基于数字图像颜色提取的土壤有机质预测研究
|
摘要
有机质是土壤质量的重要判定标准,其快速测定可为精准农业的实施提供基本的数据支撑。传统有机质测定法,通过野外取样,实验室化学分析,不仅费时费力,还效率低,完全不能满足现今社会发展对土壤信息的大量需求。通过光谱仪测定土壤反射率建立有机质估测模型,可快速预测有机质含量,但光谱仪价格较高,对操作环境要求严格,限制了其广泛应用。仅有RGB三波段的可见光传感器,相比光谱仪而言不仅价格便宜,还易于操作,借助其诸多优势,通过定量获取土壤表面颜色信息,解决土壤有机质快速测定问题,不管是从实用性还是经济性角度,都值得去探索研究。为验证可见光下提取数字图像颜色信息能快速预测有机质的可行性和适用性,采用数码相机获取土壤表面颜色,分析土壤表面组成特点,确定最佳取样面积,并比较不同土样制样标准(<1 mm和<0.5 mm)与有机质的相关性,同时选取相关性高的颜色变量,通过回归分析,建立有机质预测模型。研究结果表明,以950 pixel×950 pixel作为取样面积,可以较稳定获取土样表面颜色,并减少边缘效应对取样结果的影响;对比<1 mm和<0.5 mm土样与有机质的相关性发现,<1 mm土样的RGB三波段与有机质的相关性更高,适合作为土壤颜色获取时的制样标准。在RGB三波段中, Red波段表现了与有机质的最高相关性,其相关系数为-0.70;通过对RGB三波段进行数学变换与超红(ExR)计算,可增加各颜色值与有机质的相关性,其中ExR指数表现了与有机质最高的相关性,其相关系数为-0.86。单变量建模过程中, ExR倒数模型获得了最好预测效果;多变量建模时,各颜色标准差参与建模,使颜色信息描述更为全面,明显提高了模型的预测精度,获得了最好的建模效果,其R~2=0.80, RMSE=0.51,检验结果R■=0.84, RMSE_(val)=0.54,能较好反应研究区有机质变异特点。通过黑土检验所建模型的预测效果,仅单变量Red波段模型表现出较好预测结果,其检验结果表明红波段是有机质较为敏感波段,在不同土类中具有通用性。虽然研究中所建模型难以扩展到对其他土类的预测,但对同类土壤预测效果表明,数码相机作为颜色定量获取的工具,具有实现快速预测土壤有机质的潜力。
As an important criterion for determining soil quality, the rapid determination of soil organic matter(SOM) can provide basic data support for the implementation of precision agriculture. Traditional determination method of SOM, through field sampling and laboratory chemical analysis, not only time-consuming and laborious, but also inefficient, cannot meet the large-scale demand for soil information in current social development. Although the prediction model of SOM can be established based upon the characteristics of spectral reflectance of soil affected by SOM to realize the rapid prediction for SOM, the spectrometer is of high price and strict operation environment, which limits its wide application. Then visible-light sensor with RGB is cheap and easy to operate. Therefore it is worth exploring and studying the rapid determination of SOM from the perspective of practicality and economy, with the help of many advantages of visible-light sensor. Therefore, in order to verify the feasibility and applicability of extracting color information of digital images for fast prediction of SOM, the paper uses a digital camera to obtain the soil surface color information, analyses the characteristics of soil surface composition, determines the optimal sampling area, compares the correlation between different sample preparation standards(<1 mm and <0.5 mm) and SOM, selects the high correlation of color variables, and establishes the prediction model of SOM through regression analysis. The results show that the 950×950 pixel as the sampling area can obtain the color of the soil surface more stably and reduce the influence of the edge effect on the sampling result. In the correlation analysis between the soil samples <1 mm and <0.5 mm and SOM, the RGB bands of <1 mm soil samples have a higher correlation with SOM and are suitable as sample preparation standards for soil color acquisition. In the three bands of RGB, the red band showed the highest correlation with SOM, with a correlation coefficient of-0.70. The correlation between color and SOM was enhanced by the mathematical transformation of the RGB band and the excessred(ExR) calculation, in which the ExR index shows the highest correlation with SOM with a correlation coefficient of-0.86. In a single variable modeling process, the best predictive effect is obtained by ExR reciprocal model. In the multivariable modeling, the standard deviations of each color were involved in the modeling, which causes the color information description to be more comprehensive, and the best modeling results are obtained that can better reflect the variation of SOM within the study area, its R~2=0.80, RMSE=0.51, the validation result R■=0.84, RMSE_(val)=0.54. Based on the prediction results of the model for black soil, only the single-variable red band model shows a good prediction effect, and the test results show that the red band is a sensitive band of SOM and has its universality in different soil types. Although the model built in this study cannot be extended to predict other soil types, the prediction of the same soil shows that the digital camera, as a quantitative color imetric tool, has the potential to rapidly predict SOM content.
