CAJ | 학술논문

针对标准BP神经网络易陷入局部极小值的问题，本文结合全局随机搜索最优解的粒子群优化算法，建立了一种3D动漫造型评价模型，并将其应用到3D动漫造型的生成过程。该模型充分利用粒子群算法的全局寻优特性，优化BP网络的权值和阈值，使网络的均方误差小于或等于目标设定值。实验结果表明，本文方法在保证BP网络能收敛到全局最优解的前提下，加快了BP网络的收敛速度和收敛精度，并在3D动漫造型的进化中具有较好的评价性能，提高了造型的生成质量。
침대표준BP신경망락역함입국부겁소치적문제，본문결합전국수궤수색최우해적입자군우화산법，건립료일충3D동만조형평개모형，병장기응용도3D동만조형적생성과정。해모형충분이용입자군산법적전국심우특성，우화BP망락적권치화역치，사망락적균방오차소우혹등우목표설정치。실험결과표명，본문방법재보증BP망락능수렴도전국최우해적전제하，가쾌료BP망락적수렴속도화수렴정도，병재3D동만조형적진화중구유교호적평개성능，제고료조형적생성질량。
This paper constructs a 3D animation modeling evaluation model with Particle Swarm Optimization (PSO) algorithm and BP network in view of the issues of easy falling of standard BP neural network into local minimum and the global searching of PSO. We apply the model to the generation of 3D animation modeling. It fully utilizes the characteristic of global searching of PSO and optimizes the weights and thresholds of BP network, which makes mean-square error less than or equal to the preset value. Experimental results show that the approach improves the convergence rate and convergence precision of BP network based on the guarantee of the global optimization result. It has preferable evaluation capability in the evolution of 3D animation modelings and improves the quality of 3D animation modelings.