摘要
针对现有图斑变化检测方法难以满足土地监管的实际生产需求问题,提出了一套基于深度学习与矢-栅数据的图斑变化检测自动化处理系统。首先利用高分遥感影像和矢量数据,以强边界约束为支持生成自适应大小的自标注样本集,并利用不同分类器对训练样本进行纯化;然后利用改进的深度卷积神经网络模型实现图斑的细粒度分类;最后根据制定的判别规则,筛选预测值与真实值不一致的矢量图斑,作为最终的变化检测结果。以无锡洋溪河地区为研究区域,结果表明,该系统不仅能够准确高效定位变化图斑,而且大幅减轻了人工工作量,在耕地“非农化”“非粮化”以及违法用地检查等变化检测领域具有较高的应用价值。
Aiming at the problem that the existing polygon change detection methods can not meet the actual production needs of land supervision,an automatic polygon change detection processing system based on depth learning and vector-grid data is proposed.First,using high resolution remote sensing images and vector data,a self-labeled sample set with adaptive size is generated with strong boundary constraints,and different classifiers are used to purify the training samples.Then,the improved depth convolution neural network model is used to realize the fine granularity classification of patches.Finally,according to the established discriminant rules,the vector patches whose predicted values are inconsistent with the true values are selected as the final change detection results.Taking Yangxi River area in Wuxi as the research area,the results show that the system can not only accurately and efficiently locate change polygons,but also greatly reduce the manual workload,and has high application value in the change detection fields such as“non-agricultural”“non grain”and illegal land use inspection.
作者
单浩宇
王春晓
尹鹏程
李二珠
张连蓬
史嘉诚
刘伟
SHAN Hao-yu;WANG Chun-xiao;YIN Peng-cheng;LI Er-zhu;ZHANG Lian-peng;SHI Jia-cheng;LIU Wei(School of Geography,Geomatics and Planning,Jiangsu Normal University,Xuzhou Jiangsu 221116,China;Hainan Geometics Center,Ministry of Natural Resources,Haikou Hainan 570203,China;Xuzhou Natural Resources and Planning Bureau,Xuzhou Jiangsu 221018,China)
出处
《现代测绘》
2022年第6期16-19,共4页
Modern Surveying and Mapping
基金
江苏省自然资源厅科技创新项目(2021046)
自然资源部国土卫星遥感应用重点实验室开放基金资助(KLSMNR-K202206)
徐州市重点研发计划(社会发展)项目(KC20172)
江苏师范大学研究生科研创新计划项目(2021XKT0133)
关键词
变化检测
深度学习
高分辨率遥感影像
自动化处理系统
change detection
deep learning
high resolution remote sensing image
automatic processing system
作者简介
第一作者:单浩宇,硕士研究生,研究方向为深度学习、遥感图像信息处理;通讯作者:刘伟,副教授,研究方向为空间数据质量检查、遥感图像处理以及GIS开发。
