基于数字图像技术反演中国东北黑土有机质含量  

Reversing Organic Matter Contents in Black Soils in Northeast China Using Digital Image Technology

作　　者：王亚丹 张凤[2] 胡文友[2] 于东升[2] 迟凤琴 张超 徐英德 杨顺华 俞元春[1] 姜军[2] 徐仁扣[2] WANG Yadan;ZHANG Feng;HU Wenyou;YU Dongsheng;CHI Fengqin;ZHANG Chao;XU Yingde;YANG Shunhua;YU Yuanchun;JIANG Jun;XU Renkou(Co-innovation Center for the Sustainable Forestry in Southern China/College of Ecology and Environment,Nanjing Forestry University,Nanjing 210037,China;Key Laboratory of Soil and Sustainable Agriculture,Chinese Academy of Sciences,Nanjing 211135,China;Heilongjiang Black Soil Conservation and Utilization Research Institute,Harbin 150086,China;College of Land Science and Technology,China Agricultural University,Beijing 100193,China;College of Land and Environment,Shenyang Agricultural University,Shenyang 110866,China)

机构地区：[1]南方现代林业协同创新中心/南京林业大学生态与环境学院,南京210037 [2]土壤与农业可持续发展重点实验室(中国科学院),南京211135 [3]黑龙江省黑土保护利用研究院,哈尔滨150086 [4]中国农业大学土地科学与技术学院,北京100193 [5]沈阳农业大学土地与环境学院,沈阳110866

出　　处：《土壤》2024年第5期1051-1056,共6页Soils

基　　金：国家重点研发计划项目(2021YFD1500202);中国科学院战略性先导科技专项项目(XDA2801010104);国家重点农业科技项目(NK2022180104)资助。

摘　　要：以东北黑土为研究对象,利用数码相机获取黑土的数字图像,遴选与土壤有机质(SOM)含量相关的红(R)、绿(G)、蓝(B)颜色分量,并通过逐步多元回归(SMRM)和神经网络模型(NNM)建立基于数字图像的SOM含量预测模型。结果表明:黑土数字图像的各颜色分量原始值与SOM含量的相关系数绝对值(r)依次为:R>G>B,分别为0.67、0.65、0.50。原始值经对数和开平方数值变换后,r增加,而经倒数和平方变换后,r降低。据此,基于数字图像R、G、B颜色分量的原始值和各变换值建立了预测SOM含量的SMRM模型,训练集和验证集决定系数(R2)分别为0.43~0.50和0.46~0.50,均方根误差(RMSE)分别为1.28%~1.39%和1.31%~1.39%(P<0.001),其中基于对数和开平方变换值的模型拟合程度和预测精度更高。同时,利用NNM模型基于黑土R、G、B颜色分量的原始值反演SOM含量,发现多层感知器算法模型得到的黑土SOM实测值和预测值之间R2均为0.49,RMSE为1.31%和1.28%(P<0.001)。因此,SMRM和NNM均能通过黑土数字图像的R、G、B颜色分量反演SOM含量,其是快速获取中国东北黑土SOM含量的一套可操作的预测方法。In the present study,digital images of black soil were identified by their red(R),green(G),and blue(B)color components that correlate with SOM content,and then used to construct predictive stepwise multiple regression models(SMRM)and neural network methodologies(NNM)for SOM content.Our findings revealed that the absolute value of correlation coefficients(r)between each original color component and SOM content followed the order:R>G>B,with r of 0.67,0.65 and 0.50,respectively.The r value increased after logarithmic and square root transformations,but decreased following reciprocal and square changes.The determination coefficient(R2)for SMRM training and validation sets with and without transformations fall within the range of 0.43 to 0.50 and 0.46 to 0.50,and the root mean square error(RMSE)ranged 1.28%–1.39%,and 1.31%–1.39%,respectively(P<0.001).Specifically,SMRM incorporating logarithmic and square root transformations of R,G and B color components demonstrated superior predictive performance and higher accuracy.Subsequently,multi-layer perceptron neural networks using original values of R,G and B color components successfully estimated SOM content,with R2 of 0.49 and 0.49,and RMSE of 1.31%and 1.28%for the training and validation sets,respectively(P<0.001).Therefore,both SMRM and NNM provided effective estimates in SOM content for black soil using its digital image.Our findings provide an operational prediction model for the rapid assessment of SOM content of black soil in northeast China.

关 键 词：黑土 有机质 数字图像 逐步多元回归 神经网络模型 

分 类 号：S151.9[农业科学—土壤学]