摘要
利用无人机遥感技术对小麦的叶面积指数进行反演，可以快速估算小麦的叶面积指数。本文从叶面积指数的反演模型进行研究，首先利用六旋翼无人机搭载ADC Lite相机获取实验区遥感图像，并对图像进行预处理；接着，提取图像的植被指数和LAI-2200冠层分析仪的叶面积指数，并选取建模样本和验证样本；最后利用植被指数与叶面积指数分别建立经验模型、BP神经网络模型和支持向量机模型，并对构建的模型进行精度评价，选取最佳的叶面积指数反演模型。本文的主要研究内容和结论如下：（1）图像采集与预处理。利用六旋翼无人机搭载多光谱相机进行图像采集，通过Pix4D mapper软件对采集得到的多光谱图像进行拼接处理。 （2）经验模型分析。利用经验法，构建植被指数与叶面积的经验模型，并利用绝定系数、均方根误差进行精度评价，寻出较好的叶面积指数反演模型。结果表明，NDVI-LAI线性模型的精度较好，决定系数达0.657。其次是SA VI-LAI线性模型、SA VI-LAI对数模型和RVI-LAI线性模型。
（3）BP神经网络模型分析。利用BP神经网络法，构建了NDVI-BP、RVI-BP、DVI-BP、SA VI-BP、TVI-BP、VI-BP六个BP神经网络模型，并进行模型精度评价。结果表明，将NDVI、RVI、DVI、SA VI和TVI植被指数组合作为输入量的BP神经网络模型效果较好，决定系数达0.78，均方根误差达0.567。
贺麓成
（4）支持向量机模型分析。利用支持向量机法，将NDVI、RVI、DVI、SA VI和TVI植被指数组合作为输入量，叶面积指数为输出量，进行支持向量机模型的构建，并模型进行精度评价。结果表明，支持向量机模型效果较好，决定系数达0.828，均方根误差达0.411。
（5）模型对比与优选。为了得到最佳的叶面积指数反演模型，将NDVI-LAI线性模型、SA VI-LAI线性模型、SA VI-LAI对数模型、RVI-LAI线性模型、VI-BP神经网络模型、VI-SVM模型进行精度评价。结果表明，BP神经网络的预测精度达85.62%，较NDVI-LAI线性模型提高了2.5%，而支持向量机预测精度达89.74%，较BP神经网络模型精度提高了4.12%。因此植被指数与LAI的构建SVM模型是LAI反演的最佳模型。利用支持向量机模型对4个观测阶段的小麦遥感图像进行了LAI反演，得到LAI空间分布图。
关键词：无人机遥感；多光谱图像；经验模型；BP神经网络；支持向量机
RESEARCH ON THE INVERSION METHODS OF WHEAT LEAF AREA INDEX BASED ON UNMANNED AERIAL VEHICLE REMOTE SENSING
ABSTRACT
Retrieving wheat leaf area index by using unmanned aerial vehicle remote sensing was estimated cr
op leaf area index quickly. This paper describes the inversion models of leaf area index . It included the following steps. First, obtained the pictures of the study area by using six wing unmanned aerial vehicle equipment with ADC Lite,meanwhile mosaicking and preprocessing the multispectral images. Second, extracting the vegetation indexs from the multispectral images and extracting leaf area index from LAI-2200,meanwhile choosing samples to model and validate.Finally, built the models of the retrieving leaf area index by using vegetation indexs and leaf area index.the models were empirical models, back propagation neural network models and support vector machine model.And models were evaluated the accuracy of prediction results,meanwhile determining the best model of retrieving leaf area index. The main content and conclusions of the paper were as follows.
（1）The multispectral images by using the unmanned aerial vehicle remote sensing system, were mosaicked in the Pix4D mapper software.
（2）The empirical models of retrieving leaf area index were built by using vegetation indexs and leaf area index. And models were evaluated the accuracy of prediction results by using the determination coefficient and the root of mean square error. The results show that the linear model of normalized difference vegetation index and leaf area index was best,and the determination coefficient was 0.657. The linear and logarithm model of soil-adjusted vegetation index and leaf area
inex were next.The linear model of ratio vegetation Index and leaf area index was followed.
