Research on Semantic Image Segmentation Network Combining Attention Mechanism and Multi-Scale Features
Year: 2024
Type: article
Abstract: 针对卷积神经网络在图像语义分割时存在部分语义信息丢失、边界定位精度较低等问题, 提出联合注意力机制和多尺度特征的卷积神经网络.首先基于注意力机制将网络提取到的多尺度特征进行加权融合, 然后采用扩张卷积和全局平均池化聚合多尺度目标信息, 最后采用边界精细粒度特征提取模块对分割边界进行优化.在多尺度PASCAL VOC2012和高分辨率Cityscapes数据集上的实验结果表明, 所提网络的分割效果显著优于骨干网络ResNet-101, 平均交并比分别提高12.2个百分点和9.3个百分点.
Authors: Rui Zhang, Mengxuan Liu, Xiaoman Meng, Yichao Wu
Institution: China United Network Communications Group (China)
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W7106009633
FWCI:
Cites:
Cited by:
Related to: