轻量级双侧输入D-WNet航空影像建筑物变化检测

Lightweight Bilateral Input D-WNet Aerial Image Building Change Detection

针对传统语义分割网络变化检测结果易受阴影及其他地物干扰、建筑物边界分割较为粗糙的问题,提出一种轻量级双侧输入的变化检测网络D-WNet。新网络从W-Net出发,采用深度可分离卷积块和空洞空间金字塔池化模块代替原本繁琐的卷积和下采样过程,利用右侧线特征编码器加强高低维特征的融合,同时在解码器上采样部分引入通道和时空注意力机制获取网络在不同维度下的有效特征,得到的D-WNet在性能方面有明显提升。在公开的WHU和LEVIR-CD建筑物变化检测数据集上进行实验,并与W-Net、U-Net、ResNet、SENet和DeepLabv3+语义分割网络进行对比。实验结果表明,D-WNet在交并比、F1值、召回率、准确率和运行时间等5项指标中综合表现优异,对阴影干扰及建筑物边缘区域具有更精确的变化检测结果。

A lightweight dualinput change detection network,DWNet,is proposed to address the issues of traditional semantic segmentation networks being susceptible to interference from shadows and other ground objects,as well as the rough boundary segmentation of buildings.The new network starts with WNet and uses deep separable convolutional blocks and hollow space pyramid pooling modules to replace the originally cumbersome convolutional and downsampling processes.It utilizes a rightline feature encoder to enhance the fusion of highdimensional and highdimensional features and introduces channels and spatiotemporal attention mechanisms in the sampling section of the decoder to obtain effective features of the network in different dimensions.The resulting DWNet has significantly improved performance.Experiments were conducted on the publicly available WHU and LEVIRCD building change detection datasets,and the results were compared with the WNet,UNet,ResNet,SENet,and DeepLabv3+semantic segmentation networks.The experimental results show that DWNet performs well in five indicators(intersectiontointersection ratio,F1 value,recall rate,accuracy rate,and running time)and has more accurate change detection results for shadow interference and building edge areas.