（3）Built the six back propagation neural network models of vegetation indexs and leaf area index. Five vegetation indexs and the combination of five vegetation indexs inputted in the back propagation neural network. And models were evaluated the accuracy of prediction results by using the determination coefficient and the root of mean square error. The results show that the combination of five vegetation indexs as the input was the best back propagation neural network model,and the determination coefficient was 0.78, the root of mean square error was 0.567.
（4）The support vector machine model of vegetation indexs and leaf area index was
built,which the combination of five vegetation indexs as the input and leaf area index as output.Prediction results were better by using the support vector machine model,and the determination coefficient was 0.828, the root of mean square error was 0.411.
（5）The accuracy of prediction were evaluated by using the better models.The better models were the linear model of normalized difference vegetation index and leaf area index, the linear model and logarithm model of soil-adjusted vegetation index and leaf area inex, The linear model of ratio vegetation Index and leaf area index, the back propagation neural network model of vegetation index
s and leaf area index ,the support vector machine model of vegetation indexs and leaf area index.The results show that back propagation neural network model precision is 85.62%, and support vector machine model precision was 89.74%. The back propagation neural network model precision improved 2.5% than the best empirical model of vegetation index and leaf area index,but the support vector machine model precision improved 4.12% than the back propagation neural network model. Hence, the support vector machine model of vegetation indexs and leaf area index was the best inversion model. The four multispectral images were inverted by using the support vector machine model.
KEYWORDS: unmanned aerial vehicle remote sensing;multispectral images;empirical model; back propagation neural network; support vector machine
目录
第一章绪论 (1)
1.1研究的目的和意义 (1)
1.2国内外研究概括 (1)
1.2.1 叶面积指数获取方法 (1)
花儿与少年百科1.2.2 遥感数据来源 (4)
1.2.3 存在问题 (5)
1.3研究内容与方法 (5)
1.3.1 研究内容 (5)
1.3.2 研究方法 (6)
1.3.3 技术路线 (6)
第二章数据采集与数据预处理 (8)
2.1研究区概括 (8)
2.2实验方案 (8)
2.3无人机航测系统 (9)
2.3.1 无人机平台与传感器 (9)刘震云百度百科
2.3.2 无人机航拍路线设计 (10)
2.3.3 地面传感器 (10)
2.3.4 数据采集 (11)
2.4数据预处理 (12)
2.4.1 图像拼接过程 (12)
2.4.2 多光谱畸变精度验证 (13)
2.4.3试验区遥感图像校正和裁剪 (13)
2.4.4 植被指数的选取与提取 (14)
2014浙江高考英语2.5本章小结 (15)
第三章基于经验模型的LAI反演模型研究 (16)
3.1经验模型 (16)
3.1.1 经验模型 (16)
3.1.2 模型结果检验 (16)
3.2基于经验法模型的结果与分析 (17)
3.2.1 不同生育阶段LAI分析 (17)
3.2.3 植被指数与LAI相关性分析 (18)
3.2.4 整个观测阶段反演模型 (19)
3.2.5 各个观测阶段反演模型 (22)
3.2.6 基于经验模型的预测结果 (28)
3.3本章小结 (29)
第四章基于BP神经网络的LAI反演模型研究 (30)
4.1BP神经网络模型 (30)
4.1.1 BP神经网络 (30)
4.1.2 LAI反演模型的BP神经网络设计 (31)
4.2基于BP神经网络模型的结果与分析 (33)
4.3基于BP神经网络模型的预测结果 (35)
4.4本章小结 (36)
太师庄中学第五章基于支持向量机的LAI反演模型研究 (37)
5.1支持向量机模型 (37)
5.1.1 支持向量机 (37)
5.1.2 LAI反演的支持向量机设计 (39)
5.2基于支持向量机模型的结果与分析 (40)
5.3基于支持向量机模型的预测结果 (42)
5.4各模型精度对比分析 (43)
5.5叶面积指数反演结果 (44)
5.6本章小结 (46)
第六章结论与展望 (47)
6.1结论 (47)
6.2创新点 (47)
6.3展望 (48)
参考文献 (49)
致谢 (53)
郑州轻工业学院图书馆作者简介 (54)
豫公网安备41160202000603
站长QQ:729038198
关于我们
投诉建议