摘要
叶片氮素含量是评价植被生长状况的重要指标,快速、准确监测核桃树冠层氮素含量的变化,对及时掌控树体长势、实施精准管理具有重要意义。本研究通过低空无人机遥感平台搭载GS 2型成像光谱仪,获取了果实膨大期5年生核桃林地的高光谱遥感影像数据。利用ENVI 5.3软件对观测范围内的核桃、土壤以及阴影区域进行识别提取,根据不同地物的波谱差异寻找核桃与土壤、阴影区域之间无交集且差异较大的波段区间,确定冠层的范围,并通过支持向量机方法验证其提取精度;根据NDVI、RVI和DVI植被指数筛选指示冠层氮素含量的特征敏感波段,分析了9种光谱参数对核桃冠层氮素含量的估算能力及其相关性,并将筛选的特征敏感波段作为BP神经网络模型的输入变量,进行了核桃冠层氮素含量的估算。结果表明:当B_(100)(550.7)处的光谱反射率大于0.10,且B_(233)(779.4)处的光谱反射率大于0.70时,可有效识别和确定核桃树冠层范围,制图精度高达96.43%。在分析核桃树冠层氮素含量与NDVI、RVI、DVI植被指数相关关系的基础上,确定了B_(33)(440.6)、B_(165)(660.7)、B_(186)(697.0)和B 347(986.4)为指示氮素含量的特征敏感波段。9种光谱参数中,以B_(347)(986.4)和B_(186)(697.0)重构的NDVI(986.4,697.0)在核桃林地冠层氮素含量的诊断中更接近实测值,估算模型精度最高。基于BP神经网络建立的估算模型较9种光谱参数具有更高的估算精度,测试集R^(2)达0.805,具有一定的估算可靠性。
Leaf nitrogen content is an important index to evaluate the growth of vegetation.It is of great significance to understand the change of nitrogen content in walnut canopy quickly,efficiently and accurately,so as to control the growth of trees in time and implement precise management.Taking the expanding period of walnut fruit as an example,the hyperspectral remote sensing image data of 5-year-old walnut forest land was obtained by GS 2 imaging spectrometer on the low altitude UAV remote sensing platform.The ENVI 5.3 software was used to identify and extract the walnut,soil and shadow in the observation range,and according to the spectral differences of different objects to find the non-intersection and large difference band between walnut,soil and shadow to determine the canopy range,and verify its extraction accuracy through support vector machine method.According to the NDVI,RVI and DVI vegetation indexes,the characteristic sensitive bands indicating the nitrogen content of the canopy were screened,and the correlation and estimation ability of 9 spectral parameters with the nitrogen content of the walnut canopy.Using the screened feature-sensitive bands as input variables of BP neural network model,the nitrogen content of walnut canopy was estimated.The screened feature-sensitive bands was used as the input variable of BP neural network model to estimate the nitrogen content of walnut canopy.The results showed that when the spectral reflectance at B_(100)(550.7)was more than 0.10 and that at B 233(779.4)was more than 0.70,the canopy range of walnut could be identified and determined effectively.Its drawing accuracy was as high as 96.43%.Based on the correlation between nitrogen content in walnut canopy and NDVI,RVI and DVI vegetation indexes,B_(33)(440.6),B_(165)(660.7),B_(186)(697.0)and B_(347)(986.4)were determined as the characteristics of indicating nitrogen content sensitive band.The estimation models based on the three reconstructed vegetation indices NDVI(986.4,697.0),RVI(986.4,697.0),and DVI(660.7,440.6)all reached extremely significant levels.Among them,NDVI(986.4,697.0)constructed by two bands of B_(347)(986.4)and B 186(697.0)was more close to the measured value in the diagnosis of nitrogen content in walnut forest canopy,and the accuracy of the estimation model was the highest.The estimation model based on BP neural network had higher estimation accuracy than the nine spectral parameters,the R^(2)of verification reached 0.805.The estimation model had the highest accuracy and certain estimation reliability.
作者
王鑫梅
张劲松
孟平
杨洪国
孙圣
WANG Xinmei;ZHANG Jinsong;MENG Ping;YANG Hongguo;SUN Sheng(Research Institute of Forestry,Chinese Academy of Forestry,Beijing 100091,China;Collaborative Innovation Center of Sustainable Forestry in Southern China,Nanjing Forestry University,Nanjing 210037,China)
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2021年第2期178-187,共10页
Transactions of the Chinese Society for Agricultural Machinery
基金
中央级公益性科研院所基本科研专项基金项目(CAFZC2017M005、CAFBB2017ZX002)。
关键词
核桃冠层
氮素含量
无人机遥感
特征敏感波段
植被指数
BP神经网络
walnut canopy
nitrogen content
UAV remote sensing
characteristic sensitive band
vegetation index
BP neural network