摘要
棉花是我国最重要的经济作物之一,其快速准确的提取是有效管理棉田信息的关键。针对高分辨率遥感影像获取不易,数据常有缺失,作物生长期遥感影像利用不充分的问题。本文基于深度学习方法,使用棉花生长周期内4月至10月的遥感卫星影像,探究遥感影像单个时相及其时相组合的数据对棉花提取的影响。通过Deeplabv3+模型对研究区各个时相的遥感影像的棉花分别进行提取,并使用分段函数评估模型对提取效果进行快速比较,按此结果排列出适合棉花提取的单时相遥感影像,依次加入到Deeplabv3+模型中,得到该模型下的遥感影像棉花提取结果最优时相组合为6月、7月、8月和10月,该组合下棉花提取准率Precision、Recall、F1指数、IoU指数分别为0.93、0.93、0.94、0.87,同时,并基于该结果与SegNet,U-Net两种模型相比,准确度分别高出18%和8%,能够明显减少与非棉花耕地的错分情况,对棉田的边缘恢复较好,该结果可以在棉花遥感数据不足及快速提取时提供参考。
Cotton is one of the most important economic crops in my country,and its fast and accurate extraction is the key to effective management of cotton field information.In view of the difficulty in obtaining high-resolution remote sensing images,data is often missing,and the use of remote sensing images during crop growth is insufficient.Based on the deep learning method,this paper uses remote sensing satellite images from April to October during the cotton growth cycle to explore the impact of remote sensing image data of a single time phase and its time phase combination on the extraction of cotton.The Deeplabv3+model was used to extract the cotton from the remote sensing images of each time phase in the study area,and the segmented function evaluation model was used to quickly compare the extraction results.According to the results,the single-phase remote sensing images suitable for cotton extraction were arranged and added to In the Deeplabv3+model,the optimal time combination of the remote sensing image cotton extraction results obtained under this model is June,July,August,and October.The precision extraction,Recall,F1 index and IoU index of cotton under this combination are 0.93,0.93,0.94,0.87,respectively.At the same time,and based on the results,compared with the SegNet and U-Net models,the accuracy is 18%and 8%higher,respectively,which can significantly reduce the misclassification of non-cotton farmland.The edge recovery is good,and this result can provide a reference when the cotton remote sensing data is insufficient and fast extraction.
作者
司凯凯
汪传建
赵庆展
杨启原
任媛媛
袁盼丽
SI Kaikai;WANG Chuanjian;ZHAO Qingzhan;YANG Qiyuan;REN Yuanyuan;YUAN Panli(College of Information Science and Technology/Geospatial Information Engineering Research Center,Shihezi University,Shihezi,Xinjiang 832003,China;School of Internet,Anhui University,Hefei,Anhui 230039,China)
出处
《石河子大学学报（自然科学版）》
CAS
北大核心
2022年第5期639-647,共9页
Journal of Shihezi University(Natural Science)
基金
国家重点研发计划项目(2017YFB0504203)。
