Mathematical Modeling based on Neural Network Learning for Object Recognition in Automated Systems

This paper aims to identify efficient methods of mathematically modeling an automated physical system using a neural network. Based on the Levenberg-Marquardt method, we built a feed-forward neural network with the capabilities of a graphics accelerator. The model also sums up and suggests a new neural network training algorithm with Bayes regularization, Nguyen-Widrow initialization, and the early stopping and control method. This greatly expands the efficiency of solving problems where knowledge of an automation system is usable.