基于多模态遥感图像的特征融合模型

打开文本图片集

中图分类号：TP183 文献标识码：A DOI： 10.7535/hbgykj.2025yx06001

reature fusion model based on multi-modal remote sensing images

WANG Jianxia¹，QIU Shaozu1，YANG Chunjin²，WU Changli³， ZHANG Xiaomingl （1.Schoolof InformationScienceandEnginering，HebeiUniversityofScienceand Technology，Shijiazhuang，Hebei050018， China；2.Hebei Taihang Machinery Industry CompanyLimited，Shijiazhuang，Hebei O521oo，China；3.Schoolof Economics and Management，Hebei University of Science and Technology，Shijiazhuang，Hebei O5ool8，China）

Abstract：To address the issues such as limited model accuracy and large parameter scale of traditional single-branch networksinsemanticsegmentationof remotesensing images，alarge-kernelconvolution-basedmulti-modal feature fusion network （LMFNet） module was proposed. An improved large-kernel MobileNetV3（GMBNetV3）was adopted as the paralel backbone，and multi-source features were fused through cross-self-atention enhancement module.The gated aggregator was utilized to integrate abstract and texture information in the decoding stage.On the public datasets Potsdam and Vaihingen， LMFNet was compared with currnt advanced multi-modalimage segmentation models in terms of performance，and ablation experiments wereconducted to verify thefunctions ofeach moduleof the model.Theresults show thatLMFNet improves the segmentation performance of mIoU by approximately 0.32 percentage points ～6.50 percentage points compared to other advanced multi-modal segmentation models，while reducing the parameter quantity by 29.3%～73.6% ，and the inference speed is increased by 1.7%～45.9% on the Potsdam dataset. The proposed model effectively fuses the differences in image features and can perform semantic segmentationof remote sensing images moreclearly，providing strong support for instance segmentation of remote sensing images in urban management.

Keywords：computer neural network；large kernel convolution；remote sensing image segmentation；multi-modal; feature fusion

高分辨率遥感图像蕴含丰富的地物细节与空间特征，常被用于场景分析以实现精准的地物识别与场景理解[1-2]。（剩余16593字）