孪生高分辨率卷积神经网络遥感影像变化检测方法  

Siamese High-resolution Convolutional Neural Network Remote Sensing Image Change Detection Method

作　　者：王晓雯 戴晨光[1] 张振超[1] 季虹良 WANG Xiaowen;DAI Chenguang;ZHANG Zhenchao;JI Hongliang(School of Gea-spatial Information,Information Engineering University,Zhengzhou 450001,China)

机构地区：[1]信息工程大学地理空间信息学院,河南郑州450001

出　　处：《测绘与空间地理信息》2023年第5期70-73,共4页Geomatics & Spatial Information Technology

基　　金：国家自然科学基金(42071340)资助。

摘　　要：针对遥感影像变化检测问题,提出了一种孪生高分辨率卷积神经网络模型。该模型首先基于孪生网络模型提取不同时相遥感影像的特征,然后将特征拼接后输入到嵌套U形网络中输出变化检测区域。为了提升变化检测效果,进一步设计了高分辨率卷积神经网络用于提取不同时相遥感影像的特征,以充分利用不同分辨率的特征来提升变化检测效果。在LEVIR-CD变化检测数据集上的大量实验表明,所提出方法能够比对比方法获得更高的变化检测精度。Aiming at the problem of remote sensing image change detection,a Siamese high-resolution convolutional neural network model is proposed.The model first extracts the features of remote sensing images in different phases based on the Siamese network model,and then splices the features and inputs them into the nested U-shaped network to output the change detection area.In order to improve the change detection effect,a high-resolution convolutional neural network is further designed to extract the features of remote sensing images of different time phases,so as to make full use of the features of different resolutions to improve the change detection effect.Extensive experiments on the LEVIR-CD change detection dataset show that the proposed method can achieve higher change detection accuracy than the comparison methods.

关 键 词：遥感影像变化检测 深度学习 高分辨率网络 UNet++ 

分 类 号：P237[天文地球—摄影测量与遥感]