短裤特征截面曲线的径向基函数神经网络模型构建

将着装人台进行三维扫描获取点云数据，截取与人体特征部位相对应的短裤特征截面。将原数据坐标点转化为极坐标系下的极角与极径值后，以极角值作为输入向量，极径值作为输出向量，构建短裤特征截面曲线的径向基函数（RBF）神经网络模型，并与反向传播（BP）神经网络、最小二乘法及三次样条函数的拟合效果进行比较。结果表明，神经网络拟合曲线的平均绝对误差比最小二乘法及三次样条函数方法小，仿真输出曲线和原始数据非常接近，且曲线光滑；RBF网络的训练速度更快，所需训练步数少，拟合效率明显优于BP神经网络。

Construction of radial basis function neural network models for typical cross section curve of shorts

Three-dimensional body scanning technique is used to collect point clouds data from the dressed mannequin and capture the shorts’ typical cross section that is corresponding to the feature points of body. By change the original coordinate point to polar angle and polar radius under the polar coordinate system, and taking the polar angle as the input and the polar radius as the output, RBF neural network model of the shorts’ typical cross section is established. Then the curve of clothing typical cross section is fitted and the fitting effect is compared with that of BP, least square method and cubic splines. The experiment results show that the mean average absolute percentage error of both neural networks is less than that of least square method and cubic splines. The simulation output curve is very close to original data and the curve is smooth. RBF network has much higher training speed, fewer training steps. and fitting efficiency superior to the BP neural network.