摘要
为实现大面积果园土壤水分含量(soil moisture cotent,SMC)的快速估测,应用无人机高光谱成像仪,对甘肃省静宁县人工灌溉的某苹果示范园与自然降雨的某苹果园进行了拍摄,获取表面土壤的光谱图像,综合比较两区共240个土壤样本点的原始光谱反射率(R)、倒数变换光谱反射率(1/R)、对数变化光谱反射率(LOG(R))、根号变换光谱反射率(√R)、一阶微分变换光谱反射率(D(R))后,人工灌溉区选择1/R、自然降雨区选择LOG(R)使用竞争自适应重加权采样法(Competitive Adaptive Reweighting,CARS)、无信息变量消除法(Uninformative Variables Elimination,UVE)和连续投影算法(Successive Projections Algorithm,SPA)进行特征波段筛选,筛选出的波段采用偏最小二乘回归(Partial Least Squares Regression,PLSR)算法分别构建两个果园的SMC预测模型,并比较不同灌溉方式下的估测精度。结果表明:人工灌溉区采用1/R、自然降雨区采用LOG(R)时,光谱反射率与SMC的相关系数最高,分别达到0.751和0.845;CARS、SPA、UVE三种波段筛选算法下,人工灌溉区采用CARS算法、自然降雨区采用SPA算法获得的预测精度最高,两者模型的决定系数R 2和均方根误差RMSE分别为0.730、0.017和0.745、0.015。针对两区土壤表层高光谱图像,采用CARS-1/R-PLSR和SPA-LOG(R)-PLSR建立SMC估测模型,得到人工灌溉区与自然降雨区R 2分别为0.7845和0.7568。最后采用SMC模型绘制出两个试验区的土壤SMC估测图,实现对果园土壤水分含量的精准掌握及精细化管理。
In order to realize the rapid estimation of soil moisture cotent(SMC)of large area orchard,uav hyperspectral imager was used to take pictures of an apple demonstration garden under artificial irrigation and an apple orchard under natural rainfall in Jingning County,Gansu Province.Soil surface soil spectral image,comprehensive comparison and a total of 240 soil samples of the original spectral reflectance(R),inverse transform spectral reflectance(1/R),spectral reflectivity logarithm change(LOG(R)),the square root of transform spectral reflectance(R),first order differential transform spectral reflectance(D(R)),1/R was selected for artificial irrigation area and LOG(R)was selected for natural rainfall area.Competitive Adaptive Reweighting(CARS)and Uninformative Variables were used Elimination(UVE)and the continuous projection Algorithm(SPA)for feature band selection.SMC prediction models of the two orchards were constructed by Partial Least Squares Regression(PLSR)algorithm,and the estimation accuracy of SMC models under different irrigation methods was compared.The results show that when 1/R is used in artificial irrigation area and LOG(R)is used in natural rainfall area,the correlation coefficient between spectral reflectance and SMC is the highest,reaching 0.751 and 0.845,respectively.Under the three band screening algorithms of CARS,SPA and UVE,CARS algorithm in artificial irrigation area and SPA algorithm in natural rainfall area achieved the highest prediction accuracy,and the determination coefficient R2 and root mean square error RMSE of the two models were 0.730 and 0.017 and 0.745 and 0.015,respectively.Based on the hyperspectral images of soil surface in the two areas,THE SMC estimation model was established by CARS-1/R-PLSR and SPA-Log(R)-PLSR,and the R2 of artificial irrigation area and natural rainfall area was 0.7845 and 0.7568,respectively.Finally,the SMC model was used to draw the ESTIMATION map of soil SMC in the two test areas to realize the accurate grasp and fine management of soil water content in the orchard.
作者
夏媛媛
冯全
杨森
郭发旭
XIA Yuan-yuan;FENG Quan;YANG Sen;GUO Fa-xu(School of Mechanical and Electrical Engineering,Gansu Agricultural University,Lanzhou Gansu 730070,China)
出处
《林业机械与木工设备》
2023年第4期24-32,共9页
Forestry Machinery & Woodworking Equipment
基金
兰州市科技局计划项目(2021-1-149)
甘肃省高等学校产业支撑引导项目(2019C-11)。
关键词
高光谱
土壤水分含量
偏最小二乘回归
波段筛选
hyperspectral soil moisture content
partial least squares regression
band screening
