基于多尺度特征交叉融合的遥感图像变化检测

Detection of Changes in Remote Sensing Image Based on Multi-scale Feature Cross Fusion

针对传统变化检测方法检测小目标时存在漏检误检、对细节信息的分割不精细等现象,提出了一种基于多尺度特征交叉融合的遥感图像变化检测方法。采用编码器-解码器结构,在编码器和解码器之间引入多尺度特征融合(MSFF)模块和多尺度注意力(MSA)机制,前者用来聚合不同尺度的信息,后者用来捕获不同尺度之间的差异和联系。在编码器最后引入细化输出(RO)模块,采用并行卷积并加入膨胀和空洞进一步细化细节信息减少边缘信息损失。在LEVIR-CD数据集上进行实验,结果表明所提方法能够在有效识别小目标的同时兼顾细节信息,在精度、F1分数、交并比、总体准确率上有明显的提升。

A remote sensing image change detection method based on multi-scale feature cross-fusion is proposed to address the issues of missed detections and false alarms for small targets as well as insufficient segmentation of fine details in traditional change detection methods.The method adopts an encoder-decoder structure with the introduction of Multi-Scale Feature Fusion(MSFF)module and Multi-Scale Attention(MSA)mechanism between the encoder and decoder.The former is used to aggregate information from different scales while the latter captures differences and correlations between different scales.A Refinement Output(RO)module is introduced at the end of the encoder using parallel convolutions with dilation and atrous convolutions to further refine the fine details and reduce edge information loss.Experimental results on the LEVIR-CD dataset demonstrate that the proposed method can effectively identify small targets while preserving fine details leading to significant improvements in accuracy F1 score intersection-over-union and overall accuracy.