Method and equipment for optimizing neural network text recognition model

Abstract

The invention provides a method and device for optimizing a neural network text recognition model. Compared with the prior art, the method comprises the steps: obtaining a first image training set tobe subjected to text recognition; training a neural network text recognition model based on the first image training set to be subjected to text recognition; performing text recognition on the first image training set until a text recognition result of the first image training set meets a preset loss threshold; performing data augmentation based on the first image training set; determining a second image training set after the first image training set is subjected to data augmentation, inputting the second image training set into the improved neural network text recognition model, wherein theimproved neural network text recognition model comprises a language model used for performing result adjustment on the neural network text recognition model, so a text recognition output result of theneural network recognition model is optimized through the language model. In this way, the text recognition accuracy of the neural network can be improved.

Description

technical field [0001] The present application relates to the field of computer technology, in particular to a technique for optimizing a neural network text recognition model. Background technique [0002] At present, the performance of the neural network is highly dependent on the training data. When the training data is more diverse and the data volume is larger, the performance of the obtained model is often better. In the existing technology, before the daily training model, the image Do augmentation to increase the diversity of pictures. However, if the operation is not done properly, the quality of the pictures may deteriorate, affecting the performance of the model, resulting in low accuracy of model recognition results. Contents of the invention [0003] The purpose of this application is to provide a method and device for optimizing a neural network text recognition model. [0004] According to one aspect of the present application, a method for optimizing a neu...

AI Technical Summary

Problems solved by technology

[0002] At present, the performance of the neural network is highly dependent on the training data. When the training data is more diverse and the data volume is larger, the performance of the obtained model is often better. In the existing technology, before the daily training model, the image Do augmentation to increase the diversity of pictures. However, improper operation may lead to poor quality of pictures, affect model performance, and result in low accuracy of model recognition results.

Images