Freeform Surface Reconstruction Based On Neural Network

Freeform surfaces are widely used in many areas, many effective algorithms and application technology were proposed. In technology of freeform surface rebuilding, the ultimate objective is to obtain the computer model of it. Using BP neural networks to rebuild freeform surface was proposed by P. Gu and X. Yan in 1995, and they got slick surfaces. Since then, BP neural network based freeform surface rebuilding becomes hotspot of research. In this paper, a method of adaptive Sigmoid function is applied in allusion to shortcomings of BP arithmetic. Using conjugate gradient with measures of restarting in special condition to adjust weights and the shape of the Sigmoid function at the same time, and it makes the curve of convergence break away from flat area rapidly, and accelerate the convergence. The advanced arithmetic is used in freeform rebuilding and the result of simulation shows that the neural network can rebuild the freeform in higher accuracy. The improved arithmetic makes considerable progress both in speed of convergence and the accuracy.