%0 Journal Article %T 农村宅基地动态变化监测技术研究 %T Study on monitoring technology of rural homestead dynamic changes %A 陆国稳,李会宾,夏天,史云 %A Lu Guowen %A Li Huibin %A XiaTian %A Shi Yun %J 中国农业信息 %J China agricultural informatics %@ 1672-0423 %V 34 %N 3 %D 2022 %P 9-19 %K 宅基地;动态变化监测;融合监测模型 %K homestead;dynamic change monitoring;fusion monitoring model %X 【目的】 宅基地是保障农民安居乐业和农村社会稳定的重要基础,实时掌握农村宅基地的使用、变化、发展信息对建设社会主义新农村、促进农村地区的发展研究,提高农民生活水平具有重要意义,但是现有的变化监测研究主要针对于城市建筑物,且依旧存在大量的人工监测方法,因此迫切需要一种能对农村宅基地进行高效动态变化监测的有效手段。【方法】 文章基于面向宅基地图像的半自动标注及训练方法,对训练样本进行半自动化提取与训练。通过构建的深度学习多尺度融合监测模型对宅基地图斑进行智能化提取,依据模型对提取出的多期宅基地图斑进行叠加分析,自动提取出宅基地新增与拆除的变化图斑,以此实现农村宅基地的动态变化监测。【结果】 面向宅基地图像的半自动标注及训练模型有效实现了训练样本集的半自动化构建与训练,通过实时的影像输入,完成了样本的动态构建与训练,提高了训练效率;深度学习多尺度融合监测模型对研究区宅基地动态变化的监测效果明显,成功提取出多期宅基地建筑物的精确轮廓,识别出宅基地的变化区域,反映出宅基地变化的数量、面积、形态特征。【结论】 基于深度学习多尺度融合监测模型,对于宅基地图斑的动态变化监测效果明显,该方法适用于大规模农村宅基地的动态变化监测,可为农村宅基地的变化监测提供了一种新的技术手段与研究思路。 %X [Purpose] Homestead is the guarantee farmers to live and work in peace and contentment,and the important foundation of the rural social stability,real time information to master the use of rural residence,change and development,the construction of new socialist rural areas,promote the development of rural areas,improve farmers’ living standard is of great significance,but the existing research is aimed at monitoring the changes in the city buildings,And there are still a large number of manual monitoring methods,so there is an urgent need for an effective means to effectively monitor the dynamic change of rural homestead.Method Based on a semi-automatic annotation and training method for homestead images,semi-automatic extraction and training of training samples are carried out. Secondly,the deep learning multi-scale fusion monitoring model was used to intelligently extract the house site map spots. Finally,the superposition analysis of the extracted multi-period house site map spots was conducted according to the model to automatically extract the new and removed changed house sites,so as to realize the dynamic change monitoring of rural house sites.[Result] The semi-automatic labeling and training model for homestead images effectively realized the semi-automatic construction and training of training sample sets. Through real-time image input,the dynamic construction and training of samples were completed,and the training efficiency was improved. Deep learning multi-scale fusion monitoring model has an obvious monitoring effect on the dynamic change of homestead in the study area,successfully extracting the accurate contour of multi-period homestead buildings,identifying the changing area of homestead,and reflecting the number,area and morphological characteristics of homestead change.[Conclusion] Based on the deep learning multi-scale fusion monitoring model,the dynamic change monitoring effect of homestead map spot is obvious. This method is suitable for the dynamic change monitoring of large-scale rural homestead,which provides a new technical means and research idea for the change monitoring of rural homestead. %R 10.12105/j.issn.1672-0423.20220302 %U http://www.cjarrp.com/zgnyxx/ch/reader/view_abstract.aspx %1 JIS Version 3.0.0