摘要
目前,状态空间模型的Mamba架构因独特的扫描方法,可有效解决卷积神经网络架构固有感受野限制和Transfomer架构存在的计算复杂度的问题,在遥感影像变化检测任务中迅速采用。Mamba架构主要考虑全局上下文信息,忽略影像的局部细节信息。因此,为实现变化信息的有效提取,提出基于并行卷积通道注意力遥感变化检测模型(PCA-RSM),该模型融合特征图的局部和全局信息,并引入通道注意力机制增加特征图在通道维度的关注度。在公开数据集WHU-CD与LEVIR-CD的实验结果表明,PCA-RSM相较于主流模型,在小目标与复杂背景目标场景具有更优异的表现,在WHU-CD数据集中精确率、召回率、F1分数、IoU指标分别为94.25%、89.85%、92.00%、85.18%,在LEVIR-CD数据集中其4个指标分别达到91.04%、88.81%、89.91%、81.68%,表现出PCA-RSM可有效区分小目标和复杂目标的变化信息,提高遥感影像变化检测效果。
At present,the Mamba architecture based on the state space model has been rapidly applied in the task of remote sensing image change detection due to its unique scanning method,which can effectively solve the inherent receptive field limitations of the convolutional neural network architecture and the problem of secondary computational complexity existing in the Transformer architecture.However,the Mamba architecture primarily focuses on global contextual information while overlooking the crucial role of local details in remote sensing image change detection tasks.Therefore,to achieue the effective extraction of change information,we propose a remote sensing change detection Model named Parallel Convolution Attention-Remote Sensing Model(PCA-RSM),which integrates the local and global information of the feature map and introduces the channel attention mechanism to increase the attention of the feature map in the channel dimension.Experimental results on the public datasets WHU-CD and LEVIR-CD demonstrate that PCA-RSM outperforms mainstream models in scenarios involving small targets and complex backgrounds.In the WHU-CD dataset,it has achieved a precision of 94.25%,recall of 89.85%,F1-score of 92.00%,and IoU of 85.18%.In the LEVIR-CD dataset,it attained precision of 91.04%,recall of 88.81%,F1-score of 89.91%,and IoU of 81.68%.These results indicate that PCA-RSM can effectively identify change information for both small and complex targets,thereby enhancing the performance of remote sensing image change detection.
作者
朱锋
林飞
冯福
李伯昊
曹巧
王雪
ZHU Feng;LIN Fei;FENG Fu;LI Bohao;CAO Qiao;WANG Xue(Hefei Institutes of Collabrative and Innovation for Intelligent Agriculture,Hefei 231131,China;Zhongke Hefei Intelligent Agricultural Valley Co.,Ltd,Hefei 230036,China;Hefei Institutes of Physical Science,Chinese Academy of Sciences,Hefei 230031,China;The Third Geological Brigade of Sichuan Province,Chengdu 610000,China)
出处
《测绘工程》
2025年第6期40-48,55,共10页
Engineering of Surveying and Mapping
基金
国家重点研发计划项目(2023YFD1702100,2024YFD1700100)
生态环境部试点项目(2023ADDFZ00164)。
关键词
遥感影像
变化检测
状态空间模型
并行卷积
通道注意力机制
remote sensing image
change detection
state space model
parallel convolution
channel attention
