Recognising electronic symbols using neural networks

Neural networks are being implemented for the recognition of electronic symbols with the objective of carrying out a comparative study with the performance of other techniques including template matching, chain vectors and Hough transform. Whilst the error back propagation is commonly used for training, it is a well known fact that an very large number of iterations are required before the network is properly trained. Another disadvantage of this method is that it is highly susceptible to being trapped in a local minimum of the error hypersurface. Various methods of training will be investigated. The neural network techniques will be integrated with a Circuit Diagram Interpreter (CDI) for the understanding of scanned schematics. The symbols will be in one of eight orientations, and further investigations will be carried out on the effects of the pose of the symbol with respect to the performance of the recognition module