高光谱成像的马铃薯叶片含水率分布可视化
|
摘要
为了快速检测马铃薯叶片的水分含量,并探究受到干旱胁迫时叶片含水率变化情况,利用高光谱成像对马铃薯叶片含水率进行检测和可视化研究。采集71个叶片,用烘干法对叶片水分梯度进行控制,共得到355个样本。使用高光谱分选仪器采集叶片862.9～1 704.2 nm(256个波长)的光谱成像数据,采用称重法测量含水率。利用Sample set partitioning based on joint X-Y distance(SPXY)算法将总样本按照2∶1的比例划分为建模集(240个样本)和验证集(115个样本)。对采集的数据进行光谱特征分析,本文分别用CA和RF两种算法,各筛选得到15个特征波长。基于CA筛选出相关系数高于0.96的15个波长分别为1 406.82, 1 410.12, 1 403.62, 1 413.32, 1 416.62, 1 419.82, 1 400.32, 1 423.12, 1 426.32, 1 429.62, 1 432.82, 1 436.12, 1 439.32, 1 442.52和1 445.8 nm。基于RF算法筛选被选概率高于0.3的15个特征波长,按照被选择概率值从大到小排列,分别为1 071.62, 1 041.12, 1 222.52, 1 465.22, 1 397.02, 1 449.02, 1 034.32, 1 523.22, 976.42, 1 172.52, 979.82, 1 165.82, 1 037.72, 1 426.32和869.8 nm。用CA和RF算法筛选到的特征波长建立PLSR模型,分别记为CA-PLSR模型和RF-PLSR模型。利用高精度模型检测结果,对马铃薯叶片含水率进行可视化分析,首先计算马铃薯叶片图像每个像素点的含水率,得到灰度图像,然后对灰度图像进行伪彩色变换,绘制出叶片含水率可视化彩色图像。为了体现马铃薯叶片烘干处理中含水率变化进程,用HSV彩色模型对样本叶片的伪彩色图像进行分割,获得分割图像结果,显示出在某含水率区间的叶片面积比例。结果显示, CA算法选取的15个波长均在1 400.3～1 450.0 nm范围内, CA-PLSR模型的建模精度(R■)为0.975 5、建模集均方根误差(RMSEC)为2.81%,验证集精度(R■)为0.933 2、验证集均方根误差(RMSEV)为2.31%。RF算法选取的特征波长分布范围较CA法选取范围广,具有局部"峰谷"特性,且RF-PLSR模型的建模集精度(R■)为0.983 2、 RMSEC为2.32%,验证集精度(R■)为0.947 1、 RMSEV为2.15%。选取RF-PLS模型计算马铃薯每个像素点的含水率,得到伪彩色变换图像,观察可知随着烘干时间的增加含水率逐渐下降;并能够从叶片结构角度看到,随着水分胁迫的加强,叶片从边缘开始失水,逐渐向叶片中间蔓延,其中叶茎和叶脉的含水率较其他部位高。计算得到叶片伪彩色图像中含水率大于90%, 80%和70%的像素点占整个叶片图像的比例。利用高光谱成像技术可以实现马铃薯叶片的含水率检测与分布可视化表达,为监测马铃薯生长状况以及叶片含水率分析提供新的理论根据。
In order to quickly detect the water content of potato leaves and explore the change of leaf water content under drought stress, the hyperspectral imaging technology was utilized to detect and visualize the moisture content of potato leaves in this paper. 71 leaves were collected and the water gradient of the leaves was controlled by a drying method. A total of 355 samples were obtained. The hyperspectral sorting instrument was used to collect potato leaves spectral and image data of 862.9～1 704.2 nm(256 wavelengths). The water content was measured by weighing method. According to a certain proportion, Sample Set Partitioning Based on Joint X-Y Distance(SPXY) algorithm was used to divide the sample into a model set and a validation set. For the calibration set, the feature wavelengths were extracted by using Coefficient Analysis(CA) and Random Frog(RF) algorithms respectively, and Partial Least Squares Regression(PLSR) models were established respectively. The calibration set and validation set determines coefficient R~2 and the RMSE(Root Mean Square Error) were used as the evaluation index. The gray image of the potato leaves water content was calculated using the results of the detection model. The visualization analysis of potato leaves water content was realized based on the pseudo color image transformation and segmentation. The average reflectance of each sample leaf was calculated by ENVI software, obtaining a total of 355 sample's spectral data. According to the proportion of 2∶1, the total samples were divided into calibration set(240 samples) and validation set(115 samples) by SPXY