基于变量优选和ELM算法的土壤含水量预测研究

引用本文

蔡亮红, 丁建丽. 基于变量优选和ELM算法的土壤含水量预测研究[J]. 光谱学与光谱分析, 2018,38(7): 2209-2214.
CAI Liang-hong, DING Jian-li. Prediction for Soil Water Content Based on Variable Preferred and Extreme Learning Machine Algorithm[J]. Spectroscopy and Spectral Analysis, 2018,38(7): 2209-2214.
Doi:10.3964/j.issn.1000-0593(2018)07-2209-06 复制到剪切板

Permissions

《光谱学与光谱分析》期刊社所有

基于变量优选和ELM算法的土壤含水量预测研究

蔡亮红^1,², 丁建丽^1,^2,^*

1. 新疆大学资源与环境科学学院, 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

*通讯联系人　　e-mail: 2187736938@qq.com

作者简介: 蔡亮红, 1991年生, 新疆大学资源与环境科学学院硕士研究生 e-mail: 1173716776@qq.com

收稿日期: 2017-08-14

基金: 国家自然科学基金项目(41771470, U1303381), 自治区重点实验室专项基金项目(2016D03001), 自治区科冀支疆项目(201591101), 教育部促进与美大地区科研合作与高层次人才培养项目

摘要

土壤水分含量(SMC)的快速估测对干旱半干旱地区的精准农业发展具有重要的意义。以渭干河-库车河绿洲为靶区, 采用小波变换(WT)对反射光谱进行1~8层小波分解, 通过相关性分析确定最大分解层数, 再通过竞争性自适应重加权(CRAS)、连续投影算法(SPA)和CARS-SPA耦合算法进行特征波长筛选。基于全波段构建BP神经网络模型和基于特征波长构建BP神经网络、支持向量机、随机森林和极限学习机模型, 并进行对比分析。结果显示: (1)随着小波分解的进行, 总体上L6在去噪的同时还尽可能的保留了光谱原始特征, 为最大分解层; (2)小波变换和CARS-SPA算法的结合使其在建立模型时较为彻底的去除噪声和无信息变量, 同时消除变量间的共线性; (3)在所有的SMC预测模型中, 相对于BP神经网络、 SVM, ELM和RF具有更好的预测能力, 其中L6-CARS-SPA-ELM精度最高, 其RMSEC=0.015 1,Rc2=0.916 6, RMSEP=0.014 2,Rp2=0.935 4, RPD=2.323 9。这体现出ELM预测模型对非线性问题的强解析能力和模型的稳健性, 为该研究区SMC的预测提供新的思路。

关键词: 光谱学; 土壤水分; 小波变换; 变量优选; 极限学习机

中图分类号:S127 文献标志码:A

Prediction for Soil Water Content Based on Variable Preferred and Extreme Learning Machine Algorithm

CAI Liang-hong^1,², DING Jian-li^1,^2,^*

1. College of Resources & Environmental Science, Xinjiang University, Urumqi 830046, China;

2. Key Laboratory of Oasis Ecology, Xinjiang University, Urumqi 830046, China

*Corresponding author

Abstract

The rapid estimation of soil moisture content (SMC) is of great significance to precision agriculture in arid areas, and hyperspectral remote sensing technology had been widely used in the estimation of soil moisture content due to its non-destructive, rapid, and high spectral resolution characteristics. Meanwhile, there are many prediction models of soil moisture content, such as BP, SVM, RF and so on, but the prediction model has some shortcomings. Recently, the extreme learning machine(ELM) as a new algorithm began to emerge in the field of soil property prediction. In the present study, a total of 39 soil samples at 0~20 cm depth were collected from delta oasis in Weigan-Kuqain, Xinjiang Province. We brought back to the laboratory to dry it naturally, groundnd and passed through a 2 mm hole scree, and then the sample holders were clear black boxs in 12 cm diameter and 1.8 cm deep, which were filled and leveled at the rim with a spatula. Reflectance of soil samples were measured using ASD Fieldspec 3 Spectrometer in a dark room. We used the following steps to process soil reflectance: First, discrete wavelet transformation (DWT) was used to decompose the original spectral in 8 levels using db4 wavelet basis by MATLAB programming language. In order to select the maximum level of DWT, correlation coefficients between SMC and the spectra of each level was computed. Secondly, On the basis of wavelet transform, CARS (the adaptive variable weighting algorithm), SPA (successive projections algorithm) and CARS-SPA were used to filter the redundant variables, the wavelength variables with better correlation with SMC were screened out. Thirdly, On the basis of the preferred wavelengths, BP neural network, SVM (support vector machine), RF (random forest) and ELM (extreme learning machine) prediction models were employed to build the hyperspectral estimation models of SMC, and the advantages and disadvantages of the model were further analyzed. Statistical parameters of root mean square error of calibration (RMSEC), determination coefficient of calibration (Rc2), root mean square error of prediction (RMSEP), determination coefficient of predicting (Rp2) and relative prediction deviation (RPD) were selected as comparison criteria. The results showed that: (1) With the increase of the number of decomposed layers, the correlation between soil reflectance and SMC showed a trend of increasing first and then decreasing, and L6 was the most significant band at 0.01 level. In general, the characteristic spectrum of L6 was denoised at the same time, and the spectral detail was preserved to the maximum extent. So the maximum decomposition order of the wavelet was 6 order decomposition; (2) On the basis of L6, the CARS, SPA and CARS-SPA algorithms were used to optimize the variables, and the number of selected wavelength variables were 81, 23 and 12, respectively. The predictive models constructed by three algorithms were better than those of the whole-band model. The prediction model based on the CARS-SPA was the most accurate in the corresponding model. It can be seen that the CARS-SPA coupling algorithm not only simplified the model complexity, but also increased the robustness of the model; (3) Compared with the BP, SVM, RF and ELM, In all the SMC predicting models, there were 6 models with predictive ability, Sort by: L6-CARS-SPA-ELM>L6-CARS-SPA-RF>L6-CARS-ELM>L6-CARS-RF>L6-SPA-ELM>L6-SPA-RF. Results showed that ELM performed much better than BP, SVM and RF in predicting SMC in this study. At the same time, the L6-CARS-SPA-ELM model had the highest accuracy, and the model had RMSEC=0.015 1,Rc2=0.916 6, RMSEP=0.014 2,Rp2=0.935 4, RPD=2.323 9. It was shown that the combination of wavelet transform and CARS-SPA algorithm made it possible to remove the noise as much as possible and to remove the noise completely when the model was established. At the same time, and ELM model was a new method to predict other soil properties.

