Implementation of Deep learning Methods to Marathi Hand Written Characters and its Pattern Recognition by Using Generative AI

Abstract

In the field of Machine learning, Powerful AI Systems trained and tested to carry out the task have utilized many models in areas including pattern recognition, natural language processing, and computer vision. Deep learning has provided outstanding pattern matching and recognition systems in several domains, including character recognition. In this research we have collected our own published dataset like MHCD_GIETV1 and MHCD_GIETV2. Initially we have taken a  samples of vowels characters like Bara-khadi.We have used deep learning methods of Inception,ResNet,VGG16and leveraged a custom convolutional neural networks that enhance to find the accuracy of Vgg16 i.e. 65.96%,Inception 83.42% and ResNet 65.96%.After comparison we found Inception is better accuracy but we are not happy about our results. We have proposed a customized convolutional neural networks with the increase of hidden layer by extending our epcho and achieving an accuracy of 96%.To extend our research we used Generative AI by using LSTM with GRU and obtained a result in a remarkable patterns matching an accuracy of 98% probability. This research provides valuable resources for future explorations in other languages and related domain such as sentiment analysis and boarder image recognition applications.