping, precision agriculture and soil resource investigation, the study of aerial hyperspectral data to provide scientific prediction results by aerial hyperspectral have become the focus of research, especially in the case of black soil rich in nutrients in Northeast China. Compared with the main components of soil in the black soil, selenium is a trace element, whose effect on the normal growth of crops is as important as a large number of elements, and it is also a necessary nutrient element for human. In this paper, an indirect extraction model based on the main component is created for the retrieval of selenium content. This model can significantly increase the regression coefficient of selenium content and reduce the error between the measured value and the predicted value. The data source is CASI-1500 aerial hyperspectral imaging system with a spectral range of 380～1 050 nm, and a spatial resolution of 1.5 m. 60 soil samples were collected from the Jiansanjiang area of Heilongjiang. The data of selenium, organic matter, total iron, pH and calcium oxide content were obtained. The BP neural network was selected to establish the inversion model of spectrum and content. In addition, the law of spectral change in the visible and near infrared range of different content of black soil composition was analyzed, and the rule that the spectral reflectance would increase gradually as the content of selenium increased. However, when the selenium content was low, the law would gradually weaken until the other components are disturbed. The spectral characteristics of organic matter were opposite to that of selenium. In general, the reflectance decreases as the content increases, which is closely related to the spectral properties of organic matter. The spectra of the total iron showed similar laws with the organic matter spectrum, indicating that the two have high correlation. The spectral characteristics and detection values of different pH values and calcium oxide contents did not show obvious characteristics, and the law of reflection was not obvious. The correlation coefficients of nutrient contents in different nutrient contents of 60 sampling points were obtained by bands. The results show that the correlation coefficient of each band of pH is the highest, the mean value is 0.63, the second is the correlation coefficient of total iron, 0.54, the correlation coefficient of organic matter and calcium oxide is close to 0.42 and 0.47, while the average correlation coefficient of selenium element content and bands is the lowest, which is 0.38. The first 5 bands with higher correlation coefficients are selected as modeling bands. The characteristics of selenium are 447, 437, 456, 466 and 475 nm; the characteristic bands of organic matter are 447, 456, 466, 437 and 475 nm; the characteristic bands of the whole iron are 752, 695, 800, 762 and 733 nm, and the characteristics of pH are 905, 752 and 695 nm. By calculating the correlation coefficient of sample point selenium content and other components, selenium has a positive correlation with organic matter, and the correlation coefficient is 0.79. The correlation coefficient is negatively correlated with total iron, pH and calcium oxide, and the correlation coefficients are-0.80,-0.94 and-0.69, respectively. In view of the high precision of the inversion of organic matter, total iron, pH and calcium oxide, while the content of selenium is low and the accuracy of direct inversion is insufficient, a method of extracting the functional relationship of selenium elements by extracting the content of four components is designed, and the content of selenium elements is indirectly retrieved. First, the five components and characteristic spectra are analyzed by using neural network, and the regression coefficients R~2 and RMSE of each component are calculated. It is concluded that total iron and pH have higher inversion accuracy, while organic matter and calcium oxide coefficient are lower than 0.8, but they are also significantly higher than those of selenium. A regression model for the content of selenium and other four components was obtained, and Se=0.522 9+0.041 8 Som-0.016 6 Fe_2O_3-0.035 6 pH-0.005 CaO. The selenium element was extracted indirectly, the regression coefficient increased from 0.516 to 0.724, the root mean square error was reduced from 0.182 to 0.136 based on this model, which improved the accuracy of the selenium content inversion, and provided a new technique for the precise mapping of selenium elements in a large scale.
引文
[1] Mezned N,Mechergui M,Alayet F,et.al.Geoscience & Remote Sensing Symposium,2016,4541.
[2] ZHANG Dong-hui,ZHAO Ying-jun,LU Dong-hua,et al(张东辉,赵英俊,陆冬华,等).Chinese Journal of Soil Science(土壤通报),2018,49(1):17.
[3] Daniel ?í?ala,Tereza Zádorová,Jií Kapika.Remote Sens.,2017,9(1):28.
[4] SONG Di-si,SHENG Hao,ZHOU Qing,et al(宋迪思,盛浩,周清,等).Chinese Journal of Soil Science(土壤通报),2016,47(1):1.
[5] DAI Hui-min,GONG Chuan-dong,DONG Bei,et al(戴慧敏,宫传东,董北,等).Acta Pedologica Sinica(土壤学报),2015,52(6):153.
[6] XU Qiang,CHI Feng-qin,KUANG En-jun,et al(徐强,迟凤琴,匡恩俊,等).Soils(土壤),2016,48(5):992.
[7] ZHENG Guang-hui,JIAO Cai-xia,SHANG Gang,et al(郑光辉,焦彩霞,赏刚,等).Spectroscopy and Spectral Analysis(光谱学与光谱分析),2016,36(10):3222.
[8] Sarah Malec,Derek Rogge,Uta Heiden,et al.Remote Sens.,2015,7(9):11776.
[9] XUE Li-hong,ZHOU Ding-hao,LI Ying,et al(薛利红,周鼎浩,李颖,等).Acta Pedologica Sinica(土壤学报),2014,51(5):993.
[10] Veronika Kopaková,Eyal Ben-Dor,Nimrod Carmon,et al.Remote Sens.,2017,9(2):134.
[2] ZHANG Dong-hui,ZHAO Ying-jun,LU Dong-hua,et al(张东辉,赵英俊,陆冬华,等).Chinese Journal of Soil Science(土壤通报),2018,49(1):17.
[3] Daniel ?í?ala,Tereza Zádorová,Jií Kapika.Remote Sens.,2017,9(1):28.
[4] SONG Di-si,SHENG Hao,ZHOU Qing,et al(宋迪思,盛浩,周清,等).Chinese Journal of Soil Science(土壤通报),2016,47(1):1.
[5] DAI Hui-min,GONG Chuan-dong,DONG Bei,et al(戴慧敏,宫传东,董北,等).Acta Pedologica Sinica(土壤学报),2015,52(6):153.
[6] XU Qiang,CHI Feng-qin,KUANG En-jun,et al(徐强,迟凤琴,匡恩俊,等).Soils(土壤),2016,48(5):992.
[7] ZHENG Guang-hui,JIAO Cai-xia,SHANG Gang,et al(郑光辉,焦彩霞,赏刚,等).Spectroscopy and Spectral Analysis(光谱学与光谱分析),2016,36(10):3222.
[8] Sarah Malec,Derek Rogge,Uta Heiden,et al.Remote Sens.,2015,7(9):11776.
[9] XUE Li-hong,ZHOU Ding-hao,LI Ying,et al(薛利红,周鼎浩,李颖,等).Acta Pedologica Sinica(土壤学报),2014,51(5):993.
[10] Veronika Kopaková,Eyal Ben-Dor,Nimrod Carmon,et al.Remote Sens.,2017,9(2):134.
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |