摘要
针对变化检测常用的面向对象影像分析(object based image analysis,OBIA)技术中规则数量巨大、时空普适性差、深度学习方法样本获取困难且精度难以满足工程化需求的现状,提出了一种将面向对象影像分析和对象卷积神经网络(object convolutional neural network,OCNN)相结合的耕地变化检测方法。以基期(T_(1))耕地矢量为约束条件,当期(T_(2))高分辨率影像OCNN土地覆被分类结果为主要判断依据,结合知识规则,进行耕地地块层和对象层变化检测。为了验证本文提出的耕地变化检测方法的有效性,采用两个实验区域,将OCNN分别与卷积神经网络(convolutional neural network,CNN)、VGG(visual geometry group)检测结果进行比较。结果表明,该方法在效率和精度上都显著优于基于CNN与VGG网络的方法。
Rulesets used for object based change detection(OBCD) has the disadvantage of poor spatial-temporal generalization.Meanwhile,deep learning for change detection of multi-temporal images is difficult to obtain samples and cannot meet accuracy requirements of engineering projects.To address above issues,a novel farmland change detection method built on a combination of object based image analysis(OBIA) and object convolutional neural networks(OCNN) is proposed.T_(1) farmland vector is used as constraint condition.OCNN based land cover classification result of T_(2) high resolution image is used as proof of changes.In the same time,knowledge rule is integrated in the model to gain the parcel and object hierarchical change detection.In order to verify the effectiveness of the proposed farmland change detection method,OCNN is compared with convolutional neural networks(CNN) and VGG(visual geometry group)in two experimental areas.The results show that the proposed method is significantly better in efficiency and accuracy.
作者
徐志红
关元秀
王善华
容俊
唐紫晗
XU Zhihong;GUAN Yuanxiu;WANG Shanhua;RONG Jun;TANG Zihan(Zhejiang Natural Resources Surveying and Registration Center,Hangzhou 310012,China;PIESAT Information Technology Co.Ltd.,Beijing 100195,China)
出处
《遥感信息》
CSCD
北大核心
2022年第5期15-22,共8页
Remote Sensing Information
基金
浙江省自然资源厅2022年度科技项目(2022-80)。
关键词
耕地
深度学习
面向对象影像分析
对象卷积神经网络
变化检测
farmland
deep learning
object based image analysis
object convolutional neural networks
change detection