Keyword: Spectroscopy; Soil moisture; WT (wavelet transformation); Variable selection; ELM (extreme learning machine)

文章图片

引言

土壤水分含量(soil moisture content, SMC)是土壤系统中物质和能量循环的载体, 对土壤特性、植被生长分布以及区域生态系统有着重要的影响^{[1, 2]}。然而, 土壤水分非常容易受环境的影响, 并且SMC传统的监测方法由于费时费力, 难以用于田间尺度和满足精准农业的要求^[3], 因此, SMC的监测需要一种高效、精准的方法。近年来, 高光谱遥感技术以其大面积、非接触、时效性等优势, 在SMC的监测研究中得到重视^[4]。然而, 通过高光谱技术所获取的土壤光谱原始数据存在明显的光谱噪声和严重的散射现象, 土壤高光谱中必然存在与SMC不相关的噪声^[5], 这将会增加SMC的探测难度, 再加上高光谱本身的海量数据, 因此, 较为彻底的去噪和变量优选成为建立精度较高模型的关键。

研究以渭库-绿洲土壤样品为研究对象, 基于小波变换(wavelet transform, WT)对反射光谱进行分解, 并根据各层特征光谱与SMC之间的相关关系来确定最大分解尺度, 在此基础上引入竞争性自适应重加权(competitive adaptive reweighted sampling, CRAS)、连续投影算法(successive projections algorithm, SPA)和CARS-SPA耦合算法等三种变量优选方法进行特征波段的筛选, 并建立(extreme learning machine, ELM)预测模型, 为了检验模型的有效性, 将ELM模型与基于BP神经网络、支持向量机(support vector machine, SVM)和随进森林(random forest, RF)的土壤水分含量预测模型进行比较, 以期选择最优模型, 为土壤水分含量等研究及当地精准农业提供科学支撑和参考。

1 实验部分

1.1 土壤样本

以新疆塔里木盆地中北部的渭-库绿洲作为实验区, 并且依据其土壤、地形等特征, 布设了39个采样点, 同时通过GPS将采样点的坐标进行记录, 目的在于后期验证。每个采样点采用5点混合法进行采样, 采集深度为0~20 cm, 每个采样点采集2份样本, 分别装入铝盒和塑料袋中, 将铝盒中的土壤样本带回实验室进行土壤含水量的测定, 测定方法为烘干法, 即将土壤样本置于温度为105 ℃恒温箱中进行48 h烘干。另一份样本放置在遮阴处, 避免土壤水分的流失, 以保持土壤样本的野外实际状态, 并及时获取其光谱数据。光谱数据的详细测定方法见文献[6]中1.1。

1.2 小波变换

由于原始光谱数据存在大量的噪声信息, 需要对其进行去噪, 而小波变换作为一种新的数据处理方法, 其具备傅里叶分析方法的优势, 同时又填补了该方法的缺点^[7], 小波变换的表达式见文献[6]中1.2。于雷等^[8]的研究结果表明, 在进行小波变换时, 当小波母函数为db4时的分解效果最佳。本研究利用这个结论对原始光谱数据进行8层分解, 同时获取其相应的特征光谱(L1~L8)。

