摘要
针对图像语义分割中存在的边缘模糊和准确度较低的问题,提出一种基于边缘感知强化空间细节的图像语义分割方法,在语义分割网络中引入边缘检测模块,以捕获更加精细的空间细节。模型采用编码器-解码器结构,使用空间金字塔池化模块(atrous spatial pyramid pooling,ASPP)提取语义信息;提出双向多级聚合模块(bi-directional multi-level aggregation,BMLA)生成边缘特征,并使其强化空间细节;设计一种新型的基于注意力机制的特征融合模块(attention feature fusion module,AFFM),将强化后的空间特征与语义特征融合。采用Cityscapes和ADE20K数据集进行实验,对比其他主流语义分割算法,该方法在分割性能上具有不错的竞争力。
To address the issues of blurred edges and low accuracy in image semantic segmentation,this paper proposes an image semantic segmentation method based on edge-aware reinforcement of spatial details.An edge detection module is introduced into the semantic segmentation network to capture finer spatial details.The model adopts an encoder-decoder architecture and uses the atrous spatial pyramid pooling(ASPP)module to extract semantic information.A bi-directional multi-level aggregation(BMLA)module is proposed to generate edge features and reinforce spatial details.A novel attention-based feature fusion module(AFFM)is designed to integrate the enhanced spatial features with the semantic features.Experiments on the Cityscapes and ADE20K datasets show that,compared with other mainstream semantic segmentation algorithms,the proposed method demonstrates strong competitiveness in segmentation performance.
作者
刘伯红
蒋佳跞
LIU Bohong;JIANG Jialuo(School of Computer Science and Technology,Chongqing University of Posts and Telecommunications,Chongqing 400065,P.R.China)
出处
《重庆邮电大学学报(自然科学版)》
北大核心
2025年第1期57-66,共10页
Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)
基金
国家自然科学基金项目(62272075)。
关键词
语义分割
边缘检测
编码器-解码器
注意力机制
semantic segmentation
edge detection
encoder-decoder
attention mechanism
