Integrated neural network and machine vision approach for intelligent state identification

An interfacing of neural networks (NNs) and machine vision to provide the next state of a system given an image of the present state of the system is presented. This interfacing is applied to a loading operation. First, a NN is trained for part recognition under conditions of rotation, location, object distortion, and background noise given an image of the part. Then, a second NN, given the output of the first NN and an image of a pallet being loaded, is trained for optimal part loading onto the pallet under conditions of noise in the image. The paradigm used is backpropagation. It was found that backpropagation performed well in this present state/next state identification. It was able to successfully train both networks. The various training styles used in both NN are examined.<>