1.3 特征波长选择算法

由于高光谱数据量大, 所以小波系数重构后仍存在大部分冗余和共线信息, 导致计算量大和模型复杂, 对光谱的定量分析造成严重影响^[9]。采用CARS算法, SPA算法和CARS-SPA耦合算法筛选最优变量, 降低数据的共线性和冗余, 使构建的模型简化。 CARS算法借助自适应重加权采样技术(adaptive reweighted sampling, APS)、指数衰减函数(exponentially decreasing function, EDP)和十折交互检验确定最优变量子集, 它可筛选出对土壤属性较敏感的波长变量, 并可以解决变量筛选时的组合爆炸问题。 SPA算法是一种前向选择算法, 当校正模型的交互验证均方根误差RMSECV最低时, 相应的波长变量集为筛选的共线性最小的最佳特征波长变量集。 CARS-SPA耦合算法是通过CARS算法先筛选出土壤属性的特征波长, 但波长变量之间仍然存在共线性, 为了更加有效的筛选出最少的特征变量, 经CARS筛选特征波长后再利用SPA进行二次特征波长筛选。

1.4 建模分析方法与精度评价

引入ELM构建SMC预测模型, 并且比较BP, SVM, RF和ELM模型的稳定性。由于BP, SVM和ELM严格意义上均属于神经网络系列的预测模型, 而RF属于非线性回归模型, 因此选择两种具备代表性的本质同类算法BP和SVM, 一种本质异类算法RF, 使得构建的模型更具有现实意义。

BP是多层前馈神经网络^[10], 作为常用的预测模型中的一种, 其结构简练、准确度较高; SVM最早是一种用有导师学习方式来解决二分类问题的模型, 它能找出多维空间数据的分离最优超平面, 并尽可能让所有训练集到该超平面的误差最小^[11]; RF模型是建立在决策树基础上, 通过多次bootstrap 抽样获得随机样本, 并通过这些样本分别建立相对应的决策树, 对于回归问题, 其最终的结果为所有决策树得到的结果的均值^[12], RF模型建立的过程中涉及到2个关键的参数: 决策树的数量(ntree)和节点分裂的次数(mtry), 本文预测设定ntree为1 000, mtry为2; ELM算法是近年提出的单隐层前馈神经网络^[13], 它在算法执行过程中随机设定输入层与隐含层之间的权值和阈值, 不用重复更新迭代网络的输入权值以及隐元的偏置, 并可以得到最优解, 因此其运算时间短、移植性强^[14]。

为了评估模型的效果, 选择如下参数来进行评价, 如决定系数(determination of coefficients, R²)、均方根误差(root mean square error, RMSE)和相对分析误差(residual predictive deviation, RPD), 即相对于建模集来说, 分别为 $R_{p}^{2}$ , RMSEC, 相对于预测集来说, 分别为 $R_{c}^{2}$ , RMSEP和RPD。 R²与模型稳定性成正比, 而RMSE与其呈反比。 RPD表示预测效果, 分为三个等级, 分别为较好(RPD≥ 2)、一般(1.4≤ RPD< 2)和无预测能力(RPD< 1.4)^[15]。

2 结果与讨论

2.1 样本土壤含水量状况

通过对研究区SMC进行统计可知, 如表1。其中建模集样本和验证集样本的土壤水分含量的平均值分别为0.146和0.148, 其整体样本土壤含水量平均值为0.147, 变异系数(coefficient of variation, CV)为0.388, 属于中等变异。

表1 土壤样品SMC统计特征 Table 1 Statistical characteristics of SMC of soil samples

2.2 小波变换最大分解尺度

通过db4母函数对土壤光谱进行8层分解, 同时对相应分解层的小波系数进行重构, 得到L1— L8, 共8层特征光谱, 分解结果如图1。在图1(a)(L0为原始光谱)中有2个较为明显的水分吸收峰(1 400和1 900 nm)和2个较为微弱的吸收峰(450和2 200 nm)。图1(b)中由于原始光谱反射率自身的噪声进行传递, 使得在350~400 nm表现为“ 小毛刺” , 但是当分解到图1f时, 由于噪声传递被削弱, 导致此时的噪声已经被大量去除; 在小波分解的过程中, 由于细节信息不断被滤除, 使得光谱曲线越来越平滑, 同时一些土壤水分特征峰也逐渐消失, 在图1(g)中1 400和1 900 nm处仍然有显著的吸收峰, 但是在图1(h)中完全不能呈现。

	Figure Option View Download New Window
	图1 小波变换1— 8层重构光谱Fig.1 Reconstruction spectra of original spectrum under 1— 8 wavelet level

