基于成像高光谱数据的温室水稻重金属胁迫诊断研究 |
| |
引用本文: | 张双印,王云将,欧阳炜,费腾. 基于成像高光谱数据的温室水稻重金属胁迫诊断研究[J]. 安徽农业科学, 2018, 46(1): 5-9. DOI: 10.3969/j.issn.0517-6611.2018.01.002 |
| |
作者姓名: | 张双印 王云将 欧阳炜 费腾 |
| |
作者单位: | 武汉大学资源与环境科学学院,湖北武汉,430079;华中农业大学资源与环境学院,湖北武汉,430070 |
| |
摘 要: | [目的]通过温室水稻叶片高光谱影像数据,从Cd和Pb不同梯度的交叉胁迫中诊断具体的胁迫类别和胁迫梯度。[方法]经过双因素方差分析筛选出特征波段,比较SVM和BP神经网络在诊断能力上的强弱。[结果]在几种预处理方法中,对光谱二阶微分预处理可以对Cd和Pb胁迫达到很好的诊断效果,预处理后挑选出6个对Cd胁迫敏感的特征波段以及10个对Pb胁迫敏感的特征波段。基于SVM的诊断Cd胁迫的精度达86%,对3个具体梯度的诊断精度达75%、90%、96%,对Pb胁迫的诊断精度达85%,3个梯度分别为83%、85%、88%;基于BP神经网络的Cd胁迫诊断精度达88%,3个梯度为69%、75%、75%;对Pb胁迫的诊断精度达88%,3个梯度为81%、69%、69%。[结论]从植被高光谱影像数据诊断重金属Cd和Pb胁迫是可行的,且SVM的诊断精度整体优于BP神经网络。
|
关 键 词: | 高光谱 重金属诊断 SVM BP神经网络 |
Diagnosis of Heavy Metal Stress in Leaf of Rice in Greenhouse Based on Hyperspectral Image |
| |
Abstract: | [Objective] This study aimed to diagnose specific stress categories and stress gradients from the cross-stress of Cd and Pb with high spectral imagery data of greenhouse rice leaves.[Method] After double factor variance analysis,the characteristic bands for diagnosis were selected,and two models of SVM and BP neural network were compared in terms of diagnostic ability.[Result] The results showed that with the pretreatment of 2nd spectral derivative,SVM could achieve very good diagnostic effect for Cd and Pb stress.6 characteristic bands were identified sensitive to Cd stress,and 10 characteristic bands were identified sensitive to Pb stress.The accuracy of diagnostic Cd stress based on SVM was 86%,and the diagnostic accuracy of three gradients were 75%,90% and 96%,while the accuracy of diagnostic Pb stress based on SVM was 85%,and the diagnostic accuracy of three gradients were 83%,85% and 88%.The diagnostic accuracy of Pb stress based BP neutral network was 88%,and the diagnostic accuracy of three gradients were 69%,75% and 75%,while the diagnostic accuracy of Pb stress was 88%,and the diagnostic accuracy of three gradients were 81%,69% and 69%.[Conclusion] It is feasible to diagnose heavy metal Cd and Pb stress from hyperspectral spectral imaging data of vegetation,and the accuracy of SVM is satisfied. |
| |
Keywords: | |
本文献已被 CNKI 万方数据 等数据库收录! |
|