矿区土壤重金属Pb、Cd污染状况高光谱分类建模
|
摘要
针对矿区土壤重金属含量高度变异性及样本不均衡导致重金属污染状况分类误差较大的问题,本文在光谱预处理及光谱变换基础上,采用主成分分析(PCA)对光谱进行降维处理,并通过SMOTE算法生成虚拟样本均衡各污染等级样本,最后应用随机森林(RF)对Cd、Pb进行回归与分类。研究结果表明:定量反演重金属Pb、Cd含量精度很低;在定性分析试验中对降维前光谱样本应用SMOTE算法,土壤重金属Pb、Cd污染等级分类精度较原始样本分类精度均有较大提升,且少数类别误判率也降低明显。其研究为大面积监测矿区土壤重金属污染状况提供了一种有效、精确的方法。
In view of the high variability of soil heavy metal content and the imbalanced samples lead to the high classification error of heavy metal pollution in mining area. Based on the spectral preprocessing and spectral transformation,this paper uses principal component analysis( PCA) for spectral dimension,and applies SMOTE algorithm to generate virtual samples balance each pollution grade sample,and heavy metal Cd and Pb are regressed and classified by random forest( RF). The results show that the quantitative inversion precision of heavy metals Pb and Cd is bad. In the qualitative analysis experiment,SMOTE algorithm was applied to spectral samples before dimension reduction. The classification accuracy of Pb and Cd pollution levels in soil was greatly improved compared with that of the original samples,and the misjudgment rate of a few categories was also significantly decreased. The study provides an effective and accurate method for monitoring soil heavy metal pollution in a large area.
In view of the high variability of soil heavy metal content and the imbalanced samples lead to the high classification error of heavy metal pollution in mining area. Based on the spectral preprocessing and spectral transformation,this paper uses principal component analysis( PCA) for spectral dimension,and applies SMOTE algorithm to generate virtual samples balance each pollution grade sample,and heavy metal Cd and Pb are regressed and classified by random forest( RF). The results show that the quantitative inversion precision of heavy metals Pb and Cd is bad. In the qualitative analysis experiment,SMOTE algorithm was applied to spectral samples before dimension reduction. The classification accuracy of Pb and Cd pollution levels in soil was greatly improved compared with that of the original samples,and the misjudgment rate of a few categories was also significantly decreased. The study provides an effective and accurate method for monitoring soil heavy metal pollution in a large area.
引文
[1]蒋明,郭云开,钱佳,等.不同采样间隔下分数阶微分对土壤重金属高光谱数据的影响[J].测绘通报,2018(10):37-40.
[2]郭云开,周烽松,丁美青,等.水稻冠层与土壤高光谱反演土壤重金属对比研究[J].遥感信息,2017,32(2):173-179.
[3]贾亚琪,程志飞,刘品祯,等.煤矿区周边农田土壤重金属积累特征及生态风险评价[J].土壤通报,2016,47(2):474-479.
[4] KOOISTRA L,WEHRENS R,LEUVEN R,et al.Possibilities of visible-near-infrared spectroscopy for the assessment of soil contamination in river floodplains[J].Analytica Chimica Acta,2001,446(1-2):97-105.
[5] KEMPER T,SOMMER S. Estimate of heavy metal contamination in soils after a mining accident using reflectance spectroscopy[J]. Environmental Science&Technology,2002,36(12):2742-2747.
[6]王晓华,邓喀中,杨化超.土壤重金属污染信息提取遥感模型的建立———以水口山矿区铅锌污染为例[J].测绘通报,2013(3):29-31.
[7]郭云开,刘宁,刘磊,等.土壤Cu含量高光谱反演的BP神经网络模型[J].测绘科学,2018,43(1):135-139.
[8]陶超,王亚晋,邹滨,等.土壤重金属铅、锌高光谱反演模型可迁移能力分析研究[J].光谱学与光谱分析,2018,38(6):1850-1855.
[9]赵楠,张小芳,张利军.不平衡数据分类研究综述[J].计算机科学,2018,45(S1):22-27.
[10] XIE X L,LI A B.Identication of soil prole classes using depth-weighted visible-nearinfrared spectral reectance[J].Geoderma,2018,325(12):90-101.
[11] CHAWLA N V,BOWYER K W,HALL L O,et al. SMOTE:synthetic minority over-sampling technique[J]. Journal of Articial Intelligence Research,2002,16(1):321-357.
[12]郭玉宝,池天河,彭玲,等.利用随机森林的高分一号遥感数据进行城市用地分类[J].测绘通报,2016(5):73-76.
[13]冯文卿,眭海刚,涂继辉,等.高分辨率遥感影像的随机森林变化检测方法[J].测绘学报,2017,46(11):1880-1890.
[14]涂宇龙,邹滨,姜晓璐,等.矿区土壤Cu含量高光谱反演建模[J].光谱学与光谱分析,2018,38(2):575-581.
[15]宋海峰,陈广胜,杨巍巍.基于PCA的高光谱遥感图像分类[J].测绘工程,2017,26(12):17-20.
[2]郭云开,周烽松,丁美青,等.水稻冠层与土壤高光谱反演土壤重金属对比研究[J].遥感信息,2017,32(2):173-179.
[3]贾亚琪,程志飞,刘品祯,等.煤矿区周边农田土壤重金属积累特征及生态风险评价[J].土壤通报,2016,47(2):474-479.
[4] KOOISTRA L,WEHRENS R,LEUVEN R,et al.Possibilities of visible-near-infrared spectroscopy for the assessment of soil contamination in river floodplains[J].Analytica Chimica Acta,2001,446(1-2):97-105.
[5] KEMPER T,SOMMER S. Estimate of heavy metal contamination in soils after a mining accident using reflectance spectroscopy[J]. Environmental Science&Technology,2002,36(12):2742-2747.
[6]王晓华,邓喀中,杨化超.土壤重金属污染信息提取遥感模型的建立———以水口山矿区铅锌污染为例[J].测绘通报,2013(3):29-31.
[7]郭云开,刘宁,刘磊,等.土壤Cu含量高光谱反演的BP神经网络模型[J].测绘科学,2018,43(1):135-139.
[8]陶超,王亚晋,邹滨,等.土壤重金属铅、锌高光谱反演模型可迁移能力分析研究[J].光谱学与光谱分析,2018,38(6):1850-1855.
[9]赵楠,张小芳,张利军.不平衡数据分类研究综述[J].计算机科学,2018,45(S1):22-27.
[10] XIE X L,LI A B.Identication of soil prole classes using depth-weighted visible-nearinfrared spectral reectance[J].Geoderma,2018,325(12):90-101.
[11] CHAWLA N V,BOWYER K W,HALL L O,et al. SMOTE:synthetic minority over-sampling technique[J]. Journal of Articial Intelligence Research,2002,16(1):321-357.
[12]郭玉宝,池天河,彭玲,等.利用随机森林的高分一号遥感数据进行城市用地分类[J].测绘通报,2016(5):73-76.
[13]冯文卿,眭海刚,涂继辉,等.高分辨率遥感影像的随机森林变化检测方法[J].测绘学报,2017,46(11):1880-1890.
[14]涂宇龙,邹滨,姜晓璐,等.矿区土壤Cu含量高光谱反演建模[J].光谱学与光谱分析,2018,38(2):575-581.
[15]宋海峰,陈广胜,杨巍巍.基于PCA的高光谱遥感图像分类[J].测绘工程,2017,26(12):17-20.