Development and Performance Assessment of Bio-inspired based ANN Model for Handwritten English Numeral Recognition

In many application areas such as online sorting of postal address and pin codes, automatic recognition of name plates of vehicles recognition of the handwritten numerals plays an important role. The performance of recognition depends on the types of models and the training scheme employed. In this paper, the problem of classification of digits has been addressed and hybrid classifiers have been developed for this purpose. The basic model structure is used for artificial neural network (ANN) but its associated weights are updated using genetic algorithm (GA) and particle swarm optimization (PSO) algorithm. The bio-inspired based weights adjustment scheme aims to obtain optimal weights as compared to gradient descent-based learning such as backpropagation as well as adaptive moment optimization. The proposed models have been simulated using MNIST (Modified National Institute of Standards and Technology) dataset and loss and accuracy values have been evaluated. The comparison of the results reveals that the GA and PSO based ANN models outperforms the corresponding stochastic gradient descent and adam optimizer. It is further noticed that the GA based ANN classifier exhibits the best performance compared to other three models in terms of lowest loss during training and highest accuracy during validation. The proposed method can be extended for recognition of numerals and characters of other languages.