Research And Application Of Water Book Image Recognition Algorithm Based On Deep Learning

Language is one of the spiritual civilizations of a nation.The text recognition technology plays an important role in promoting the dissemination of text and cultural inheritance.In Guizhou Province,China,there is a kind of text,known as Shui,which is an ancient type of hieroglyph.Due to the scarcity of inheritors of Shui characters,the inheritance of Shui characters is interrupted,thus the digital protection of Shui characters is urgently needed.The traditional text recognition technologies are mostly based on manually extracted features,which involves a lot of expertise and human labor,and the recognition result is not satisfactory.With the advent of deep learning techniques,many researchers have tried to apply their power to text recognition.However,there are several challenges confronting us when using deep learning for Shui characters recognition.As far as Shui character recognition is concerned,first,there is no complete dataset about Shui characters.Because there are very few professionals to collect and process Shui characters.Second,many models and algorithms have been proposed,but most are designed for specific problems.And few researchers have tried to apply deep learning methods to Shui character recognition.Finally,the hyper-parameter settings during deep learning model training process greatly affect the performance of the model,though much effort has been made to optimize the hyperparameters of deep learning models.In view of the above problems,this paper studies the deep neural network model.Based on the deep convolutional neural network(CNN),using a new hyper-parameter optimization algorithm,this paper implements Shui character recognition.The specific work is as follows:1.A hyper-parameter optimization algorithm based on population evolution is proposed.Drawing inspiration from evolutionary algorithms and combining the advantages of grid search and manual configuration,this paper proposes a hyper-parameter optimization algorithm based on population evolution.The algorithm uses individual evolution,group selection,asynchronous parallel iterative training to achieve population evolution results.Compared with the grid search method,the experimental results demonstrate the effectiveness of the proposed algorithm.2.A novel architecture based on CNN is designed.This paper analyzes the composition of CNN structure in detail and discusses the selection problems of loss function and activation function.In this paper,a novel CNN architecture is proposed which has 11 layers,and a hyper-parameter optimization algorithm is adopted for training which is based on population evolution proposed.The proposed model is tested on the Shui character dataset,and the effectiveness of the model is verified by the experiment.3.A completely new dataset of Shui characters is established.The dataset provides experimental data for the study of deep learning.The paper introduces the characteristics of Shui characters and presents a detailed overview of data collection,data preprocessing,data annotation,and the process of establishing the Shui character dataset.Moreover,a Shui character recognition prototype system is designed to integrate data processing and text recognition functions,which provides convenience for the study and dissemination of Shui characters.In summary,the hyper-parameter optimization algorithm based on population evolution proposed in this paper provides a reference method for neural network hyper-parameter setting.The study of Shui character recognition can be used as an empirical case for minority language protection and text recognition,which provides a foundation for other researchers to study.