Design And Analysis Of Tibetan Recognition System Based On The Improved Algorithm Of Convolution Neural Network

Tibetan language has a long history.The preservation of Tibetan script has important significance for the study and inheritance of Tibetan culture.It also plays an important role in the cultural history of our country.The current Chinese and English word recognition technology has matured,but the Tibetan recognition technology is lagging behind.The Tibetan handwriting recognition input is an important way for the Tibetan computer to automatically recognize the input,overcomes the inadequacy of the keyboard input method,can realize the Tibetan computer's intelligent input,and enriches the human-computer interaction.With the development of mobile terminals and smart terminals,Tibetan handwriting input urgently requires more efficient identification technologies.Therefore,the study of handwritten Tibetan recognition technology not only has social value,but also has a broader market value.The traditional handwriting Tibetan recognition technology still stays in the traditional text recognition technology,including preprocessing,feature extraction and feature matching to further identify Tibetan characters.Handwritten Tibetan characters have many similar characters,and there are many external interference factors such as personal differences,stroke adhesion,and paper quality differences.Traditionally,traditional recognition methods include feature engineering and classification.Feature extraction is based on the character characteristics and writing order of handwritten Tibetan characters,and classifiers are used for classification.However,the overall ability to express Tibetan character features is not Strong,and the ability to resist interference is poor,so the recognition efficiency is not high.In recent years,due to the high classification accuracy and strong learning ability ofneural networks,and the robustness and fault-tolerance of noise data,the recognition rate of bp neural networks in handwritten Tibetan recognition is generally higher than that of traditional methods.As a classifier,the image is preprocessed,the dimension-reduced data is input to the input of the neural network,training is performed by connecting multiple fully connected layers,and finally classified.Since bp neural network training data needs multiple iterations,the network convergence speed is slow,and the recognition efficiency of handwritten Tibetan texts is not high.With the continuous development of machine learning,the more powerful convolutional neural network for image feature extraction provides a new solution for handwritten Tibetan recognition technology.Convolution image acquisition in the recognition process is directly obtained from the original image,reducing the image pre-processing work,and can simultaneously feature extraction and classification.This paper proposes a handwritten Tibetan recognition method based on convolutional neural network.Compared with the traditional Tibetan recognition technology,the feature extraction and the function of the classifier are integrated,which reduces the external factors in the feature extraction process..I collected 900 sets of handwritten Tibetan characters as data sets,which were divided into 34 categories.Five commonly used deep convolutional neural network models were selected for training.Combined with the characteristics of handwritten Tibetan characters,the structure of the model was correspondingly improved and optimized.The amount of parameters was greatly reduced,and the generation of overfitting was reduced.A higher recognition rate.Using the model fusion technique,the full-connected layers under the global average pooling of all single models are stitched,and training is performed while memory is allowed,and the processing results are merged to obtain better handwriting than the single model.Tibetan recognition rate and accuracy.Under the same hardware environment and the same data set,I reproduced the method of using bp neural network to recognize handwritten Tibetan and conducted training.Through the experimental comparison,the Tibetan recognition method based on the improved convolutional neural network algorithm proposed in this paper is not only fast and convenient to train,but also greatly improves the handwritten Tibetan recognition rate,with a recognition rate of99.5%.