摘要
为了解决传统啦啦操动作风格评价因过度依赖裁判主观判断而导致的不准确性,文章提出了一种基于感知学习的啦啦操动作风格识别与评价方法。研究对象为正负样本不均衡的啦啦操初学者,他们被随机分为实验组与对照组。同时,采用基于PU-CNN和循环神经网络的模式识别算法对标准姿势输入数据进行特征提取,并建立三种感知学习类型的识别评价模型,从而验证PU-CNN和循环神经网络的模式识别算法的有效性。在感知学习过程中,基于PU-CNN和循环神经网络的模式识别算法与专家评委的打分相近;与DT算法、P-CNN算法及LRCN算法相比,该算法的识别准确率最高,达97.6%,GPU运行时间最短,为2.2小时。
In order to address the inaccuracies in traditional cheerleading style evaluation stemming from an over-reliance on referee′subjective judgments,this paper proposes a method of recognizing and evaluating cheerleading styles based on Perceptron Learning.Cheerleading beginners with uneven positive and negative samples are randomly divided into experimental group and control group.A pattern recognition algorithm based on PU-CNN and recurrent neural network has been employed to extract features from the standard posture input data,and three types of Perceptron Learning style recognition evaluation models have been established to verify the effectiveness of the style recognition algorithm based on PU-CNN and recurrent neural network.In the Perceptron Learning,the scores from the pattern recognition algorithm closely align with those from expert judges.Compared to DT algorithm,P-CNN algorithm and LRCN algorithm,this algorithm achieves the highest recognition accuracy rate of 97.6%and the shortest GPU runtime of 2.2 hours.
作者
徐静
XU Jing(College of Physical Education,Huainan Normal University,Huainan 232001,China)
出处
《景德镇学院学报》
2024年第3期48-52,共5页
Journal of JingDeZhen University
基金
淮南师范学院校级重点项目(2022XJZD012)。
关键词
啦啦操
动作风格识别
感知学习
计算机视觉
机器学习
cheerleading
action style recognition
Perceptron Learning
computer vision
machine learning
