现有基于高光谱遥感技术的土壤重金属含量反演模型, 大多是采用同一试验区且有限的样本点进行定量反演建模。 但考虑到实际应用需求, 该类模型在不同试验区是否具有较好的迁移推广能力是目前迫切需要回答的问题。 如不可行, 是否存在其他可行手段用于土壤重金属污染评估？ 为回答上述问题, 选取湖南省郴州市和衡阳市两铅锌矿区作为实验研究区, 并首先利用郴州地区采样点分别对Pb和Zn两种重金属进行定量回归建模和定性分类建模, 然后比较两种模型在衡阳实验区的可迁移能力。 实验结果表明: (1)基于偏最小二乘回归(PLSR)的定量回归模型可迁移能力较差。 分别采用四种光谱预处理方式建模, 发现回归模型对异地采样的预测精度很低, 难以正确反演衡阳试验区重金属Pb和Zn的含量。 (2)基于支持向量机(SVM)分类的定性反演模型具有一定的可迁移能力, 以郴州地区采样数据训练得到的SVM分类模型能有效判定衡阳试验区Pb、 Zn的污染状况, 分类精度分别达到84.78%和86.96%。 结果表明, 在快速检测土壤重金属污染状况的问题上, 定性分类是一种更加切实可行的方式。

The existing model of soil heavy metal content reversal model by hyperspectral remote sensing technology is mostly based on the limited sample points of the same study area. However, considering the practical application requirements, whether the model has a good migrate ability is an urgent question. If it is not feasible, is there any other feasible means for soil heavy metal pollution assessment? In order to answer the above-mentioned questions, this paper selects two lead-zinc mines in Chenzhou City and Hengyang City as research areas. The quantitative inversion and qualitative classification of heavy metals Pb and Zn were carried out using the sampling sites in Chenzhou area to compare the two models in Hengyang City of the migrate ability. Experiments show that: (1) Quantitative inversion model based on Partial least squares regression (PLSR) has poor migration ability. The regression model was established by four spectral preprocessing methods. It was found that the prediction accuracy of the model was very low, and it was difficult to correctly invert the contents of Pb and Zn in Hengyang research area. (2) Support vector machine (SVM) classification of qualitative inversion model has a certain ability to migrate. Based on Chenzhou area sampling data, training SVM classification model can effectively predict the Hengyang research area Pb and Zn pollution situation, the prediction accuracies are 84.78% and 86.96%, respectively. The results show that qualitative classification is a more practical way to detect soil heavy metal pollution rapidly.