基于改进高分辨率网络的多语义图像分割方法
A Multi Semantic Image Segmentation Method Based on Improved High Resolution Networks
张少杰 1彭富明 1方斌 1张子祥 1相福磊 1何浩天1
作者信息
摘要
针对室外复杂场景下图像分割难度较大的问题,提出一种基于HRNet的多语义图像分割模型(HR_DfeNet).该模型通过引入通道注意力和空间注意力模块优化特征提取,通过改进金字塔池化模块设计ASPP_M模块形成高分辨率特征提取分支,并与多种注意力机制融合.在Cityscape数据集上,HR_DfeNet相较于传统分割模型表现出不同程度的分割优化效果.
Abstract
To address the difficulty of image segmentation in complex outdoor scenes,this paper proposes a multi semantic image segmentation model based on HRNet(HR_DfeNet),which optimizes feature extraction by introducing channel attention and spatial attention modules,designs a high-resolution feature extraction branch by improving the pyramid pooling module and ASPP_M module,and integrates with multiple attention mechanisms.On the Cityscape dataset,HR_DfeNet exhibits varying degrees of segmentation optimization performance compared to traditional segmentation models.
关键词
室外复杂场景/图像分割/注意力模块/金字塔池化模块Key words
outdoor complex scenes/image segmentation/attention module/pyramid pooling module引用本文复制引用
基金项目
国家重点研发计划项目(2021YFE0194600)
江苏省科技计划项目(BZ2023023)
出版年
2024
国家科技期刊平台
NSTL
万方数据