对小波变换后L1— L8与SMC进行相关性分析可知(表2), 表中相关性均通过0.01(阈值为± 0.408)的显著性检验, 其中L1与SMC两者之间有393个波段通过显著性, 在L1~L6中其显著性波段数量呈现增加趋势, 当进行到L6时其波段数量为602个, 达到最大, 当进行到L4时相关系数为0.619, 达到最大正相关, L6之后其显著波段数量及其相关系数发生骤减。总的来说, L6在去噪的同时还尽可能的保留了光谱原始特征。故本研究选择第6层为最大分解层数, 同时以L6为基础进行后续分析。

表2 SMC与各层特征光谱相关分析 Table 2 Correlation analysis between SMC and spectra from wavelet analysis in each level

2.3 特征波长选取

在小波变换后L6的基础上, 分别基于CARS, SPA和CARS-SPA算法筛选SMC的特征波长, 如图2所示。从图2可知, 随着CARS算法进行特征波长筛选的进行, 在1~36次运行中, RMSECV逐渐趋于0, 这体现出L6与SMC之间的大量的无效信息被滤除, 当运行次数达到36次以后, 其值开始增大, 可能是由于在波长变量筛选的同时, 也滤除了与SMC相关的主要变量。总体上来说, 在运行36次时RMSECV最小, 为0.012 6, 所对应的特征波长为81个。图2(b)和(c)分别是利用SPA算法对全波段和CARS筛选结果进行二次筛选的RMSE曲线图, 随着筛选变量的增加, RMSE整体上呈下降趋势, 图2(b)和(c)中空心方块表示通过SPA算法筛选出的最优特征波长, 分别为23和12个。

	Figure Option View Download New Window
	图2 变量优选过程Fig.2 Variable optimization process

由表3可知, 在小波变换后L6的基础上, 通过CARS算法筛选出的特征波长数量要比用SPA, CARS-SPA算法筛选出的要多。 SPA算法筛选出23个波长变量, 占全波段的1.07%, 占CARS算法筛选变量的28.40%。在经过CARS和SPA算法后, 所筛选的变量个数均减少, 本研究在CARS算法筛选变量的基础上引入SPA算法进行变量的二次筛选, 得到12个波长变量, 占全波段的0.56%, 占CARS的14.81%, 占SPA的52.17%, 使得波长变量个数大大减少。

表3 L6特征光谱变量优选结果 Table 3 The optimal variables of the spectral characteristics of the levels of L6

2.4 基于特征波长建模及其比较分析

在小波变换后L6的基础上, 基于CARS, SPA和CARS-SPA算法筛选出SMC的特征波长分别为81, 23和12个, 并且分别构建基于特征波段的BP, SVM, RF和ELM四种SMC的预测模型, 为了更好的突出变量优选的优势, 引入全波段的BP模型进行对比, 结果见表4。从表4可以看出, 基于全波段的BP预测模型的精度是最低的, 说明通过变量筛选后的模型的建模波段大大减少, 其稳定性更好。

表4 土壤水分含量预测结果 Table 4 Results of estimation for SMC

Model	Variable selection methods	Variable number	Calibration sets		Prediction sets
Model	Variable selection methods	Variable number	RMSEC	$R_{c}^{2}$	RMSEP	$R_{p}^{2}$	RPD
BP	Full L6-CARS L6-SPA L6-CARS-SPA	2 151 81 23 12	0.045 2 0.021 1 0.024 0 0.031 2	0.682 5 0.927 5 0.868 7 0.813 9	0.045 6 0.035 0 0.044 8 0.030 0	0.681 4 0.814 7 0.761 3 0.844 4	0.723 7 0.942 9 0.736 6 1.100 0
SVM	L6-CARS L6-SPA L6-CARS-SPA	81 23 12	0.018 5 0.018 3 0.018 5	0.832 6 0.838 5 0.833 2	0.031 9 0.044 4 0.025 2	0.865 7 0.869 3 0.872 1	1.034 5 0.743 2 1.309 5
RF	L6-CARS L6-SPA L6-CARS-SPA	81 23 12	0.024 0 0.022 0 0.024 1	0.930 6 0.940 4 0.936 6	0.020 8 0.023 8 0.015 5	0.902 9 0.892 7 0.925 5	1.586 5 1.386 6 2.129 0
ELM	L6-CARS L6-SPA L6-CARS-SPA	81 23 12	0.019 7 0.011 3 0.015 1	0.911 2 0.940 9 0.916 6	0.015 6 0.023 3 0.014 2	0.916 7 0.897 2 0.935 4	2.115 4 1.416 3 2.323 9

表4 土壤水分含量预测结果 Table 4 Results of estimation for SMC

在相同的SMC预测模型中, 对于不同筛选方法均表现为: L6-CARS-SPA> L6-CARS> L6-SPA的排列顺序, 相对于CARS算法来说, SPA算法所筛选的特征波长较少, 但基于SPA算法所构建的模型中精度最高的L6-SPA-ELM的RPD仅为1.416 3, 模型精度较差, 这是由于光谱数据存在大量与SMC不相关的冗余信息, 而其与真正有效信息呈非共线关系, 以致SPA算法获得的23个特征波长中存在部分冗余信息, 从而降低了SPA相应模型的预测精度。而CARS-SPA算法获得的12个特征波长所构建的模型的精度均要高于相应模型, 这说明利用SPA算法对CARS算法所获取的特征波段进行二次筛选, 不仅可以简化模型, 而且可以提升模型精度。