As an important criterion for determining soil quality, the rapid determination of soil organic matter(SOM) can provide basic data support for the implementation of precision agriculture. Traditional determination method of SOM, through field sampling and laboratory chemical analysis, not only time-consuming and laborious, but also inefficient, cannot meet the large-scale demand for soil information in current social development. Although the prediction model of SOM can be established based upon the characteristics of spectral reflectance of soil affected by SOM to realize the rapid prediction for SOM, the spectrometer is of high price and strict operation environment, which limits its wide application. Then visible-light sensor with RGB is cheap and easy to operate. Therefore it is worth exploring and studying the rapid determination of SOM from the perspective of practicality and economy, with the help of many advantages of visible-light sensor. Therefore, in order to verify the feasibility and applicability of extracting color information of digital images for fast prediction of SOM, the paper uses a digital camera to obtain the soil surface color information, analyses the characteristics of soil surface composition, determines the optimal sampling area, compares the correlation between different sample preparation standards(<1 mm and <0.5 mm) and SOM, selects the high correlation of color variables, and establishes the prediction model of SOM through regression analysis. The results show that the 950×950 pixel as the sampling area can obtain the color of the soil surface more stably and reduce the influence of the edge effect on the sampling result. In the correlation analysis between the soil samples <1 mm and <0.5 mm and SOM, the RGB bands of <1 mm soil samples have a higher correlation with SOM and are suitable as sample preparation standards for soil color acquisition. In the three bands of RGB, the red band showed the highest correlation with SOM, with a correlation coefficient of-0.70. The correlation between color and SOM was enhanced by the mathematical transformation of the RGB band and the excessred(ExR) calculation, in which the ExR index shows the highest correlation with SOM with a correlation coefficient of-0.86. In a single variable modeling process, the best predictive effect is obtained by ExR reciprocal model. In the multivariable modeling, the standard deviations of each color were involved in the modeling, which causes the color information description to be more comprehensive, and the best modeling results are obtained that can better reflect the variation of SOM within the study area, its R~2=0.80, RMSE=0.51, the validation result R■=0.84, RMSE_(val)=0.54. Based on the prediction results of the model for black soil, only the single-variable red band model shows a good prediction effect, and the test results show that the red band is a sensitive band of SOM and has its universality in different soil types. Although the model built in this study cannot be extended to predict other soil types, the prediction of the same soil shows that the digital camera, as a quantitative color imetric tool, has the potential to rapidly predict SOM content.
引文
[1] McBratney A B, Stockmann U, Angers D A, et al. Soil Carbon. Progress in Soil Science. Springer, Cham, 2014.
[2] Rossel R A V, Webster R. European Journal of Soil Science, 2012, 63(6): 848.
[3] Reeves JB III. Geoderma, 2010, 158(1-2): 3.
[4] FANG Shao-wen, YANG Mei-hua, ZHAO Xiao-min, et al(方少文, 杨梅花, 赵小敏, 等). Acta Pedologica Sinica(土壤学报), 2014, 51(5): 1003.
[5] Stiglitz R, Mikhailova E, Post C, et al. Geoderma, 2017, 286: 98.
[6] Adhikari K, Hartemink A E. Geoderma, 2016, 262: 101.
[7] Gelder B K, Anex R P, Kaspar T C, et al. Soil Science Society of America Journal, 2011, 75(5): 1821.
[8] Gregory S D L, Lauzon J D, O’Halloran I P, et al. Canadian Journal of Soil Science, 2006, 86(3): 573.
[9] ViscarraRossel R A, Fouad Y, Walter C. Biosystems Engineering, 2008, 100(2): 149.
[10] LIU Chao, YUAN Man, ZHUANG Wen-hua, et al(刘超, 袁满, 庄文化, 等). China Sciencepaper(中国科技论文), 2015, 10(9): 1071.
