基于状态空间模型和卷积注意力的遥感影像变化检测  

Remote Sensing Image Change Detection Based on State Space Model and Convolutional Block Attention Module

作　　者：蔡林 徐义春 张上[2,3] CAI Lin;XU Yichun;ZHANG Shang(Hubei Key Laboratory of Intelligent Vision Based Monitoring for Hydroelectric Engineering,China Three Gorges University,Yichang Hubei 443002;Hubei Province Engineering Technology Research Center for Construction Quality Testing Equipments,China Three Gorges University,Yichang Hubei 443002;College of Computer and Information Technology,China Three Gorges University,Yichang Hubei 443002)

机构地区：[1]水电工程智能视觉监测湖北省重点实验室,湖北宜昌443002 [2]湖北省建筑质量检测装备工程技术研究中心,湖北宜昌443002 [3]三峡大学计算机与信息学院,湖北宜昌443002

出　　处：《软件》2025年第3期17-24,共8页Software

基　　金：湖北省大学生创新创业训练计划(S202311075047)。

摘　　要：变化检测(CD)旨在从双时相图像中识别表面变化,基于深度学习的方法在CD领域取得了实质性的突破,但其结果还是容易受到外部因素的影响,导致检测图中出现伪变化和噪声。最近出现的状态空间模型Mamba应用在视觉领域时实现了数据依赖的全局视觉上下文,同时具有较低的计算复杂度。本文提出了一种基于状态空间模型的高分辨率双时相遥感影像变化检测网络模型CMamba-CDNet。该模型通过提取多尺度的特征并进行融合来结合浅层和深层信息,应用Mamba模型充分学习图像的全局上下文信息,引入卷积注意力模块增强特征判别性,有效解决了变化检测中的伪变化和噪声问题。在LEVIR-CD和WHU-CD数据集上的实验结果显示,CMamba-CDNet在F1和Io U指标上均优于现有的方法,分别达到了95.245%、91.24%和96.778%、93.85%,证明了其在变化检测任务中的优越性能。Change detection(CD)aims to identify surface changes from bi-temporal images.Deep learning-based methods have achieved substantial breakthroughs in the field of CD,but their results are still susceptible to external factors,leading to false changes and noise in detection maps.The recently emerged state space model Mamba,when applied in the visual domain,realizes data-dependent global visual context while maintaining low computational complexity.This paper proposes a high-resolution bi-temporal remote sensing image change detection network model based on the state space model,CMamba-CDNet.This model combines shallow and deep information by extracting multi-scale features and fusing them.The Mamba model is applied to fully learn the global contextual information of the image,and a convolutional attention module is introduced to enhance feature discriminability,effectively solving the problems of pseudo changes and noise in change detection.Experimental results on the LEVIR-CD and WHU-CD datasets show that CMamba-CDNet outperforms existing methods in terms of F1 and IoU metrics,reaching 95.245%,91.24%,and 96.778%,93.85%respectively,proving its superior performance in change detection tasks.

关 键 词：深度学习 变化检测 状态空间模型 遥感图像处理 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]