Physical Education Department, Shanghai University of Finance and Economics, Shanghai 200433, China.
Comput Intell Neurosci. 2022 Apr 27;2022:4921846. doi: 10.1155/2022/4921846. eCollection 2022.
In practice, PE teaching evaluation based on probabilistic convolutional neural network still faces some practical problems. At present, the existing research mainly focuses on how to improve the accuracy of PE (physical education) teaching evaluation, but ignores the balance between accuracy and speed of the model, which is the key to achieve efficient PE teaching estimation. Aiming at the problem of optimization contradiction existing in the traditional probabilistic stochastic convolution neural network regression method, a position adaptive probabilistic stochastic convolution neural network regression method was proposed. Firstly, the basic principle of probabilistic and random convolution neural network regression method is given. Secondly, the contradiction and reasons between hot trial regression and coordinated regression are analyzed. It is found that the process heat will return to the optimization with irreconcilable contradiction with the coordinates due to the lack of learning parameters when the hot trial transforms the coordinates. The optimization contradiction will make the model tonot obtain the exact coordinates of the nodes. Then, based on the above analysis, the learnable parameters are introduced into the Softmax function, and the position adaptive Softmax model is proposed. Combining the model with the probabilistic stochastic convolution neural network regression method, the position adaptive probabilistic stochastic convolution neural network integral regression method is obtained. In order to reduce the training cost of this method, a simplified training strategy is proposed. Finally, the simulation software MATLAB is used for verification, and the functions of sample maintenance, probabilistic stochastic convolution neural network training, and neural network evaluation are realized. The experimental data show that the probabilistic stochastic convolutional neural network is feasible for teaching quality evaluation, meets the accuracy requirements, and indeed provides a convenient and practical tool for PE teaching quality evaluation.
在实践中，基于概率卷积神经网络的体育教学评估仍然面临一些实际问题。目前，现有研究主要集中在如何提高体育教学评估的准确性，但忽略了模型的准确性和速度之间的平衡，这是实现高效体育教学估计的关键。针对传统概率随机卷积神经网络回归方法中存在的优化矛盾问题，提出了一种位置自适应概率随机卷积神经网络回归方法。首先，给出了概率随机卷积神经网络回归方法的基本原理。其次，分析了热点试验回归与协调回归之间的矛盾及其原因。研究发现，由于热点试验在转换坐标时缺乏学习参数，过程热将由于与坐标不可调和的矛盾而返回到优化中。优化矛盾会使模型无法获得节点的确切坐标。然后，基于上述分析，将可学习参数引入到 Softmax 函数中，提出了位置自适应 Softmax 模型。将该模型与概率随机卷积神经网络回归方法相结合，得到了位置自适应概率随机卷积神经网络积分回归方法。为了降低该方法的训练成本，提出了一种简化的训练策略。最后，使用仿真软件 MATLAB 进行验证，实现了样本维护、概率随机卷积神经网络训练和神经网络评估等功能。实验数据表明，概率随机卷积神经网络可用于教学质量评估，满足精度要求，确实为体育教学质量评估提供了方便实用的工具。
