Implementation of Artificial Neural Networks with Reverse Propagation Algorithms for Recognizing Number Patterns

Abstract

The number recognition identical with some research related to pattern recognition, especially the use of classification techniques Artificial Neural Network (ANN). This research aims to design application of number recognition using Artificial Neural Network (ANN) with back propagation  algorithm. To get the best pattern recognition, be done some of stages. First, collected of data by taking the characteristics of numeric characters. Second, done segmentation  on each character with a size of 10 x 8 pixels, each image segmentation then converted into a black and white image format and represented in binary form 0-1, the binary value of each character  formed into a vector as an input to the process of the ANN . Third, prepared ANN architecture for training and testing process. The amount of data used in this research were 1400 data, the 75% is used as training data and 30% is used as a data testing. The results obtained from the introduction of ANN models with up to 80 layers of inputs, 30 hidden layers and 11 output layers, the model was able to produce validation accuracy of 99.49% and 97.62% testing.