algorithm. Spectral characteristics of the collected data were analyzed. Two algorithms, CA and RF, were used to select 15 characteristic wavelengths, respectively. Based on the CA, the selected 15 wavelengths with the correlation coefficient higher than 0.96 were 1 406.82, 1 410.12, 1 403.62, 1 413.32, 1 416.62, 1 419.82, 1 400.32, 1 423.12, 1 426.32, 1 429.62, 1 432.82, and 1 441.12, 1 493.32, 1 442.52 and 1 445.8 nm. Based on the RF algorithm, the 15 feature wavelengths that the selected probability higher than 0.3 were 1 071.62, 1 041.12, 1 222.52, 1 465.22, 1 397.02, 1 449.02, 1 034.32, 1 523.22, 976.42, 1 172.52, 979.82, 1 165.82, 1 037.72, 1 426.32 and 869.8 nm. The PLSR model was established using the characteristic wavelengths filtered by the CA and RF algorithms, marked RF-PLSR and RF-PLSR model respectively. The water content of potato leaves was analyzed visually using the more precise model. First, the each pixel water content of the potato leaf image was calculated to obtain a gray image. Then, the gray image was pseudo-color transformed to draw a visual color image of the leaf water content. In order to reflect the change of potato leaves water content in the drying process, HSV model was used to segment the pseudo-color image of sample leaf. The segmentation image results, showing the proportion of leaf area in a certain water content interval, were obtained. The results showed that the 15 wavelengths selected by the CA algorithm were in the range of 1 400.3 to 1 450.0. The calibration accuracy of CA-PLSR was 0.975 5, and the RMSEC(Root Mean Square Error of Calibration) was 2.81%, and the validation accuracy was 0.933 2, and the RMSEV(Root Mean Square Error of Validation) was 2.31%. The range of characteristic wavelengths selected by the RF algorithm, with local "peak valley" characteristic, was wider than that of the CA algorithm. The calibration accuracy of RF-PLSR model was 0.983 2, and the RMSEC was 2.32%, and the validation accuracy was 0.947 1, and the RMSEV was 2.15%. The RF-PLS model is selected to calculate the water content of each pixel in the potato leaf images. According to the pseudo-color image, it could be seen intuitively that the water content gradually decreased with the drying time increasing. From the perspective of leaf tissue structure, with the strengthening of water stress, the leaves began to lose water from the edge and gradually spread to the middle, in which the water content of leaf stems and veins was higher than that of other parts. The pseudo-color image was segmented by HSV model based on the color difference of the image. The proportion of pixels with water content greater than 90%, 80%, and 70% in the leaf pseudo-color image to the entire leaf image. Using the hyperspectral imaging technology can realize the water content detection and distribution visualization of potato leaves, which provides a new theoretical basis for the potato growth analysis and potato leaf water content analysis.
