The Application Of Deep Learning In Handwritten Numeral Recognition

In the case of pattern recognition,handwritten numeral recognition is an important research topic in pattern recognition,and it has a very wide application in today’s information society.However,the research on numeral recognition is still in the development stage,and the recognition effect is not ideal.Therefore,the research of handwritten numeral recognition is of great practical significance.Now,in many applications the deep learning have achieved relatively good results,recent research results show that deep learning is effective in the field of machine learning and artificial intelligence,especially in the field of image recognition and speech recognition.Therefore,the application of deep learning in handwritten numeral recognition is of great practical significance.This paper is based on stack auto-encoder in deep learning,with self-learning method of initialization parameters,and then gradually tuning to solve the learning problem of the deep networks.A handwritten numeral recognition classification system with high precision and supervised depth feature fusion is proposed.The influence of different feature fusion techniques of sample data on the accuracy of handwritten numeral recognition classification model is discussed.The MNIST features,PCA features and HOG features are effectively combined together,thenextracts twice with the deep auto-encoder networks(stack auto-encoder)and implements the model classification.The experimental results show that the model classification results after merging features and twice extraction are better than only using artificial feature extraction method as the input of sparse autoencoder networks,increase the accuracy of model recognition,prove the feasibility of feature selection in deep learning model application,further explain the deep learning effectiveness in handwritten numeral recognition applications.