摘要
充分融合遥感影像的光谱、时相和空间特征,可有效提取农作物的结构信息。首先以2022年10月—2023年6月的Sentinel-2遥感影像为数据源,根据冬小麦关键物候信息与对应的归一化植被指数的时间序列变化情况构建冬小麦种植结构信息提取指数模型;然后利用2020—2022年国土变更调查数据和高分辨率遥感影像快速制作耕地样例库;最后根据该样例库和2023年高分辨率遥感影像,利用Mask-R-CNN模型提取南通市2023年耕地地块,进一步优化指数模型提取的冬小麦分布结果。提取结果与南通市官方统计数据的相对误差为5.66%。基于海安市2022年度国土变更调查外业举证数据抽样验证冬小麦提取结果,总体精度为93.90%,Kappa系数为0.87。研究方法可为冬小麦的快速提取与分析提供参考。
The structural information of crops can be effectively extracted by fully integrating the spectral,phase and spatial features of remote sensing images.Firstly,taking Sentinel-2 remote sensing images from October 2022 to June 2023 as data sources,we established an index model of winter wheat planting structure information extraction according to the time series changes of key phenological information of winter wheat and corresponding normalized difference vegetation index.Then,we used the territory change survey data from 2020 to 2022 and high-resolution remote sensing images to produce a cultivated land sample library.Finally,based on this sample library and high-resolution remote sensing images in 2023,we used Mask-R-CNN model to extract the cultivated land plots of Nantong City in 2023,to further optimize the distribution results of winter wheat extracted by the index model.The relative error of extracted results is 5.66%,compared with the official statistical data of Nantong City.Based on the field evidence data of Hai’an City in 2022,the overall accuracy of winter wheat is 93.90%,and the Kappa coefficient is 0.87.The research method in this paper provides a reference for the rapid extraction and analysis of winter wheat.
作者
刘善磊
李梦梦
孙长奎
LIU Shanlei;LI Mengmeng;SUN Changkui(Provincial Geomatics Centre of Jiangsu,Nanjing 210013,China;Key Lab of Natural Resources Monitoring,Department of Natural Resources of Jiangsu Province,Nanjing 210013,China)
出处
《地理空间信息》
2025年第7期111-114,共4页
Geospatial Information
基金
江苏省自然资源科技资助项目(2022005)。
关键词
时间序列
深度学习
信息提取
冬小麦
time series
deep learning
information extraction
winter wheat
