慢特征分析引导的多级注意力自编码器遥感图像变化检测  

Remote sensing image change detection based on slow feature analysisguided multi-level attention autoencoder

作　　者：刘金玲 陈红珍 李辰征 聂宏宾 李立钢 LIU Jinling;CHEN Hongzhen;LI Chenzheng;NIE Hongbin;LI Ligang(Key Laboratory of Electronics and Information Technology for Space Systems,National Space Science Center,Chinese Academy of Sciences,Beijing 100190,China;School of Computer Science and Technology,University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院国家空间科学中心复杂航天系统电子信息技术重点试验室,北京100190 [2]中国科学院大学计算机科学与技术学院,北京100049

出　　处：《测绘通报》2024年第9期67-73,共7页Bulletin of Surveying and Mapping

基　　金：民用航天技术预先研究项目(D030312)。

摘　　要：遥感图像变化检测是识别和量化地表变化的一种重要途径,是遥感技术的主要应用之一。然而在不同光照、季节等成像条件下获取的遥感图像中,同一物体会表现出不同的外观,使得变化检测算法难以准确判别真实地表变化。针对此问题,提出了基于慢特征分析引导的多级注意力自编码器(SFAMAA)遥感图像变化检测方法。首先,设计了一种U型卷积自编码器并引入多级通道注意力机制,扩大网络感受野的同时使其聚焦重要通道的信息,增强网络对全局信息和变化信息的感知能力;然后,设计了一种慢特征分析损失函数引导网络训练,使得网络可以有效抑制因成像条件差异导致的伪变化。在公开数据集SZTAKI上进行试验验证,试验结果表明,本文方法可有效抑制噪声和伪变化,对不同光照、季节等成像条件下获取的遥感图像具有较高的精度和良好的稳健性。Remote sensing image change detection is an important way to identify and quantify surface changes,and is one of the main applications of remote sensing technology.However,in remote sensing images obtained under different imaging conditions such as lighting and seasons,the same object may exhibit different appearances,making it difficult for change detection algorithms to accurately distinguish real ground changes.A remote sensing image change detection method based on slow feature analysis guidance multi-level attention autoencoder(SFAMAA)is proposed to address this issue.Firstly,a U-shaped convolutional autoencoder is designed and a multi-level channel attention mechanism is introduced to expand the network s receptive field while focusing on important channel information,enhancing the network s perception of global and changing information;In addition,a slow feature analysis loss function is designed to guide network training,enabling the network to effectively suppress pseudo changes caused by differences in imaging conditions.Experimental verification is conducted on the public dataset SZTAKI,and the results show that the proposed method can effectively suppress noise and pseudo changes,and has high accuracy and good robustness for remote sensing images obtained under different imaging conditions such as lighting and seasons.

关 键 词：自编码器 通道注意力 慢特征分析 变化检测 

分 类 号：P23[天文地球—摄影测量与遥感]