Research on deep learning video game simulation algorithm based on western wind music

Abstract

Deep learning technology and graphics computing technology have made important progress in recent years, making video game technology develop. Western orchestral music based on video game technology has been significantly improved in both human–computer experience and game fun. Therefore, an optimized reinforcement learning (Deep Q Networks, DQN) video game simulation algorithm is proposed. The first is to optimize the network activation function, integrate the ReLU and Softlus functions to design a new piecewise activation function, and use it as a fully connected function of the network. Secondly, the Gabor filter is improved and replaces the traditional filter of the network. Finally, the improved filter and the game image are used to obtain the directional features, and the dimensionality reduction of the features is realized by the kernel principal component analysis (PCA) method, the network training parameters are obtained, and the game simulation is completed. In order to verify the improved DQN game simulation model, two games were selected to participate in the experiment. In the game test of the orchestra simulator, the proposed ReLU-Softplus is the best in the orchestra experience, with the highest score of 126, while the average score of ReLU, Sigmaid, Tanh and ReLU-Softplus are 59.6, 57.2, 52.6 and 64.7. Compared with the other three functions, ReLU-Softplus has a 13.2% improvement. It can be seen that ReLU-Softplus has a better experience of wind games. We still take the pipe game as the test object, adopt the improved Gabor and ReLU-Softplus, and use KPCA to reduce the dimension of the game data features. When the curvature coefficient is 0.3, the DQN model has the highest score in the game test, with a score of 159.6, and the best average score is 93.5, which is better than 73.6 without the improved Gabor. It can be seen that the improved DQN game simulation algorithm can significantly optimize the game scoring effect. The research content has important reference value for improving the experience of western wind music video games and improving the game scoring mechanism.