在不同的SMC模型中, 其RPD≥ 2的较好模型共三种, 按其预测精度排序为: L6-CARS-SPA-ELM> L6-CARS-SPA-RF> L6-CARS-ELM, 而1.4≤ RPD< 2效果一般的预测模型也有三种, 按其预测模型预测精度排序为: L6-CARS-RF> L6-SPA-ELM> L6-SPA-RF, 其余模型基本无预测能力。其中基于L6-CARS-SPA-ELM模型预测精度最高, 其RMSEC=0.015 1, $R_{c}^{2}$ =0.916 6, RMSEP=0.014 2, $R_{p}^{2}$ =0.935 4, RPD=2.323 9, 而在所有基于特征波长的预测模型中, L6-SPA-BP模型预测精度最低, 其预测集RPD仅为0.736 6。相对于BP, SVM和RF, ELM的预测能力最高, RF次之。可能是由于BP难以获取全局最优解, 易陷于局部极小的状态, 容易产生过拟合和稳健性不足等问题, SVM在很大程度上核函数的选择, RF拥有调参少、高精度和移植性好等优势, 不容易发生“ 过度拟合” , 但是随着特征学习增量及树的棵树增多, RF在保证预测精度的同时运算效率明显下降。而ELM具有训练误差小、速度快、泛化性能强, 这体现出ELM预测模型对非线性问题的强解析能力和模型的稳健性, 使得ELM在SMC预测中凸显出来。

综上所述, 在小波变换后L6的基础上, 相对于CARS和SPA, 在CARS算法基础上进行SPA算法的二次筛选, 其建模波段最少, 而模型稳定性较好, 这有助于简化模型结构, 并且提升稳定性。在不同的预测模型中, L6-CARS-SPA-ELM模型的预测效果最优, 其次是L6-CARS-SPA-RF。

图3为小波变换后L6, 基于CARS-SPA算法进行变量筛选后的ELM预测模型建模集和预测集检验结果。

	Figure Option View Download New Window
	图3 实测值与估算值的比较Fig.3 Comparison between measured and estimated SMC values

3 结论

利用小波变换进行预处理, 并引入CARS, SPA和CARS-SPA算法进行特征波长优选, 并基于所选的特征波长构建BP, SVM, RF和ELM四种SMC预测模型, 通过比较分析, 得到如下结论:

(1)随着小波分解的进行, 总体上L6在去噪的同时还尽可能的保留了光谱原始特征, 为最大分解层。

(2)在L6的基础上, 引入CARS, SPA和CARS-SPA算法进行特征波长优选, 总体上, 基于优选波长所构建的模型的精度均比全波段要高, 其中基于L6-CARS-SPA构建的模型在相应的SMC预测模型中的精度最高, 说明小波变换和CARS-SPA算法的结合使其在建立模型时较为彻底的去除噪声和无信息变量, 并消除变量间的共线性。

(3)在所有的SMC预测模型中, 具有预测能力的共六种, 排序为: L6-CARS-SPA-ELM> L6-CARS-SPA-RF> L6-CARS-ELM> L6-CARS-RF> L6-SPA-ELM> L6-SPA-RF, 由此可知, 相对于BP, SVM, ELM和RF的预测能力更好, 其中L6-CARS-SPA-ELM精度最高, 其RMSEC=0.015 1, $R_{c}^{2}$ =0.916 6, RMSEP=0.014 2, $R_{p}^{2}$ =0.935 4。这体现出ELM预测模型对非线性问题的强解析能力和模型的稳健性, 为该研究区SMC的预测提供新的思路。

The authors have declared that no competing interests exist.

