Neural Network Models Performance Analysis of Large-Scale Text Recognition∗

Abstract

The continuous development of computer technology leads to booming image data and throws a tricky question to scholars about how to process these data intelligently. Luckily, it is a dream come true to the recognition of images with the help of progressive deep-learning technology. Nowadays, image recognition based on neural networks is widely used, and recognizing a large scale of text information is one of the critical applications. Therefore, this paper will first review the development history of image recognition technology and introduce the concept of the convolutional neural network model. After that, it will analyze the performance of multiple algorithms in recognizing a large amount of text information based on Reginal Convolutional Neural Network, Spatial Pyramid Pooling, Fast Region Convolutional Neural Network, and Faster Convolutional Neural Network. Last but not least, it also points out the prospect of the future development direction of the current image processing technology and its defections. Analysis shows that the biggest drawback of deep learning technology is its dependence on training data. More specifically, when the training data is incomplete, it will be hard for the network model to maintain its recognition accuracy, especially in large-scale text recognition. To further improve the image recognition technology, we should put the effort into constructing a deep neural network model, optimize the training data, reduce the model training parameters, and improve the model accuracy.