光谱学与光谱分析, 2020, 40 (1): 247, 网络出版: 2020-04-04  

基于参数优化SVM方法识别盐生植被钠离子光谱特征

Identification of Sodium Ion Spectral Characteristics of Halophytes Based on Parameter Optimized SVM Method
邓来飞 1,2,*张飞 1,2,3齐亚霄 1,2袁婕 1,2
作者单位
1 新疆大学资源与环境科学学院智慧城市与环境建模自治区普通高校重点实验室, 新疆 乌鲁木齐 830046
2 新疆大学绿洲生态教育部重点实验室, 新疆 乌鲁木齐 830046
3 中亚地理信息开发利用国家测绘地理信息局工程技术研究中心, 新疆 乌鲁木齐 830002
摘要
新疆盐渍土地分布广、 面积大, 在这些盐渍土地上生长着多种类型的盐生植物, 它们对改良盐渍土地、 维护生态稳定、 促进生态平衡具有重要的现实意义。 有关研究表明, 许多盐土植物大量吸收钠, 钠与钾都能增加细胞渗透压, 以适应高盐环境, 产生膨压而促进细胞的伸长, 因而对其生长是有益的, 能部分代替钾的功能。 因此掌握盐生植物的钠特征, 有助于了解盐生植物对生态环境的长期适应和响应, 使用高光谱技术实现有效诊断叶片钠特征。 首先, 对实测冠层高光谱数据, 采用离散小波变换(DWT)和db5小波对原始光谱进行9层小波分解, 求取最佳分解层数为5层。 其次, 对光谱数据进行5层db5小波分解, 对分解后的高频分量和低频分量建立了小波植被指数, 筛选出可敏感表征钠离子含量的小波植被指数。 最后, 利用SVR, LS-SVR, PSO-SVR和PSO-LS-SVR模型建立盐生植被钠离子含量的估算模型, 并与由原始光谱构建的光谱植被指数建立的估算模型进行比较。 此外, 引入偏最小二乘回归模型PLSR作为对比, 评价参数优化的支持向量机方法在高光谱技术估测盐生植被叶片钠离子含量的优势。 结果表明: (1)5种模型预测结果表明, PSO能有效优化SVR和LS-SVR模型参数(c, g), 提高模型精度和预测能力, 优化后的模型具有预测精度高、 泛化能力强以及稳健性能好等特点。 (2)综合小波指数构建的模型是综合多尺度、 多分辨率数据的反演模型, 其能从不同侧面反映植被的信息, 因而综合小波指数构建的4种模型优于单一小波指数构建的模型。 (3)对比两种类型的植被指数反演结果, 单一小波植被指数构建Na+含量的预测模型可取得较好的预测效果, 单一光谱指数估测Na+含量效果不佳, 这是因为小波变换可以减少原始光谱的噪声, 凸显光谱的细节信息, 增强其反演Na+含量的精度; 综合小波植被指数构建的模型精度和预测效果优于综合光谱指数构建的模型, 原始光谱经小波变换后, 可凸显更多的细节信息, 提高高光谱反演叶片Na+含量的能力。
Abstract
There are a wide range of saline soils in Xinjiang. It covers a large area. Various types of halophytes which have prominent significance for improving the saline lands, maintaining ecological stability and promoting ecological balance grow on these saline soils. Studies have shown that many halophytes absorb a large amount of sodium. Both sodium and potassium can increase the cell osmotic pressure to adapt to the high-salt condition, producing turgor pressure and promoting cell elongation. So it is beneficial to its growth and can partially replace the function of potassium. Thus, mastering the sodium characteristics of halophytes is helpful to understand the long-term adaptation and response of halophytes to the ecological environment. In this paper, HyperSpectral technique was used to effectively explore the characteristics of leaf sodium. Firstly, the discrete wavelet transform (DWT) and db5 wavelet were used to decompose the original spectral in 9 layers, and the optimal decomposition layer is 5 layers. Secondly, the original spectral were decomposed by db5 wavelet in 5 layers, and the wavelet vegetation indices were created by the decomposed high-frequency components and low-frequency components. We selected the wavelet vegetation indices which could sensitively characterize sodium ion content of halophytes. Finally, the SVR, LS-SVR, PSO-SVR and PSO-LS-SVR models were used to estimate the sodium ion content of halophytes vegetation. The results were compared to the models created by the spectral vegetation indices of original spectral. In addition, we used the partial least squares regression model as a comparison to evaluate the advantages of the parameter-optimized support vector machine method in estimating the sodium ion content of the leaves of the halophytes vegetation using hyperspectral techniques. The results showed that: (1) The prediction results of the five models showed that PSO can effectively optimize the parameters (c, g) of SVR and LS-SVR models, and improve the accuracy and prediction ability of the models. The optimized models had the advantages of high prediction accuracy, strong generalization ability and good robustness performance. (2) The model constructed by the multiple wavelet index was an inversion model of integrated multi-scale and multi-resolution data, which can reflect the vegetation information from different aspects. Therefore, the four models constructed by the multiple wavelet index were superior to the models constructed by the single wavelet index. (3) Contrasting the inversion results of two types index, the Na+ content prediction model constructed by a single wavelet vegetation index can achieve a better prediction results. The single spectral index is not effective in estimating Na+ content, which is because the wavelet transform can reduce the noise of the original spectral and highlight the detailed feature of the spectral, improving the prediction results. The model accuracy and prediction effect of the integrated wavelet vegetation index were better than those of the integrated spectral index. More spectral subtle feature can be highlighted by wavelet transform, thus improving the ability of retrieving Na+ content in leaves by HyperSpectral method.

邓来飞, 张飞, 齐亚霄, 袁婕. 基于参数优化SVM方法识别盐生植被钠离子光谱特征[J]. 光谱学与光谱分析, 2020, 40(1): 247. DENG Lai-fei, ZHANG Fei, QI Ya-xiao, YUAN Jie. Identification of Sodium Ion Spectral Characteristics of Halophytes Based on Parameter Optimized SVM Method[J]. Spectroscopy and Spectral Analysis, 2020, 40(1): 247.

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!