摘要
为了快速感知并分析田间作物生长状况,采用先进的半导体镀膜工艺光谱成像传感器,研究了玉米植株冠层叶绿素含量分布式检测方法。试验采用IMEC 5×5成像传感器,拍摄47株苗期玉米植株冠层,获取673~951 nm范围内的25个波长的光谱图像。实验中,利用SPAD-520叶绿素仪非破坏性地测量叶绿素含量,每株玉米冠层叶片设置2~3个采样点,每点测量3次取平均,共计242个样本数据。对光谱图像数据,经4灰度级标准板提取并校准反射率。为了实现玉米植株与花盆、土壤背景的有效分离,在分析不同对象光谱反射率与图像像素特征的基础上,提出了一种基于谱图特征组合的植株分割方法,即基于植被指数的图像初步分割与区域标记计算的冠层精细分割的植株提取算法。首先,计算各像素点归一化植被指数(NDVI),并开展基于NDVI的植株冠层分割方法分割结果优于基于最大类间方差法的全局阈值自适应分割算法。其次,采用边缘保持中值滤波算法剔除初步分割后图像中存在的噪声点后,基于区域标记算法进行精细分割,获得掩膜并最终得到仅保留玉米植株冠层的光谱图像。分别采用相关分析法(CA)和随机蛙跳(RF)算法选取反射光谱特征波长,并构建750~951 nm近红外(NIR)和673~750 nm红色(R)选中波长集合,遍历NIR和R集合组合计算比值植被指数(RVI),差值植被指数(DVI),归一化植被指数(NDVI)和SPAD转换指数(TSPAD)。然后,再次采用CA和RF算法筛选植被指数,利用SPXY算法将样本按照7∶3比例划分为建模集和验证集,并建立了叶绿素含量指标检测CA+RF-PLSR模型。结果表明,其建模集R^2C为0.573 9, RMSEC为3.84%,验证集R^2V为0.420 2, RMSEC为2.3%。利用建模结果对多光谱图像进行处理,绘制玉米叶片SPAD值伪彩色分布图,实现叶绿素含量分布可视化。研究表明采用镀膜型光谱成像数据,分析对象光谱与图像特征,探讨玉米冠层
In order to quickly analyze the growth of the crop in the field,the spectral imaging sensor was used to detect the chlorophyll content of the maize canopy.The imagesof 47 maize plants were photographed using an IMEC 5×5 imaging unit multispectral camera.The camera was designed based on the coating principle to obtain spectral images of 25 wavelengths in the range of 673~951 nm.At the same time,the chlorophyll content was measured by SPAD-520 device.There were 2~3 sampling points in each leaf,and they were measured 3 times at each point so that 242 sample data were collected.A linear inversion formula was established based on the relationship between the gray value of multi-spectral images and the gray plate standard reflectance.The gray plate standard was made up of 4 gray level standard plates.In order to separate the plant from flowerpots and soil background,a combination method was studied.Although the canopy was segmented using OTSU method,it was not useful.After analyzing the spectral reflectance characteristics of different objects,a plant extraction algorithm was proposed based on normalization difference vegetation index(NDVI)image and region marker calculation.Firstly,the initial segmentation was conducted based on NDVI calculation on each pixel.Secondly,the noise points were eliminated by the edge-preserved median filtering algorithm.Thirdly,the region algorithm was used to obtain a mask and finally segment the multi-spectral images of theplant canopy.The characteristic wavelengths were selected based on CA(Correlation Analysis,CA)and RF(random Frog,RF)algorithm,which was used to construct the Near-Infrared(NIR)and Red(R)data set.The vegetationindices were calculated by the traversing NIR and R sets including the Ratio Vegetation Index(RVI),the Normalized Difference Vegetation Index(NDVI),the Difference Vegetation Index(DVI),and the SPAD Transfer Index(TSPAD).According to the proportion of 7∶3,the total samples were divided into calibration and validation setby SPXY(Sample set partitioning based on j
作者
龙耀威
李民赞
高德华
张智勇
孙红
Qin Zhang
LONG Yao-wei;LI Min-zan;GAO De-hua;ZHANG Zhi-yong;SUN Hong;Qin Zhang(Key Laboratory of Modern Precision Agriculture System Integration Research,Ministry of Education,China Agricultural University,Beijing 100083,China;Center for Precision&Automated Agricultural System,Washington State University,Pullman WA 99350,USA)
出处
《光谱学与光谱分析》
SCIE
EI
CAS
CSCD
北大核心
2020年第7期2253-2258,共6页
Spectroscopy and Spectral Analysis
基金
广西创新驱动发展专项资金项目(桂科AA18118037)
国家自然科学基金项目(31971785,31501219)
“海外名师”高端外国专家项目(MS2017ZGNY004)
中国农业大学研究生培养项目(ZYXW037,HJ2019029,JG2019004)
中央高校基本科研业务费专项资金项目(2020TC036)资助。
