光谱学与光谱分析, 2020, 40 (2): 556, 网络出版: 2020-05-12  

SVM和BP检测滨海湿地土壤有机质

Estimation of Soil Organic Matter in Coastal Wetlands by SVM and BP Based on Hyperspectral Remote Sensing
作者单位
1 江苏海洋大学测绘与海洋信息学院, 江苏 连云港 222005
2 河南财经政法大学资源与环境学院, 河南 郑州 450046
摘要
近年来, 虽然随着高光谱技术的出现可以快速获取土壤中的养分含量, 但不同的土壤类型对估算的精度会有很大的差异。 滨海湿地土壤类型受海洋环境影响较大, 其高光谱反射率与内陆土壤类型的表现会有所不同, 也就造成了同样的估算模型在反演滨海湿地土壤的养分含量时, 反演精度的降低, 随着近年来海洋资源的开发与滨海湿地生态恢复工作的不断推进, 探索一种合适的估算模型来快速准确的获取土壤中的养分含量变得更加紧迫。 该研究旨在验证利用可见-近红外高光谱反射率构建非线性模型来反演滨海湿地土壤类型中有机质(soil organic matter, SOM)含量的可行性。 以江苏省盐城大丰麋鹿国家级自然保护区的第三核心区土壤作为研究对象, 将土壤样本的光谱反射率进行5点Savitzky-Golay(S-G)平滑滤波处理, 再进行一阶微分R′、 倒数的一阶微分(1/R)′、 倒数的二阶微分(1/R)″、 对数的一阶微分(lgR)′四种微分变换后, 应用相关系数和显著性水平(p<0.01)提取土壤有机质含量的敏感波段, 利用台湾大学林智仁教授开发的MATLAB软件中的LIBSVM工具包构建SVM(support vector machine)支持向量机估算模型, 并利用MATLAB2018b软件中自带的BP(back propagation)反向传播神经网络构建估算模型, 最后利用决定系数R2和均方根误差RMSE进行模型的预测精度验证。 结果表明: 原始光谱通过5点S-G平滑滤波、 微分变换与相关系数法可以较好的提取出有效波段, 其中基于(1/R)′光谱变换提取的滨海湿地土壤有机质特征波段为498~501, 1 180~1 182, 1 946, 1 947和2 323~2 326 nm; 对比发现SVM的估算精度优于BP神经网络; 利用光谱的(1/R)′微分形式构建的SVM模型估算滨海湿地土壤SOM含量的精度最高, 决定系数R2与RMSE分别为0.93和0.23, 并且均通过了p<0.01的显著性检验。 因此利用高光谱构建SVM非线性模型来快速估算滨海湿地土壤中的养分含量具有一定的可行性。
Abstract
In recent years, although the nutrient content in the soil can be quickly obtained with the emergence of hyperspectral technology, but different soil types have great differences in the accuracy of estimation. The soil type of coastal wetland is greatly affected by the marine environment, and its hyperspectral reflectance and inland soil type will be different. This will reduce the precision in the same estimation model when inverting the nutrient content of coastal wetland soil types. With the development of marine resources and the ecological restoration of coastal wetlands in recent years, it is urgent to explore a suitable estimation model to quickly and accurately obtain nutrient content in soil. This study aimed to verify the use of visible-near infrared hyperspectral reflectivity to construct a nonlinear model so as to invert the feasibility of organic matter (SOM) in coastal wetland soils. The topsoil in the third core area of Dafeng Elk National Nature Reserve in Yancheng, Jiangsu province was taken as the investigated object. The sensitive bands corresponding to Soil Organic Matter (SOM) content were retrieved based on correlation coefficient after 5 point S-G filtering and four differential transformations of R′, (1/R)′, (1/R)″, (lgR)′ by spectral reflectance of soil samples. The estimation models of SOM by Support Vector Machine (SVM) and BP neural network were determined, and the prediction accuracy of the model was verified by using the decision coefficient R2 and the root mean square error RMSE. The research results indicated that the effective bands can be identified by S-G filtering, differential transformation and correlation coefficient method based on the original spectra of soil samples. The characteristic bands of SOM based on transformations (1/R)′ were 498~501, 1 180~1 182, 1 946, 1 947, 2 323~2 326 nm. Estimation accuracy of SVM was better than that of BP neural network for SOM in Yancheng coastal wetland. The estimation model of SOM by SVM based on (1/R)′ spectra had the highest precision, with the determination coefficients (R2) and root mean square error (RMSE) of 0.93 and 0.23. Therefore, it is suitable to use hyperspectral remote sensing to quickly estimate the nutrient contents of topsoil in coastal wetland.

张森, 卢霞, 聂格格, 李昱蓉, 邵亚婷, 田燕芹, 范礼强, 张钰娟. SVM和BP检测滨海湿地土壤有机质[J]. 光谱学与光谱分析, 2020, 40(2): 556. ZHANG Sen, LU Xia, NIE Ge-ge, LI Yu-rong, SHAO Ya-ting, TIAN Yan-qin, FAN Li-qiang, ZHANG Yu-juan. Estimation of Soil Organic Matter in Coastal Wetlands by SVM and BP Based on Hyperspectral Remote Sensing[J]. Spectroscopy and Spectral Analysis, 2020, 40(2): 556.

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!