参考文献

文献列表

[1]	ZOU Wen-xiu, HAN Xiao-zeng, JIANG Heng, et al(邹文秀, 韩晓增, 江恒, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2011, 27(9): 196. [本文引用:1]
[2]	ZHANG Ding-hai, LI Xin-rong, CHEN Yong-le(张定海, 李新荣, 陈永乐). Acta Ecologica Sinica(生态学报), 2016, 36(11): 3273. [本文引用:1]
[3]	SUN Yue-jun, ZHENG Xiao-po, QIN Qi-ming, et al(孙越君, 郑小坡, 秦其明, 等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2015, 35(8): 2236. [本文引用:1]
[4]	Yin Z, Lei T, Yan Q, et al. Computers & Electronics in Agriculture, 2013, 99(99): 101. [本文引用:1]
[5]	YU Lei, ZHU Ya-xing, HONG Yong-sheng, et al(于雷, 朱亚星, 洪永胜, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2016, 32(22): 138. [本文引用:1]
[6]	CAI Liang-hong, DING Jian-li(蔡亮红, 丁建丽). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2017, 33(16): 144. [本文引用:2]
[7]	GUO Yan, HOU Su-zhen, WANG Ping, et al(郭彦, 侯素珍, 王平, 等). Arid Zone Research(干旱区研究), 2015, 32(6): 1047. [本文引用:1]
[8]	Kaewpijit S, Moigne J L, Elghazawi T. Proceedings of SPIE-The International Society for Optical Engineering, 2002, 388(4): 56. [本文引用:1]
[9]	ZHANG Chu, LIU Fei, KONG Wen-wen, et al(张初, 刘飞, 孔汶汶, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2013, 29(20): 270. [本文引用:1]
[10]	WANG Tong-tong, ZHANG Jian, TU Chuan, et al(王彤彤, 张剑, 涂川, 等). Environmental Science & Technology(环境科学与技术), 2013, 36(8): 175. [本文引用:1]
[11]	ZHANG Qiang, HUANG Sheng-zhi, CHEN Xiao-hong(张强, 黄生志, 陈晓宏). Acta Pedologica Sinice(土壤学报), 2013, 50(1): 59. [本文引用:1]
[12]	Guo P T, Li M F, Luo W, et al. Geoderma, 2015, 237-238: 49. [本文引用:1]
[13]	Cambria E, Liu Q, Li K, et al. IEEE Intelligent Systems, 2013, 28(6): 30. [本文引用:1]
[14]	Deng C W, Huang G B, Jia X U, et al. Science China, 2015, 58(2): 20301. [本文引用:1]
[15]	XUE Li-hong, ZHOU Ding-hao, LI Ying, et al(薛利红, 周鼎浩, 李颖, 等). Acta Pedologica Sinica(土壤学报), 2014, 51(5): 993. [本文引用:1]

2011

0.0

ZOU

Wen-xiu

, HAN

Xiao-zeng

, JIANG

Heng

, et al(邹文秀, 韩晓增, 江恒, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2011, 27(9): 196.

Water is one of mainly limiting factors influencing agricultural productivity in black soil region of Northeast China which is rain-fed agriculture. In order to improve the sustainable soil water management, Standard precipitation index (SPI) was used to analysis precipitation conditions from 1952 to 2008, and the response of soil moisture to precipitation during 1999 to 2008 were considered based on long-term experiment in National Observation Station of Hailun Agroecology System, Chinese Academy of Sciences (CAS). Mean annual precipitation of 540 mm was observed in study site, with standard deviation of 121 mm and 473 mm occurring in growing season which accounted for 87.6% of mean annual precipitation. Seven dry/wet states were distinguished based on SPI(standard precipitation index), they are extremely dry state, severely dry state, moderately dry state, normal state, moderately wet state, very wet state and extremely wet state. Extremely dry state had a increasing trend during recent decade (1999-2008). Soil moisture at the end of growing season was increased remarkably in moderately wet year (2003, SPI=1.5), whereas, was in low level (512.2 mm) in extremely dry year (2004, SPI=-2.6), variation in annual precipitation also impacted the characteristics of soil water supply. More water stored in soil profile was consumed when chemical fertilizer and organic matter were applied, compared with no fertilizer. Therefore, there is great significance for the sustainable development of agricultural production to regulate the soil water according to the precipitation conditions.

水分是东北黑土区农业生产的主要限制因子，为了探讨东北黑土区可持续的土壤水分管理方式，该研究以中国科学院海伦农田生态系统野外科学观测研究站内为基础，利用SPI（standard precipitation index）值（标准降水指数）研究了该地区1952－2008年的大气降水情况，同时分析了土壤水分对大气降水（1999－2008年）的响应。1952－2008年平均年降水量为540 mm，标准差为121 mm，生长季内平均降水量为473 mm，占年平均降水量的87.60%。利用标准降水指数（SPI）将1952－2008年大气降水划分为7个不同的干/湿状态，分别是极端干旱，严重干旱、中等干旱、平常年份、中等湿润、非常湿润和极端湿润。分析近10 a（1999－2008）大气降水的SPI值发现极端干旱的年份有增加的趋势，同时对生长季末期土壤含水量的影响表现为湿润的年份（2003年，SPI值=1.5）能够显著增加土壤含水量（639.70 mm），而在极端干旱的年份（2004年，SPI=-2.6）土壤含水量达到了最低值（512.21 mm）。年际间降水量的差异也影响了土壤的供水特征特，同时肥料的施用增加了作物对土壤中水分的消耗。根据降水合理地管理东北黑土区的土壤水分对农业的可持续发展具有重要意义。

... 引言土壤水分含量(soil moisture content, SMC)是土壤系统中物质和能量循环的载体, 对土壤特性、植被生长分布以及区域生态系统有着重要的影响^[1,2] ...

