近红外高光谱成像扫描速度对拟南芥冠层含水率预测的影响  

Effect of Near Infrared Hyperspectral Imaging Scanning Speed on Prediction of Water Content in Arabidopsis

作　　者：吕梦琪 宋宇杰 翁海勇 孙大伟 董晓娅[2] 方慧[1,3] 岑海燕 LU Meng-qi;SONG Yu-jie;WENG Hai-yong;SUN Da-wei;DONG Xiao-ya;FANG Hui;CEN Hai-yan(College of Biosystems Engineering and Food Science,Zhejiang University,Hangzhou 310058,China;School of Agricultural Equipment Engineering,Jiangsu University,Zhenjiang 212013,China;Key Laboratory of Spectroscopy Sensing,Ministry of Agriculture and Rural Affairs,Hangzhou 310058,China;College of Computer Science and Technology,Zhejiang University,Hangzhou 310027,China)

机构地区：[1]浙江大学生物系统工程与食品科学学院,浙江杭州310058 [2]江苏大学农业装备工程学院,江苏镇江212013 [3]农业农村部光谱检测重点实验室,浙江杭州310058 [4]浙江大学计算机科学与技术学院,浙江杭州310027

出　　处：《光谱学与光谱分析》2020年第11期3508-3514,共7页Spectroscopy and Spectral Analysis

基　　金：浙江省重点研发计划项目(2017C02031);江苏省现代农业装备与技术协同创新中心项目(4091600007);国家自然科学基金项目(31971776)资助。

摘　　要：高光谱成像技术可以无损检测植物不同尺度的理化信息,现有研究往往以分析高光谱图像的平均光谱为主,忽略了其空间维度的信息。以模式植物拟南芥为研究对象,探究高光谱成像不同扫描速度引起的图像空间分辨率差异对植物冠层含水率测量的影响,为高光谱成像在线快速检测植物冠层含水率提供优化方案。首先利用室内在线高光谱成像系统分别在20, 30和40 mm·s^-1三种扫描速度下采集了拟南芥冠层高光谱图像,并提取拟南芥冠层平均反射光谱。其次,利用偏最小二乘算法(PLSR)建立了拟南芥冠层含水率与平均反射光谱的定量分析模型,通过决定系数(R2)、均方根误差(RMSE)、相对分析误差(RPD)对模型进行评估。比较基于原始光谱与多元散射校正算法(MSC)、 Savitsky-Golay平滑算法等预处理光谱建立的PLSR模型,选取最佳光谱预处理方法用于后续的数据处理。最后,利用连续投影算法(SPA)分析比较基于最优特征波长与全波长的模型预测准确度,探明高光谱图像扫描速度对拟南芥冠层含水率预测的影响规律。研究结果表明,当扫描速度从20 mm·s^-1提升到30 mm·s^-1时,基于MSC预处理的全波段PLSR模型预测拟南芥冠层含水率决定系数降低0.88%,小于1%;当扫描速度从20 mm·s^-1提升到40 mm·s^-1时,拟南芥冠层含水率决定系数降低2.3%。说明在适当提高扫描速度的同时,能够保证植物冠层的高含水率预测准确度。改变高光谱扫描速度可以更有效地利用高光谱图像空间维度有效信息,扫描速度适当增大后,高光谱图像的空间维度信息改变,提高实际生产应用环节的图像采集效率,减少数据处理时间。Hyperspectral imaging technology can non-destructively detect physicochemical information of plants with different dimensions.Existing researches often focus on analyzing the average spectrum of hyperspectral images,ignoring the information of their spatial dimensions.In this study,the model plant Arabidopsis thaliana was used as the research object to explore the influence of spatial resolution difference caused by different scanning speeds of hyperspectral imaging on the measurement of plant canopy moisture content,and to provide optimization for rapid online detection of plant canopy moisture content by hyperspectral imaging program.An open-line hyperspectral image of the Arabidopsis canopy was extracted using an indoor online hyperspectral imaging system at 20,30 and 40 mm·s^-1,and the average of the Arabidopsis thaliana canopy reflectance spectrum was extracted.Secondly,the quantitative analysis model of canopy water content and the average reflectance spectrum of Arabidopsis thaliana was established by Partial Least Squares Regression(PLSR).The determination coefficient(R 2),root mean square error(root),mean squared error(RMSE)and relative variance deviation(RPD)were used to evaluate the model.The PLSR model based on pre-processing spectra such as the original spectrum,Multiplicative Scatter Correction(MSC)algorithm and Savitsky-Golay smoothing algorithm is compared.The best spectral pre-processing method is selected for subsequent data processing.Finally,the successive projections algorithm(SPA)is used to analyze and compare the prediction accuracy based on the optimal feature wavelength and the full wavelength,and to determine the influence of the hyperspectral image scanning speed on the canopy water content prediction of Arabidopsis thaliana.The results show that when the scanning speed was increased from 20 to 30 mm·s^-1,the full-band PLSR model based on MSC pretreatment predicted that the coefficient of canopy moisture content in Arabidopsis was reduced by 0.88%,less than 1%.When the scanning speed

关 键 词：拟南芥 冠层含水率 近红外高光谱 扫描速度 SPA PLSR 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]