改进金字塔卷积的体育动作多模态融合识别方法  

Improved Pyramid Convolutional Multimodal Fusion Recognition Method for Sports Actions

作　　者：程小虎[1] 林娟 CHENG Xiaohu;LIN Juan(School of Public and Basic,Anhui Post and Telecommunication College,Hefei 230031,Anhui,China;Feixi County Shangpai Junior High School,Hefei 231200,Anhui,China)

机构地区：[1]安徽邮电职业技术学院公共与基础学院,安徽合肥230031 [2]肥西县上派初级中学,安徽合肥231200

出　　处：《山西师范大学学报(自然科学版)》2025年第1期43-50,共8页Journal of Shanxi Normal University(Natural Science Edition)

基　　金：国家社科基金青年项目(21CTY009);安徽省高等学校省级质量工程重点项目(2021jyxm0768);安徽省高校科研人文社科重点项目(2024AH052655)。

摘　　要：针对以单模态数据难以实现对细粒度体育动作的识别,缺乏对于全局特征的表示能力,提出了一种改进金字塔卷积多模态融合的体育动作识别方法,将RGB和骨架多个模态结合,构建了骨架RGB模态特征融合网络.首先利用图卷积模型构造Focused ST-ROI特征图,然后送入金字塔卷积特征提取,同时提取深层和浅层特征,利用多层特征聚合模块进行多模态特征融合,让模型充分利用两个模态的特征进行最终决策.验证实验结果表明,提出的模型有效提升了体育动作识别的准确率,给体育动作指导提供了新的参考方法.In response to the difficulty in identifying fine-grained sports actions using single modal data and the lack of global feature representation ability,an improved pyramid paper multimodal fusion sports action recognition method is proposed,which combines RGB and multiple modalities of the skeleton to construct a skeleton RGB modal feature fusion network.Firstly,a graph convolutional model is used to construct a Focused ST-ROI feature map,which is then fed into a pyramid convolutional feature extraction system to simultaneously extract deep and shallow features.A multi-layered feature aggregation module is used for multimodal feature fusion,allowing the model to fully utilize the features of the two modalities for final decision-making.The validation experimental results indicate that the proposed model effectively improves the accuracy of sports action recognition and provides a new reference method for sports action guidance.

关 键 词：体育动作识别 多模态融合 金字塔卷积 特征信息 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]