基于手势识别的DeepLabV3+算法研究
Research on DeepLabV3+ Algorithm Based on Gesture Recognition
王宇 1潘景浩 1巫朝明 1陈宗岩 1王雅宁 1谢跃1
作者信息
- 1. 南京工程学院 信息与通信工程学院,江苏 南京 211167
- 折叠
摘要
文章为解决手势识别研究中欠缺考虑多时相、特征多样性的问题,提出了一种基于改进DeeplabV3+模型的手势识别提取方法.通过更改模型中ASPP模块结构,使用多个不同的空洞率以及图像金字塔池化等操作,增加CBAM注意力机制模块,提升模型的提取精度和效率.在公开Freihand数据集上进行验证,结果表明:改进后的DeeplabV3+模型训练速度提高了29.2%,识别精确度提升了0.04%,相似度提升了0.68%,召回率提高了0.36%.
Abstract
In order to solve the problems that multi-temporal and feature diversity are not considered in gesture recognition research,this paper proposes a gesture recognition extraction method based on improved DeeplabV3+model.By changing the ASPP module structure in the model,using multiple different void rates and image Pyramid Pooling and other operations,CBAM Attention Mechanism modules are added to improve the extraction accuracy and efficiency of the model.The results show that the training speed of improved DeeplabV3+model is improved by 29.2%,the recognition accuracy is improved by 0.04%,the similarity is improved by 0.68%,and the recall rate is improved by 0.36%.
关键词
语义分割/手势识别/深度学习/DeepLabV3+模型Key words
semantic segmentation/gesture recognition/Deep Learning/DeepLabV3+model引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据