[11] Wu Caiwu, Yang Yue, Xia Jianxin. Archives of Agronomy and Soil Science, 2017, 63(10): 1346.
[12] Kirillova N P, Kemp D B, Artemyeva Z S. European Journal of Soil Science, 2017, 68(4): 420.
[13] Soriano-Disla J M, Janik L J, Viscarra Rossel R A, et al. Applied Spectroscopy Reviews, 2014, 49(2): 139.
[14] SHEN Bao-guo, CHEN Shu-ren, YIN Jian-jun, et al(沈宝国, 陈树人, 尹建军, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2009, 25(6): 163.
[15] Liu H J, Zhang Y Z, Zhang B. Environmental Monitoring and Assessment, 2009, 154: 147.
[16] Viscarra Rossel R A, Walter C, Fouad Y. Assessment of Two Reflectance Techniques for the Qantification of the Within-Field Spatial Variability of Soil Organic Carbon. Precision Agriculture, 2003. 697.
[17] SUN Ning, CHANG Qing-rui, LIU Meng-yun, et al(孙宁, 常庆瑞, 刘梦云, 等). Journal of Northwest Forestry University(西北林学院学报), 2011, 26(1): 56.
[18] XU Bin-bin(徐彬彬). Soils(土壤), 2000, 32(6): 281.
[19] Mouazen A M, Maleki M R, Baerdemaeker J D, et al. Soil &Tillage Research, 2007, 93(1): 13.
[20] Waiser T H, Morgan C L S, Brown D J, et al. Soil Science Society of American Journal, 2007, 71(2): 389.
[2] Rossel R A V, Webster R. European Journal of Soil Science, 2012, 63(6): 848.
[3] Reeves JB III. Geoderma, 2010, 158(1-2): 3.
[4] FANG Shao-wen, YANG Mei-hua, ZHAO Xiao-min, et al(方少文, 杨梅花, 赵小敏, 等). Acta Pedologica Sinica(土壤学报), 2014, 51(5): 1003.
[5] Stiglitz R, Mikhailova E, Post C, et al. Geoderma, 2017, 286: 98.
[6] Adhikari K, Hartemink A E. Geoderma, 2016, 262: 101.
[7] Gelder B K, Anex R P, Kaspar T C, et al. Soil Science Society of America Journal, 2011, 75(5): 1821.
[8] Gregory S D L, Lauzon J D, O’Halloran I P, et al. Canadian Journal of Soil Science, 2006, 86(3): 573.
[9] ViscarraRossel R A, Fouad Y, Walter C. Biosystems Engineering, 2008, 100(2): 149.
[10] LIU Chao, YUAN Man, ZHUANG Wen-hua, et al(刘超, 袁满, 庄文化, 等). China Sciencepaper(中国科技论文), 2015, 10(9): 1071.
[11] Wu Caiwu, Yang Yue, Xia Jianxin. Archives of Agronomy and Soil Science, 2017, 63(10): 1346.
[12] Kirillova N P, Kemp D B, Artemyeva Z S. European Journal of Soil Science, 2017, 68(4): 420.
[13] Soriano-Disla J M, Janik L J, Viscarra Rossel R A, et al. Applied Spectroscopy Reviews, 2014, 49(2): 139.
[14] SHEN Bao-guo, CHEN Shu-ren, YIN Jian-jun, et al(沈宝国, 陈树人, 尹建军, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2009, 25(6): 163.
[15] Liu H J, Zhang Y Z, Zhang B. Environmental Monitoring and Assessment, 2009, 154: 147.
[16] Viscarra Rossel R A, Walter C, Fouad Y. Assessment of Two Reflectance Techniques for the Qantification of the Within-Field Spatial Variability of Soil Organic Carbon. Precision Agriculture, 2003. 697.
[17] SUN Ning, CHANG Qing-rui, LIU Meng-yun, et al(孙宁, 常庆瑞, 刘梦云, 等). Journal of Northwest Forestry University(西北林学院学报), 2011, 26(1): 56.
[18] XU Bin-bin(徐彬彬). Soils(土壤), 2000, 32(6): 281.
[19] Mouazen A M, Maleki M R, Baerdemaeker J D, et al. Soil &Tillage Research, 2007, 93(1): 13.
[20] Waiser T H, Morgan C L S, Brown D J, et al. Soil Science Society of American Journal, 2007, 71(2): 389.