| 基于BP神经网络的橡胶苗叶片磷含量高光谱预测 |
| |
| 引用本文： | 郭澎涛,苏艺,茶正早,林清火,罗微,林钊沐.基于BP神经网络的橡胶苗叶片磷含量高光谱预测[J].农业工程学报,2016,32(Z1):177-183. |
| |
| 作者姓名： | 郭澎涛 苏艺 茶正早 林清火 罗微 林钊沐 |
| |
| 作者单位： | 中国热带农业科学院橡胶研究所,儋州,571737 |
| |
| 基金项目： | 国家天然橡胶产业技术体系（CARS-34）；海南省自然科学基金（314142） |
| |
| 摘 要： | 为验证高光谱技术在橡胶苗叶片磷素营养诊断方面的可行性,该文以砂培橡胶苗为研究对象,利用高光谱仪测得不同磷处理水平下橡胶苗叶片光谱反射率,并应用微分技术求取去噪后光谱反射率一阶和二阶导数,以叶片磷含量和光谱变量相关性分析为基础,选择出叶片磷含量敏感波段,最后以敏感波段为输入变量,结合多重线性回归、偏最小二乘回归和反向传播神经网络模型对叶片磷含量进行预测。结果表明:原始光谱反射率555和722 nm、一阶导数674、710、855、1 091、1 197、1 275、1 718、2 181和2 228 nm以及二阶导数816、890、1 339、1 357和2 201 nm为叶片磷含量敏感波段;反向传播神经网络模型预测精度最高,训练集和验证集中预测值和实测值之间的相关系数r分别为0.964和0.967,均方根误差RMSE分别为0.0139和0.00856,模型性能指数(ratio of performance to deviation,RPD)分别为3.71和3.23,证明高光谱技术可以快速、准确诊断橡胶苗叶片磷含量。
|
| 关 键 词： | 神经网络 光谱分析 模型 偏最小二乘回归 微分技术 营养诊断 |
| 收稿时间： | 2015/4/27 0:00:00 |
| 修稿时间： | 7/8/2015 12:00:00 AM |
Prediction of leaf phosphorus contents for rubber seedlings based on hyperspectral sensitive bands and back propagation artificial neural network |
| |
| Guo Pengtao,Su Yi,Cha Zhengzao,Lin Qinghuo,Luo Wei and Lin Zhaomu.Prediction of leaf phosphorus contents for rubber seedlings based on hyperspectral sensitive bands and back propagation artificial neural network[J].Transactions of the Chinese Society of Agricultural Engineering,2016,32(Z1):177-183. |
| |
| Authors: | Guo Pengtao Su Yi Cha Zhengzao Lin Qinghuo Luo Wei and Lin Zhaomu |
| |
| Institution: | Rubber Research Institute, Chinese Academy of Tropical Agriculture Science, Danzhou 571737, China,Rubber Research Institute, Chinese Academy of Tropical Agriculture Science, Danzhou 571737, China,Rubber Research Institute, Chinese Academy of Tropical Agriculture Science, Danzhou 571737, China,Rubber Research Institute, Chinese Academy of Tropical Agriculture Science, Danzhou 571737, China,Rubber Research Institute, Chinese Academy of Tropical Agriculture Science, Danzhou 571737, China and Rubber Research Institute, Chinese Academy of Tropical Agriculture Science, Danzhou 571737, China |
| |
| Abstract: | Abstract: Traditional chemical analysis for the determination of leaf phosphorus content is time consuming, expensive and labor-intensive. Hyperspectral remote sensing has the potential to rapidly and accurately predict the element concentrations or biochemical composition in foliage. Thus, the aim of this study was to test the utility of hyperspectral sensitive bands in combination with back propagation artificial neural network to estimate leaf phosphorus contents for rubber tree seedlings. A sand culture experiment was carried out to grow rubber tree seedlings. These rubber tree seedlings were cultivated with Hoagland's nutrient solutions. Hoagland's nutrient solutions were set at five levels of phosphorus concentration. They were 31 mg/kg (P4), 23.25 mg/kg (P3), 15.5 mg/kg (P2), 7.75 mg/kg (P1), and 0 mg/kg (P0), respectively. Each level was replicated five times, and each replicate included five seedlings. Leaves of rubber tree seedlings were sampled at 85, 100, 115, and 133 days after start of the culture. At each sampling time, two matured leaves were collected for each tree. Ten leaves were obtained from each replicate and were combined as a sample. Finally, a total of 100 samples were collected. At each sampling date, leaf samples were sent to laboratory soon after the collection, and their leaf hyperspectral reflectance was measured by ASD FieldSpec 3 spectrometer. Phosphorus contents of the corresponding leaves were also analyzed using the conventional chemical analysis method. A second order low-pass digital Butterworth filter with normalized cutoff frequency 0.5 was used to the original spectra to filter out the noise signals. Next, differential technology was applied to the denoising of leaf hyperspectral reflectance in order to extract the first and the second derivative spectra, respectively. After that, correlation analysis was conducted to calculate correlation coefficients between rubber tree seedling leaf phosphorus contents and leaf hyperspectral reflectance as well as its first and second derivative spectra. Finally, hyperspectral sensitive bands were selected based on the results of the correlation analysis. These selected hyperspectral sensitive bands were used as input variables, and multiple linear regression (MLR), partial least-squares regression (PLSR), as well as back propagation artificial neural network (BPANN) model were used to estimate the leaf phosphorus contents for rubber tree seedlings, respectively. Results indicated that wavelength of 555 and 722 nm of the original hyperspectral reflectance, wavelength of 674, 710, 855, 1091, 1197, 1275, 1718, 2181, and 2228 nm of the first derivative spectra, and wavelength of 816, 890, 1339, 1357, and 2201 nm of the second derivative spectra were the hyperspectral sensitive bands. BPANN model with the selected 16 hyperspectral sensitive bands had the best predicted results. Correlation coefficients (r) between predicted leaf phosphorus contents and measured leaf phosphorus contents were 0.964 and 0.967 for training dataset and test dataset, respectively. Values of root mean squared errors (RMSE) were 0.01039 and 0.00856 for training dataset and test dataset, respectively. Values of relative error (RE) were 4.51% and 4.08% for train dataset and test dataset, respectively, and values of ratio of performance to deviation (RPD) were 3.71 and 3.23 for training dataset and test dataset, respectively. The results demonstrated that hyperspectral remote sensing could be used to rapidly, and accurately predict the leaf phosphorus contents for rubber seedlings. |
| |
| Keywords: | neural network spectrum analysis models partial least-squares regression differential technology nutrient diagnosis |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
| 点击此处可从《农业工程学报》浏览原始摘要信息 |
| 点击此处可从《农业工程学报》下载免费的PDF全文 |
|
