基于差分注意力孪生金字塔Transformer的高分辨率遥感影像变化检测方法  

The High-Resolution Remote Sensing Image Change Detection Based on Differential Feature Fusion Siamese Pyramid Transformer Model

作　　者：李梦晨 吕鹏远 LI Mengchen;LV Pengyuan(School of Information Engineering,Ningxia University,Yinchuan 750021,China;Ningxia Key Laboratory of Artificial Intelligence and Information Security for Channeling Computing Resources from the East to the West,Yinchuan 750021,China)

机构地区：[1]宁夏大学信息工程学院,宁夏银川750021 [2]宁夏“东数西算”人工智能与信息安全重点实验室,宁夏银川750021

出　　处：《测绘地理信息》2024年第5期66-72,共7页Journal of Geomatics

基　　金：国家自然科学基金(42001307);宁夏自然科学基金(2022AAC03053)

摘　　要：近年来,基于深度学习的方法在遥感影像变化检测任务上取得了实质性的突破。其中,基于Transformer的方法可以直接建模影像全局信息,在变化检测领域得到了研究与应用。然而现有基于Transformer的变化检测方法存在仅关注全局信息建模而忽略了局部信息重要性的问题。针对上述问题,本文提出一种基于差分特征融合的孪生金字塔Transformer网络(DSTR)。该模型使用孪生金字塔Transformer结构建模编码器用于多尺度特征提取,使用多层反卷积构建解码器。同时提出一种差分特征融合模块,利用差分注意力机制融合不同尺度双时相特征差异信息,提高模型变化信息提取能力。在两个开源变化检测数据集上的实验表明,该方法相比于目前其他方法取得了更优的检测效果。In recent years,deep learning(DL)techniques have made substantial breakthroughs in the field of remote sensing image change detection tasks.The Transformerbased method can better model global image information and has been studied and applied in the field of change detection.However,the existing transformer-based change detection methods only focus on the modeling of global information and ignore the importance of local details.To solve the above problems,this paper proposes a Siamese pyramid Transform-er change detection network based on differential feature fusion(DSTR).The model uses Siamese Pyramid Transform-er to build an encoder for multi-scale feature extraction,and a decoder using multilayer deconvolution to recover the original resolution of the feature map.At the same time,a differential feature fusion module is proposed,which uses the differential attention mechanism to fuse the difference information of dualtemporal features at different scales to improve the model change information extraction ability.Experiments on two public change detection datasets show that the proposed method has achieved better detection results than the current relevant cutting-edge methods.

关 键 词：变化检测 Transformer方法 孪生网络 高分辨率遥感图像 

分 类 号：P237[天文地球—摄影测量与遥感]