State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430079, China.
School of Resource and Environmental Sciences, Wuhan University, Wuhan 430079, China.
Sensors (Basel). 2019 Sep 9;19(18):3889. doi: 10.3390/s19183889.
This paper proposed an optimal spectral resolution for diagnosing cadmium-lead (Cd-Pb) cross contamination with different pollution levels based on the hyperspectral reflectance of rice canopy. Feature bands were sequentially selected by two-way analysis of variance (ANOVA2) and random forests from the high-dimensional hyperspectral data after preprocessing. Then Support Vector Machine (SVM) was applied to diagnose the pollution levels using different feature bands combination with different spectral resolutions and cross validation was conducted to evaluate the distinguishing accuracies. Finally, the optimal spectral resolution could be determined by comparing the diagnosing accuracies of the optimal feature bands combination in each spectral resolution. In the experiments, the hyperspectral reflectance data of rice canopy with ten different spectral resolutions was captured, covering 16 pretreatments of Cd and Pb pollution. The experimental results showed the optimal spectral resolution was 9 nm with the highest average accuracy of 0.71 and relatively standard deviation of 0.07 for diagnosing the categories and levels of Cd-Pb cross contamination. The useful exploration provided an evidence for optimal spectral resolution selection to reduce the cost of heavy metal pollution diagnose.
本文提出了一种基于水稻冠层高光谱反射率，用于诊断不同污染水平下镉铅（Cd-Pb）交叉污染的最佳光谱分辨率。在预处理后的高维高光谱数据中，通过双向方差分析（ANOVA2）和随机森林依次选择特征波段。然后，使用支持向量机（SVM）根据不同的光谱分辨率和交叉验证，使用不同的特征波段组合来诊断污染水平，以评估区分精度。最后，通过比较每个光谱分辨率中最佳特征波段组合的诊断精度，确定最佳光谱分辨率。在实验中，捕获了具有十个不同光谱分辨率的水稻冠层高光谱反射率数据，涵盖了 16 种 Cd 和 Pb 污染的预处理。实验结果表明，最佳光谱分辨率为 9nm，诊断 Cd-Pb 交叉污染类别和水平的平均准确率最高，为 0.71，相对标准偏差为 0.07。这项有用的探索为选择最佳光谱分辨率以降低重金属污染诊断成本提供了证据。