In order to quickly detect the water content of potato leaves and explore the change of leaf water content under drought stress, the hyperspectral imaging technology was utilized to detect and visualize the moisture content of potato leaves in this paper. 71 leaves were collected and the water gradient of the leaves was controlled by a drying method. A total of 355 samples were obtained. The hyperspectral sorting instrument was used to collect potato leaves spectral and image data of 862.9～1 704.2 nm(256 wavelengths). The water content was measured by weighing method. According to a certain proportion, Sample Set Partitioning Based on Joint X-Y Distance(SPXY) algorithm was used to divide the sample into a model set and a validation set. For the calibration set, the feature wavelengths were extracted by using Coefficient Analysis(CA) and Random Frog(RF) algorithms respectively, and Partial Least Squares Regression(PLSR) models were established respectively. The calibration set and validation set determines coefficient R~2 and the RMSE(Root Mean Square Error) were used as the evaluation index. The gray image of the potato leaves water content was calculated using the results of the detection model. The visualization analysis of potato leaves water content was realized based on the pseudo color image transformation and segmentation. The average reflectance of each sample leaf was calculated by ENVI software, obtaining a total of 355 sample's spectral data. According to the proportion of 2∶1, the total samples were divided into calibration set(240 samples) and validation set(115 samples) by SPXY algorithm. Spectral characteristics of the collected data were analyzed. Two algorithms, CA and RF, were used to select 15 characteristic wavelengths, respectively. Based on the CA, the selected 15 wavelengths with the correlation coefficient higher than 0.96 were 1 406.82, 1 410.12, 1 403.62, 1 413.32, 1 416.62, 1 419.82, 1 400.32, 1 423.12, 1 426.32, 1 429.62, 1 432.82, and 1 441.12, 1 493.32, 1 442.52 and 1 445.8 nm. Based on the RF algorithm, the 15 feature wavelengths that the selected probability higher than 0.3 were 1 071.62, 1 041.12, 1 222.52, 1 465.22, 1 397.02, 1 449.02, 1 034.32, 1 523.22, 976.42, 1 172.52, 979.82, 1 165.82, 1 037.72, 1 426.32 and 869.8 nm. The PLSR model was established using the characteristic wavelengths filtered by the CA and RF algorithms, marked RF-PLSR and RF-PLSR model respectively. The water content of potato leaves was analyzed visually using the more precise model. First, the each pixel water content of the potato leaf image was calculated to obtain a gray image. Then, the gray image was pseudo-color transformed to draw a visual color image of the leaf water content. In order to reflect the change of potato leaves water content in the drying process, HSV model was used to segment the pseudo-color image of sample leaf. The segmentation image results, showing the proportion of leaf area in a certain water content interval, were obtained. The results showed that the 15 wavelengths selected by the CA algorithm were in the range of 1 400.3 to 1 450.0. The calibration accuracy of CA-PLSR was 0.975 5, and the RMSEC(Root Mean Square Error of Calibration) was 2.81%, and the validation accuracy was 0.933 2, and the RMSEV(Root Mean Square Error of Validation) was 2.31%. The range of characteristic wavelengths selected by the RF algorithm, with local "peak valley" characteristic, was wider than that of the CA algorithm. The calibration accuracy of RF-PLSR model was 0.983 2, and the RMSEC was 2.32%, and the validation accuracy was 0.947 1, and the RMSEV was 2.15%. The RF-PLS model is selected to calculate the water content of each pixel in the potato leaf images. According to the pseudo-color image, it could be seen intuitively that the water content gradually decreased with the drying time increasing. From the perspective of leaf tissue structure, with the strengthening of water stress, the leaves began to lose water from the edge and gradually spread to the middle, in which the water content of leaf stems and veins was higher than that of other parts. The pseudo-color image was segmented by HSV model based on the color difference of the image. The proportion of pixels with water content greater than 90%, 80%, and 70% in the leaf pseudo-color image to the entire leaf image. Using the hyperspectral imaging technology can realize the water content detection and distribution visualization of potato leaves, which provides a new theoretical basis for the potato growth analysis and potato leaf water content analysis.
引文
[1]HOU Xian-qing,LI Rong,HE Wen-shou,et al(侯贤清,李荣,何文寿,等).Journal of Zhejiang University(浙江大学学报),2015,41(5):558.
[2]XUE Jun-wu,REN Wen-jiang,YAN Chang-rong(薛俊武,任稳江,严昌荣).Chinese Journal of Agrometeorology(中国农业气象),2014,35(1):74.
[3]QIN Jun-hong,CHEN You-jun,ZHOU Chang-yan,et al(秦军红,陈有君,周长艳,等).Chinese Journal of Eco-Agriculture(中国生态农业学报),2013,21(7):824.
