土壤锰污染的高光谱分级评价

Assessing manganese pollution of soil using hypersepctra technology

提出一种利用高光谱技术进行土壤锰污染分级评价的方法。以FieldSpec3地物光谱仪采集矿区土壤光谱反射率150份,随机分成校正集(115份)和检验集(35份)。光谱经小波去噪和多元散射校正(MSC)处理后,以主成分分析法(PCA)降维。将降维所得的前5个主成分数据为输入变量,分别采用Fisher线性判别、Byes逐步判别、模糊模式识别以及BP-ANN判别四种方法建立了土壤锰污染分级评价模型,并利用35个未知样对模型进行检验。结果表明:Fisher线性判别与模糊模式识别预测准确率为80%,Byes逐步判别为82.86%,BP-ANN模型预测精度最高,达85.71%。说明以高光谱技术进行土壤锰污染分级评价是可行,且BP-ANN是建模的优选算法。

A new method is put forward to assess the manganese（Mn）pollution of soil by hyperspectra technology.150 soil samples were collected using a FieldSpec3 spectrometer in mine area.All samples were divided randomly into 2 groups.One group with 115 samples was used as calibration set,and another with 35 samples was used as prediction set.The samples data were pretreated with the methods of wavelet denoising and multielement scattering correction（MSC）,and then were analyzed by principal component analysis（PCA）.The top 5 principal components of PCA were used as the new variables,and analysised by fisher linear discrimination,Bayes multi-types stepwise discrimination,fuzzy pattern recognition and back-propagation artificial neural network（ANN-BP）.Then,the 35 unknown samples in the prediction set were predicted.The result shows that the discriminating rate is 80% with the methods of fisher linear discrimination and fuzzy pattern recognition,82.86% with the method of Bayes multi-types stepwise discrimination,and 85.71% with the method of BP-ANN model.Therefore,the feasibility of assessing the Mn pollution of soil in rapid and non-invasive way by hyperspectra technology is proved,and PCA combined with BP-ANN is confirmed as a preferred method.