摘要
为了探讨基于油菜高光谱间接估测油菜角果皮含水率的潜力,于2023年3月—5月采集试验田油菜光谱和实测油菜角果皮含水率,经2种光谱预处理,和3种特征波长筛选方法及其组合,并融合高光谱图像空间纹理信息,用偏最小二乘回归(PLSR)、Lasso回归、支持向量回归(SVR)和极限学习机(ELM)建立角果皮含水率的回归模型,同时对模型结果进行精度评价。研究结果表明:(1)光谱预处理能够突出光谱中的一些隐藏信息,对油菜光谱进行多元散射校正(MSC)和一阶导数(first derivative,FD)数学变换后更加有利于提取光谱敏感信息;(2)预处理后,采用竞争性自适应重加权算法(CARS)和迭代的保持有信息变量(IRIV)相结合的特征波段筛选方法,利用Lasso模型进行预测效果最好,测试集的R2为0.7720;(3)针对油菜角果皮这一结构复杂、体积小、含水率分布易受几何结构影响的关键作物对象,在纯光谱信息的基础上,引入空间纹理信息,空间纹理量化了角果表面与含水率相关的空间变异和结构细节(如皱纹、凹凸),并补偿单一像元光谱因角果形状、朝向引起的变异,提升模型回归精度和预测能力,增强模型对噪声和异常值的鲁棒性,为解决复杂小尺度作物对象生理参数精准反演提供了新的有效途径。
To explore the potential of indirectly estimating the moisture content in silique peel based on hyperspectral data,this study took the rapeseed experimental field as the research object.From March to May 2023,the rapeseed spectra and moisture content of rapeseed silique peel were collected from the experimental field.After two spectral preprocessing methods,three feature wavelength selection methods,and their combinations,hyperspectral image spatial texture information was introduced.Partial Least Squares Regression(PLSR),Lasso regression,Support Vector Regression(SVR),and Extreme Learning Machine(ELM)were used to establish a regression model for the moisture content of rapeseed silique peel,and the accuracy of the model results was evaluated.The research results indicate that:(1)Spectral preprocessing techniques can highlight some hidden information in the spectrum,and mathematical transformations such as Multiple Scatter Correction(MSC)and First Derivative(FD)are more conducive to extracting spectral sensitive information;(2)After performing preprocessing,a feature selecting method combining Competitive Adaptive Reweighted Sampling(CARS)and Iterative Retention Informative Variables(IRIV)was used.The Lasso model demonstrated the best prediction performance,with an R2 of 0.7720 for test set 3.In response to the complex structure,small volume,and geometric influence of moisture content distribution in rapeseed silique peel,spatial texture information is introduced based on pure spectral information.Spatial texture quantifies the spatial variation and structural details(such as wrinkles and bumps)related to moisture content on the surface of silique,compensates for the variation caused by the shape and orientation of silique in a single pixel spectrum,improves the regression accuracy and prediction ability of the model,enhances the robustness of the model to noise and outliers,and provides a new effective way to solve the precise inversion of physiological parameters of complex small-scale crop objects.
作者
魏薇
王丹
王博韬
谭佐军
刘泉
谢静
WEI Wei;WANG Dan;WANG Bo-tao;TAN Zuo-jun;LIU Quan;XIE Jing(College of Engineering,Huazhong Agricultural University,Wuhan 430070,China;College of Informatics,Huazhong Agricultural University,Wuhan 430070,China;College of Resources&Environment,Huazhong Agricultural University,Wuhan 430070,China)
出处
《光谱学与光谱分析》
2025年第10期2863-2874,共12页
Spectroscopy and Spectral Analysis
基金
国家自然科学基金项目(42271357)
中央高校基本科研业务费专项(2662022LXYJ005)资助。
关键词
高光谱
油菜田
油菜角果皮
含水率
光谱数据处理
空间纹理信息
Hyperspectral
Rape field
Rapeseed silique peel
Moisture content
Spectral data processing
Spatial texture information
作者简介
魏薇,女,1981年生,华中农业大学工学院教授,e-mail:weiwei1981@mail.hzau.edu.cn;通讯作者:谢静,E-mail:xiejing625@mail.hzau.edu.cn。
