激光与光电子学进展, 2019, 56 (15): 153002, 网络出版: 2019-08-05   

基于分数阶微分和连续投影算法-反向传播神经网络的小麦叶片含水量高光谱估算 下载: 969次

Hyperspectral Estimation of Wheat Leaf Water Content Using Fractional Differentials and Successive Projection Algorithm-Back Propagation Neural Network
作者单位
1 新疆大学资源与环境科学学院, 新疆 乌鲁木齐 830046
2 新疆绿洲生态教育部重点实验室, 新疆 乌鲁木齐 830046
3 新疆智慧城市与环境建模普通高校重点实验室, 新疆 乌鲁木齐 830046
摘要
为了探索分数阶微分在高光谱估算小麦叶片含水量上的可行性,在农田尺度上,利用春小麦野外光谱数据与实测叶片含水量数据,以0.2阶为步长,计算光谱0~2阶微分,并分析其与小麦叶片含水量的相关性,再利用连续投影算法(SPA)从通过0.01水平显著性检验的波段中筛选出估算叶片含水量的最佳波段组合,并建立估算春小麦叶片含水量的反射传播(BP)神经网络模型。结果表明:分数阶微分可以细化小麦叶片含水量与光谱数据相关性的变化趋势;分数阶微分处理后,相关系数通过0.01水平显著性检验的波段数量呈现先增后减的趋势,在不同的波段范围内,分数阶微分的最佳阶数也有所不同;SPA筛选出的敏感波段基本上集中在红光、近红外波段范围内,1.2阶微分后水分敏感波段数最多,达到13个;所建立的模型中,基于1.8阶微分建立的6-4-1结构的BP神经网络模型为最佳模型,其建模组均方根误差为0.701,决定系数为0.751,验证组的均方根误差为0.227,决定系数为0.917,相对分析误差为3.253,说明了分数阶微分后的模型稳定性和预测能力较整数阶微分得到明显的提升,可为高光谱定量反演春小麦叶片含水量提供参考。
Abstract
To explore the feasibility of using fractional differentials in the hyperspectral estimation of wheat leaf water content, we select the Fukang experimental science base of Xinjiang University as the study area. Based on springtime wheat-field spectral data and wheat leaf water content data, we calculate the fractional differentials of the spectrum from the 0-order to the 2-order in 0.2-order steps; further, we analyze their correlations with the water content of the wheat leaves. We then use the successive projection algorithm (SPA) to select the optimal combination of bands for estimating the leaf water content from bands passing the 0.01 significance test. Finally, we establish a back propagation (BP) neural network model for estimating the water content of spring wheat leaves. The results show that fractional differentials can refine the trend of correlation between wheat leaf water content and the wheat leaves' spectral data. After fractional differential processing, the number of bands for which the correlation coefficients pass the 0.01 significance test first increases and subsequently decreases; in addition, the optimal order of fractional differentials is also different in different bands. Sensitive bands selected by the SPA are mainly concentrated in the red and near infrared bands, and the number of water sensitive bands is highest (reaching 13) after 1.2-order differential processing. Among the models considered herein, the BP neural network model with the 6-4-1 structure based on the 1.8-order differential is the best model, with the following specifications: the root-mean-square error of the modeling group is 0.701, the determination coefficient of the modeling group is 0.751, the root-mean-square error of the verification group is 0.227, the determination coefficient of the verification group is 0.917, and the relative analysis error of the verification group is 3.253. These conclusions show that the stability and predictive ability of the model using fractional differentials are better than those of integer differentials, and it provides a well-defined reference for the quantitative inversion of hyperspectral data to estimate the water content of spring wheat.

吾木提·艾山江, 买买提·沙吾提, 马春玥. 基于分数阶微分和连续投影算法-反向传播神经网络的小麦叶片含水量高光谱估算[J]. 激光与光电子学进展, 2019, 56(15): 153002. Hasan Umut, Sawut Mamat, Chunyue Ma. Hyperspectral Estimation of Wheat Leaf Water Content Using Fractional Differentials and Successive Projection Algorithm-Back Propagation Neural Network[J]. Laser & Optoelectronics Progress, 2019, 56(15): 153002.

本文已被 4 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

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!