2016

0.0

ZHANG

Ding-hai

, LI

Xin-rong

, CHEN

Yong-le

(张定海, 李新荣, 陈永乐). Acta Ecologica Sinica(生态学报), 2016, 36(11): 3273.

Shrubs were the most important dominant plants in desert areas of China. Sand binding shrubs not only are vital to stabilize sand dunes in the desert, but also are beneficial to improve ecological restoration in the desert ecosystem. Based on a long-term monitoring and study( > 50 years) at the Shapotou Desert Research and Experimental Station,which is located on the southeast edge of the Tengger Desert, we simulated the processes of the cover of sand-binding shrub and deep soil water content using an eco-hydrological model. The model revealed that the establishment of sand-binding shrub resulted in change in the hydrological processes comparing to the mobile sand dunes. The cover of sand-binding shrub and deep soil water content had reached to a new equilibrium after 50 years the sand-binding shrub established. The balance was the cover of sand-binding shrub remained stable at (10±0.9)% and the deep soil water content remained stable at (2.58±0.2)%. In the Shapotou area of the Tengger Desert, the coverage of shrubs and biological soil crust were about 10% and 60% respectively, and the deep soil water content maintained at 3%, reflecting the maximum soil water carrying capacity of the vegetation.

灌木是我国沙漠地区主要的优势植物类型,固沙灌木的存在有益于沙丘的固定进而有利于退化沙漠生态系统的生态恢复。基于腾格里沙漠沙坡头地区50多年的人工植被区长期观测研究,建立生态-水文模型模拟了该地区固沙灌木盖度和深层土壤水分的动态变化过程,结果表明,固沙灌木的建立改变了沙区原有的生态-水文过程,在固沙灌木建立40多年后,固沙灌木的盖度和深层土壤水分达到了新的平衡状态。灌木盖度逐渐稳定在(10±0.9)%左右,而深层土壤水分稳定在(2.58±0.2)%左右。因此,在年均降雨量为186mm的腾格里沙漠沙坡头地区,土壤水分的最大植被承载力为:灌木和生物土壤结皮的盖度分别维持在10%和60%,深层土壤水分维持在3%左右。

... 引言土壤水分含量(soil moisture content, SMC)是土壤系统中物质和能量循环的载体, 对土壤特性、植被生长分布以及区域生态系统有着重要的影响^[1,2] ...

2015

0.0

... 然而, 土壤水分非常容易受环境的影响, 并且SMC传统的监测方法由于费时费力, 难以用于田间尺度和满足精准农业的要求^[3], 因此, SMC的监测需要一种高效、精准的方法 ...

2013

0.0

... 近年来, 高光谱遥感技术以其大面积、非接触、时效性等优势, 在SMC的监测研究中得到重视^[4] ...

2016

0.0

... 然而, 通过高光谱技术所获取的土壤光谱原始数据存在明显的光谱噪声和严重的散射现象, 土壤高光谱中必然存在与SMC不相关的噪声^[5], 这将会增加SMC的探测难度, 再加上高光谱本身的海量数据, 因此, 较为彻底的去噪和变量优选成为建立精度较高模型的关键 ...

2017

0.0

... 光谱数据的详细测定方法见文献[6]中1 ...

... 2 小波变换由于原始光谱数据存在大量的噪声信息, 需要对其进行去噪, 而小波变换作为一种新的数据处理方法, 其具备傅里叶分析方法的优势, 同时又填补了该方法的缺点^[7], 小波变换的表达式见文献[6]中1 ...

2015

0.0

GUO

Yan

, HOU

Su-zhen

, WANG

Ping

, et al(郭彦, 侯素珍, 王平, 等). Arid Zone Research(干旱区研究), 2015, 32(6): 1047.

2002

0.0

... 于雷等^[8]的研究结果表明, 在进行小波变换时, 当小波母函数为db4时的分解效果最佳 ...

2013

0.0

ZHANG

Chu

, LIU

Fei

, KONG

Wen-wen

, et al(张初, 刘飞, 孔汶汶, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2013, 29(20): 270.

