| 尚相春,金倩,杨可明,高伟,吴兵.基于污染种类判别特征的作物Pb、Cu污染种类判别[J].农业环境科学学报,2023,42(1):46-54. |
| 基于污染种类判别特征的作物Pb、Cu污染种类判别 |
| Discrimination of Pb and Cu pollution types in crops based on the discrimination feature of lead - copper pollution type |
| 投稿时间：2022-05-03 |
| DOI：10.11654/jaes.2022-0444 |
| 中文关键词: 高光谱 重金属污染 种类判别 玉米叶片 |
| 英文关键词: hyperspectral heavy metal pollution pollution type discrimination maize leave |
| 基金项目:淮北矿业企业委托项目（2020-117）；中央高校基本科研业务费专项（2022JCCXDC01，2022YJSDC22）；国家自然科学基金项目（41971401） |
|
| 摘要点击次数: 639 |
| 全文下载次数: 654 |
| 中文摘要: |
| 为判别作物中的重金属污染种类，开展了 Pb、Cu 胁迫下的典型作物（玉米）培育实验，获取了作物叶片的高光谱数据与Pb、Cu污染信息。首先对光谱进行分数阶、整数阶导数变换，而后利用差值比光谱指数构建特征参量以组成Pb、Cu污染种类判别特征（DFLCPT），最终基于DFLCPT数据构建了用于作物Pb、Cu污染种类判别的随机森林分类（RFC）、K-最邻近分类（KNNC）、支持向量机分类（SVC）、高斯过程分类（GPC）模型。结果表明：在依托多种导数光谱构建的差值比光谱指数（DRSI）中，以0.9阶导数光谱为基准的 DRSI[2 412，1 223，636]与样本 Pb、Cu 污染种类的相关系数绝对值最大，为 0.764 1；在依托多维度 DFLCPT（DFLCPTnD）建立的 SVC、RFC、KNNC、GPC玉米 Pb、Cu污染种类判别模型中，RFC模型的效果优于 SVC、KNNC、GPC模型，其在训练集与验证集中取得的最高正确率均为100%，精度较好，稳定性较强。研究表明，依托DFLCPT的模型在Pb、Cu污染种类判别中达到了预期效果，可为大规模的作物重金属污染种类判别提供一定的技术支撑。 |
| 英文摘要: |
| To discriminate the heavy metal pollution types present in crops, a typical crop（maize）was cultivated under Pb and Cu stress, following which hyperspectral data containing Pb and Cu pollution information were obtained from the crop leaves. Fractional-order and integer - order derivative transformations were first applied to the spectra. Then, difference ratio spectral indexes（DRSIs）were used to construct feature covariates for forming the discrimination feature of lead - copper pollution type（DFLCPT）. Finally, random forest classification（RFC）, K-nearest neighbor classification（KNNC）, support vector machine classification（SVC）, and Gaussian process classification（GPC）models were constructed on the basis of the DFLCPT data for discrimination of the Pb and Cu pollution type in the crop. It was found that among the DRSIs constructed from a variety of derivative spectra, those [2 412, 1 223, 636] based on the 0.9 derivative spectrum had the largest absolute value of the correlation coefficient between the Pb and Cu pollution types of the sample, which is 0.764 1. Among the four classification models established on the basis of multidimensional DFLCPT（DFLCPTnD）data, the RFC model had a better effect than the SVC, KNNC, and GPC models. The highest accuracy of the RFC model for the training set and verification set was 100%, with good accuracy and strong stability. The results show that the DFLCPT-based model has achieved the expected effect in the discrimination of Pb and Cu pollution types and can provide technical support for the discrimination of heavy metal pollution types in crops grown on a large scale. |
| HTML
查看全文
查看/发表评论 下载PDF阅读器 |
|
|
|