[4]Cheng T,Rivard B,Sánchez-Azofeifa A.Remote Sensing of Environment,2011,115(2):659.
[5]LI Lei-lei,WANG Hai-xia,LIN Ping,et al(李蕾蕾,王海霞,林萍,等).Chinese Journal of Experimental Traditional Medical Formulae(中国实验方剂学杂志),2013,19(19):104.
[6]Zhao C,Li H,Li P,et al.IEEE Transactions on Geoscience&Remote Sensing,2016,55(1):236.
[7]Dejonge K C,Mefford B S,Chávez J L.International Journal of Remote Sensing,2016,37(10):2294.
[8]Gerhards M,Rock G,Schlerf M,et al.International Journal of Applied Earth Observation&Geoinformation,2016,53:27.
[9]CHEN Xiang,LI Min-zan,SUN Hong,et al(陈香,李民赞,孙红,等).Transactions of the Chinese Society of Agricultural Engineering(农业工程学报),2017,33(S1):137.
[10]Das B,Sahoo R N,Pargal S,et al.Spectrochimica Acta Part A:Molecular and Biomolecular Spectroscopy,2018,192:41.
[11]SUN Hong,ZHENG Tao,LIU Ning,et al(孙红,郑涛,刘宁,等).Transactions of the Chinese Society of Agricultural Engineering(农业工程学报),2018,34(1):149.
[12]Yang H,Zhang D Y,Dong Y Y,et al.International Journal of Agriculture&Biology,2013,15(4):694.
[13]Mirik M,Anskey R J,Steddom K,et al.Journal of Applied Remote Sensing,2014,8(1):083661.
[14]Li H D,Xu Q S,Liang Y Z.Analytica Chimica Acta,2012,740(31):20.
[2]XUE Jun-wu,REN Wen-jiang,YAN Chang-rong(薛俊武,任稳江,严昌荣).Chinese Journal of Agrometeorology(中国农业气象),2014,35(1):74.
[3]QIN Jun-hong,CHEN You-jun,ZHOU Chang-yan,et al(秦军红,陈有君,周长艳,等).Chinese Journal of Eco-Agriculture(中国生态农业学报),2013,21(7):824.
[4]Cheng T,Rivard B,Sánchez-Azofeifa A.Remote Sensing of Environment,2011,115(2):659.
[5]LI Lei-lei,WANG Hai-xia,LIN Ping,et al(李蕾蕾,王海霞,林萍,等).Chinese Journal of Experimental Traditional Medical Formulae(中国实验方剂学杂志),2013,19(19):104.
[6]Zhao C,Li H,Li P,et al.IEEE Transactions on Geoscience&Remote Sensing,2016,55(1):236.
[7]Dejonge K C,Mefford B S,Chávez J L.International Journal of Remote Sensing,2016,37(10):2294.
[8]Gerhards M,Rock G,Schlerf M,et al.International Journal of Applied Earth Observation&Geoinformation,2016,53:27.
[9]CHEN Xiang,LI Min-zan,SUN Hong,et al(陈香,李民赞,孙红,等).Transactions of the Chinese Society of Agricultural Engineering(农业工程学报),2017,33(S1):137.
[10]Das B,Sahoo R N,Pargal S,et al.Spectrochimica Acta Part A:Molecular and Biomolecular Spectroscopy,2018,192:41.
[11]SUN Hong,ZHENG Tao,LIU Ning,et al(孙红,郑涛,刘宁,等).Transactions of the Chinese Society of Agricultural Engineering(农业工程学报),2018,34(1):149.
[12]Yang H,Zhang D Y,Dong Y Y,et al.International Journal of Agriculture&Biology,2013,15(4):694.
[13]Mirik M,Anskey R J,Steddom K,et al.Journal of Applied Remote Sensing,2014,8(1):083661.
[14]Li H D,Xu Q S,Liang Y Z.Analytica Chimica Acta,2012,740(31):20.
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |