计算机与数字工程2025，Vol.53Issue(2)：528-534,598,8.DOI:10.3969/j.issn.1672-9722.2025.02.040
基于U-Net多任务学习的人体分割与关键点检测研究
Research on Body Segmentation and Key Point Detection Based on U-Net Multi-task Learning
刘宇征 1佟维妍1
作者信息
- 1. 沈阳工业大学化工过程自动化学院 辽阳 111003
- 折叠
摘要
Abstract
For adolescents with idiopathic scoliosis,conventional testing methods have a low positive predictive value,re-quire radiography,and limit radiation.In recent years,more and more researchers have developed and validated the feasibility and effectiveness of deep learning algorithms in scoliosis detection.The paper proposes a multi-task learning model based on the parallel decoder structure of the U-Net network and enhances the feature extraction capability of the network by improving the convolution module in the network.The model involves two tasks,i.e.,human segmentation and detection of human key points.The design idea of the model is to acquire segmentation models by sharing knowledge between different but related tasks.The paper performs model validation and evaluation using 139 uncovered back images of the human body.The results show that the multi-task model proposed in this paper improves the segmentation intersection ratio by 2.44%over the benchmark U-Net model,improves the ACC by 1.5%over the key point detection task performed alone,and alleviates the training conditions,and enhances the generality of the model.关键词
深度学习/多任务学习/语义分割/关键点检测Key words
deep learning/multi-task learning/semantic segmentation/key points detection分类
信息技术与安全科学引用本文复制引用
刘宇征,佟维妍..基于U-Net多任务学习的人体分割与关键点检测研究[J].计算机与数字工程,2025,53(2):528-534,598,8.
