现代信息科技2024，Vol.8Issue(18)：71-74,4.DOI:10.19850/j.cnki.2096-4706.2024.18.014
基于深度流场特征和语义约束的改进SegFormer语义分割算法
Improved SegFormer Semantic Segmentation Algorithm Based on Deep Flow Field Feature and Semantic Constraint
摘要
Abstract
For the problems that the multi-scale information existing in the SegFormer network cannot be effectively utilized and the boundary contour of the prediction result is unclear,an improved semantic segmentation network architecture based on deep flow field feature and semantic constraint is proposed.Firstly,a deep flow field module is added to the decoder part to enhance the consistency of depth feature.Then in order to keep the lightweight of the original network,the boundary and foreground and background auxiliary task are added to form a semantic constraint module to improve the ability of network to extract boundary and overall contour.Finally,the boundary-guided module is added to the semantic constraint module to speed up the convergence of auxiliary task.By increasing the number of parameters by 0.1 M,the segmentation accuracy of the network is improved.关键词
SegFormer/语义分割/轻量化/深度流场/辅助任务Key words
SegFormer/semantic segmentation/lightweight/deep flow field/auxiliary task分类
信息技术与安全科学引用本文复制引用
高延海..基于深度流场特征和语义约束的改进SegFormer语义分割算法[J].现代信息科技,2024,8(18):71-74,4.