Abstract: Watermelon seed variety selection plays a vital role in watermelon planting, and the variety of watermelon seeds directly affect the yield and quality of watermelons. In this study, we aimed to identify the cultivars of watermelon seeds by using a novel, rapid, non-invasive, and low cost technique named hyperspectral imaging. 121 samples of four different cultivars of watermelon seeds were investigated, and a near-infrared hyperspectral imaging system (874-1734 nm with 256 bands) was established to acquire the hyperspectral images of the samples. A region of interest (ROI) with 15×15 pixels of the hyperspectral image of each sample was defined, and the average reflectance spectrum of the ROI were extracted. To remove the absolute noises of the spectra, only the spectral range 1 042-1 646 nm was used for analysis, and to reduce the noises existed in spectral range 1 042-1 646 nm, the extracted 121 reflectance spectra were preprocessed by Savitzky-Golay smoothing (SG), Empirical Mode Decomposition (EMD), and Wavelet Transform (WT) methods. The preprocessed spectra were then used to select sensitive wavelengths by Successive Projections Algorithm (SPA) and Genetic Algorithm-partial least squares (GA-PLS) methods. Different numbers of sensitive wavelengths were selected by different variable selection methods with different preprocessing methods. 24, 16, and 15 sensitive wavelengths were selected by SPA with spectra preprocessed by SG, EMD, and WT, respectively. Moreover, 38, 33. and 32 sensitive wavelengths were selected by GA-PLS with spectra preprocessed by SG, EMD. and WT, respectively. Partial least squares - discriminant analysis (PLS-DA) was used to build discriminant models with the full spectra, and back-propagation neural network (BPNN) and extreme learning machine (ELM) were applied to build discriminant models with the selected wavelength variables. A PLS-DA model with spectra preprocessed by EMD obtained the best identification rate among all PLS-DA models, with an identification rate of 91.57% in the calibration set and 78.95% in the prediction set. SPA-BPNN models showed relatively worse results than GA-PLS-BPNN models with the same spectral preprocessing methods. The SG-GA-PLS-BPNN model obtained the best performance among all BPNN models, with an identification rate of 92.77% in the calibration set and 86.84% in the prediction set. Compared with the PLS-DA models and the BPNN models, ELM models obtained the best results. All ELM models obtained an identification rate over 90% in the calibration set and the prediction set, and the SG-SPA-ELM model, SG-GA-PLS-ELM model, and WT-SPA-ELM model obtained the identification rate of 100% of calibration and prediction. The overall results showed that BPNN and ELM models performed better than PLS-DA models, and the ELM models with the selected wavelengths based on SG preprocessed spectra obtained the best results, with 100% classification accuracy for both the calibration set and the prediction set. The SG preprocessing method showed the best performance in all PLS-DA, BPNN, and ELM models. The results indicated that it was feasible to use near-infrared hyperspectral imaging to identify the watermelon seed varieties, and near-infrared hyperspectral imaging provided an alternate way of rapid identification of watermelon seed variety. ELM, as a single hidden layer feed-forward network, was an effective classification method in watermelon seed cultivar identification. Moreover, the results in this paper showed the great potential of hyperspectral imaging in the seed industry for on-line identification of seed cultivars and detection of the seed quality parameters.

为了研究采用近红外高光谱图像技术对西瓜种子品种快速无损鉴别的可行性，该文采用近红外高光谱图像技术，通过提取西瓜种子的光谱反射率，结合Savitzky-Golay (SG)平滑算法，经验模态分解算法（empirical mode decomposition，EMD）和小波分析（wavelet transform，WT）对提取出的光谱数据进行去除噪声处理，采用连续投影算法（successive projections algorithm，SPA）和遗传-偏最小二乘法（genetic algorithm-partial least squares，GA-PLS）进行特征波长选择。基于全波段光谱建立了偏最小二乘判别分析（partial least squares-discriminant analysis，PLS-DA），基于特征波长建立了反向传播神经网络（back-propagation neural network，BP NN) 判别模型和极限学习机（extreme learning machine，ELM）判别模型。试验结果表明，基于特征波长的BPNN模型和ELM模型的结果优于基于全部波长的PLS-DA模型，基于SG预处理光谱提取的特征波长建立的ELM模型取得最优的判别效果，建模集和预测集的判别正确率均为100%。结果表明应用近红外高光谱成像技术对西瓜种子品种鉴别是可行的，为西瓜种子的品种快速鉴别提供了一种新方法。

... 3 特征波长选择算法由于高光谱数据量大, 所以小波系数重构后仍存在大部分冗余和共线信息, 导致计算量大和模型复杂, 对光谱的定量分析造成严重影响^[9] ...

2013

0.0

... BP是多层前馈神经网络^[10], 作为常用的预测模型中的一种, 其结构简练、准确度较高 ...

2013

0.0

... SVM最早是一种用有导师学习方式来解决二分类问题的模型, 它能找出多维空间数据的分离最优超平面, 并尽可能让所有训练集到该超平面的误差最小^[11] ...

2015

0.0

... RF模型是建立在决策树基础上, 通过多次bootstrap 抽样获得随机样本, 并通过这些样本分别建立相对应的决策树, 对于回归问题, 其最终的结果为所有决策树得到的结果的均值^[12], RF模型建立的过程中涉及到2个关键的参数: 决策树的数量(ntree)和节点分裂的次数(mtry), 本文预测设定ntree为1 000, mtry为2 ...

2013

0.0

... ELM算法是近年提出的单隐层前馈神经网络^[13], 它在算法执行过程中随机设定输入层与隐含层之间的权值和阈值, 不用重复更新迭代网络的输入权值以及隐元的偏置, 并可以得到最优解, 因此其运算时间短、移植性强^[14] ...

2015

0.0

2014

0.0

... 4)^[15] ...