一种基于MBFF-Net的遥感影像建筑物提取方法

A method of building extraction from remote sensing images based on MBFF-Net

针对当前基于卷积神经网络的建筑物提取存在漏检、错检和边缘不准确的问题,提出一种多分支特征融合的建筑物提取网络MBFF-Net。首先,在VGG16-UNet网络的跳跃连接部分引入CBAM注意力机制,强化网络对建筑物特征信息的学习;然后,设计多分支特征融合模块替换解码器中的卷积块,融合不同感受野的特征信息,捕捉局部和跨通道的特征关系,丰富特征表达能力;最后,结合CBAM注意力机制和多分支特征融合模块构建MBFF-Net模型,并在WHU数据集和Inria数据集上进行验证。结果表明,与U-Net、PSPNet、SegNet、VGG16-UNet相比,MBFF-Net在IoU、Precision、Recall以及mPA 4个指标上均为最优,提取的建筑物更加完整,减少了错检和漏检现象,在建筑物提取任务中表现出良好的性能,验证了其在建筑物提取方面的可行性。

Aiming at the problems of missing detection,false detection and inaccurate edge in building extraction based on convolutional neural network,a Multi-Branch Feature Fusion building extraction network MBFF-Net is proposed.Firstly,CBAM attention mechanism is introduced into the cope and crop part of VGG16-UNet to strengthen the learning of building feature information;Then,a Multi-Branch Feature Fusion Module is designed to replace the convolution block in the decoder,fuse the feature information of different receptive fields,enrich the feature expression ability,and capture the local and cross-channel feature relationship;Finally,MBFF-Net is constructed by combining CBAM attention mechanism and Multi-Branch Feature Fusion Module,and verified on WHU Dataset and Inria Dataset.The results show that compared with U-Net,PSPNet,SegNet and VGG16-UNet,MBFF-Net is the best in the four indicators of IoU,Precision,Recall and mPA.The extracted buildings are more complete,and the phenomenon of false detection and missing detection is reduced.MBFF-Net shows a good performance in the task of building extraction,which verifies its feasibility in building extraction.