摘要
采用两期无人机可见光遥感图像,对灌浆期冬小麦倒伏图像特征及倒伏信息提取方法进行研究。从增强图像空间域方面,对图像进行二次低通滤波,获取地物散点图,以散点存在明显分界线为判定标准,选出小麦倒伏信息提取的单特征,对两单特征线性拟合构建倒伏小麦两时期提取特征参数F_1和F_2,再以两特征参数相似性构建综合特征参数F_3。将特征参数结合K-means算法提取冬小麦倒伏信息,整体精度(OA)达86. 44%以上,Kappa系数达0. 73以上,倒伏信息提取精度(F)为81. 07%以上,因此综合特征参数可作为两个时期冬小麦倒伏信息提取特征参数。分别用本文方法、支持向量机、神经网络法和最大似然法提取验证区域倒伏小麦信息,经验证,本文方法提取小麦倒伏信息整体精度(OA)达86. 29%以上,Kappa系数达0. 71以上,倒伏信息提取精度(F)达80. 60%以上;其他3种常用方法提取的整体精度(OA)为69. 68%~87. 44%,Kappa系数为0. 49~0. 72,倒伏信息提取精度(F)为65. 33%~79. 76%。结果表明,本文方法整体精度和倒伏信息提取精度均高于目前常用分类方法。因此,综合特征参数与K-means算法对冬小麦在灌浆期倒伏信息提取具有一定的准确性和适用性。
The information of crop lodging is very important for agricultural hazard assessment and agricultural insurance claims. Remote sensing is a fast and efficient technology to gain the information of crop lodging, but satellite remote sensing cannot provide available data. Recently, unmanned aerial vehicle (UAV) remote sensing system has grown rapidly, and UAV remote sensing system can get available data neatly and fleetly. There was no survey on winter wheat lodging by using multi-temporal UAV remote sensing data. Therefore, a survey method of winter wheat lodging was proposed by using images derived from the UAV remote sensing experiments, which were carried out in the winter wheat test field of Institute of Water-Saving Agriculture in Arid Areas of China (IWSA), Northwest A&F University on May 4th and 16th of 2017. Images were handled with the second low pass filter firstly to enhance the image space domain. Then the scatter diagram of lodging and unlodging wheat was gained in different feature combination coordinate systems. The single features of wheat lodging information extraction based on the well-defined boundary of the scatter diagram were selected. Feature parameters F 1 and F 2 were gained by fitting boundary points of May 4th and 16th. Using the similarities of F 1 and F 2 can obtain F 3 to extract winter wheat lodging information of two periods. Using F 1 , F 2 and F 3 combined with K-means to extract the lodging information of winter wheat. It was turned out that the overall accuracy was over 86.44%, the Kappa coefficient was over 0.73, and the lodging extracting accuracy was over 81.07%, so F 3 can be the feature parameter to extract the lodging information of two periods. To research the accuracy and versatility of this method, two verification areas were selected and the method of this paper, support vector machine (SVM), neural network and maximum likelihood method were respectively used to extract the lodging information of winter wheat. The results showed that the overall accuracy, Kappa coefficient and lodging extracting accuracy of the method were over 86.29%, 0.71 and 80.60%, and the overall accuracy, Kappa coefficient and lodging extracting accuracy of the other common methods were 69.68%~ 87.44%, 0.49~0.72 and 65.33%~79.76%, respectively. The results indicated that the overall accuracy, Kappa coefficient and lodging extracting accuracy of this method were all tower over other methods. Therefore, the proposed method was accurate and versatile to extract the lodging information of winter wheat in the watery stage.
作者
李广
张立元
宋朝阳
彭曼曼
张瑜
韩文霆
LI Guang;ZHANG Liyuan;SONG Chaoyang;PENG Manman;ZHANG Yu;HAN Wenting(College of Mechanical and Electronic Engineering, Northwest A&F University, Yangling, Shaanxi 712100, China;Key Laboratory of Agriculture Internet of Things, Ministry of Agriculture and Rural Affairs, Yangling, Shaanxi 712100, China;College of Natural Resources and Environment, Northwest A&F University, Yangling, Shaanxi 712100, China;Institute of Soil and Water Conservation, Chinese Academy of Sciences & Ministry of Water Resources, Yangling, Shaanxi 712100, China;Institute of Soil and Water Conservation, Northwest A&F University, Yangling, Shaanxi 712100, China)
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2019年第4期211-220,共10页
Transactions of the Chinese Society for Agricultural Machinery
基金
国家重点研发计划项目(2017YFC0403203)
旱区作物需水无人机遥感与精准灌溉技术及装备研发平台项目(2017-C03)
陕西省水利科技项目(2017SLKJ-7)
关键词
冬小麦
倒伏
无人机遥感
图像处理
多时相
K-MEANS
winter wheat
lodging
unmanned aerial vehicle remote sensing
image processing
multi-temporal
K-means