基于深度学习方法的草地信息智能提取研究  

Research on Intelligent Extraction of Grassland Information Based on Deep Learning Method

作　　者：唐川江 唐古拉 鲁岩 干晓宇[3] TANG Chuan-jiang;TANG Gu-la;LU Yan;GAN Xiao-yu(Sichuan Grass Industry Technology Research and Promotion Center,Chengdu 610041,China;Key Laboratory of Agri-informatics,Ministry of Agriculture/Institute of Agricultural Resources and Regional Planning,Chinese Academy of Agricultural Sciences,Beijing 100081,China;College of Architecture and Environment,Sichuan University,Chengdu 610041,China)

机构地区：[1]四川省草业技术研究推广中心,四川成都610041 [2]中国农业科学院农业资源与农业区划研究所/农业农村部农业遥感重点实验室,北京100081 [3]四川大学建筑与环境学院,四川成都610065

出　　处：《中国草地学报》2020年第3期175-180,共6页Chinese Journal of Grassland

基　　金：四川省草地资源清查。

摘　　要：四川省草地资源清查工作中,以川西部分地区为研究区,对DeepLabv3+模型和传统监督分类方法开展对比实验。试验结果表明,DeepLabv3+提取平均精度为79.28%,比传统的监督分类解译方法精度提高了5个百分点,且草地信息提取结果连续,与人工判读的结果相近,在草地信息智能化自动化提取方面具有重要的实践价值。During the investigation of grassland resource in Sichuan province,the Deeplapv3+model was compared with the traditional supervised classification method in some western parts of Sichuan province.The results showed that the average accuracy of Deeplapv3+extraction was 79.28%,which was increased by 5%compared with that of the traditional supervised classification and interpretation methods.In addition,the results of grassland information extraction method were continuous,which were close to that of artificial interpretation.Thus,it has an important value in the intelligent and automatic extraction of grassland information.

关 键 词：深度学习 DeepLabv3+网络模型 草地信息提取 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]