机电工程技术2025，Vol.54Issue(2)：25-28,71,5.DOI:10.3969/j.issn.1009-9492.2025.02.005
基于多尺度特征融合与全局注意力机制的变化检测研究
Research on Change Detection Based on Multi-scale Feature Fusion and Global Attention Mechanism
摘要
Abstract
In order to enhance the detection capability of small targets and address the issue of false positives and false negatives caused by low feature utilization,a remote sensing image change detection method based on multi-scale feature fusion and global attention mechanism is proposed.This method utilizes an encoder-decoder architecture to effectively guide different scale features,resolving the multi-scale problem.The encoder employs twin residual blocks as the feature extraction network to extract features at different stages of dual-temporal images.Additionally,a local feature enhancement module is embedded at the end of feature extraction to increase the network's receptive field and maintain semantic richness.In the decoder,a global attention upsampling module is designed to enhance the network's learning of contextual information.The obtained multi-scale change feature maps are then recombined through spatial attention blocks and channel attention blocks to obtain feature maps rich in semantic information,enhancing the model's perception of different scale features.Experiments demonstrate that this method outperforms the compared models in both subjective evaluation and objective metrics on the tested public datasets.Compared to the second-best compared model,the recall rate is improved by 2.25%,and the F1 score is improved by 1.82%,significantly improving existing issues.关键词
深度学习/遥感影像/全局注意力上采样/特征融合Key words
deep learning/remote sensing images/global attention upsampling/feature fusion分类
信息技术与安全科学引用本文复制引用
孙一竣,雷斌,丁倩钰..基于多尺度特征融合与全局注意力机制的变化检测研究[J].机电工程技术,2025,54(2):25-28,71,5.基金项目
国家自然科学基金项目(72061021) （72061021）
甘肃省科技厅项目(21JR7RA284) （21JR7RA284）
