Typical character identity identification method based on depth learning

Abstract

The invention relates to a typical character identity identification method based on depth learning, and relates to a depth learning method. The method automatically judges the identity of a typical character from an image video. The method comprises the steps that a dataset with annotation information is constructed; the dataset with the annotation information is input into an neural network generated in an acceleration area to acquire a model framework, and model training is carried out to acquire a typical character identity identification neural network model; and the trained typical character identity identification neural network model is used to identify and detect the identity of the character. The typical character identity is successfully identified and compared. The speed and accuracy of character identity identification can be improved. The method can greatly promote jail management and traffic management.

Description

technical field [0001] The present invention relates to the field of character recognition, in particular to a deep learning-based typical character recognition method. Background technique [0002] The technical background related to the present invention will be described below, but these descriptions do not necessarily constitute the prior art of the present invention. [0003] Traditional machine learning generally goes through raw data to feature extractors to classifiers or detectors, and finally obtains results; while deep learning does not require manual design of feature extractors, the machine itself can learn automatically, which is very suitable for changes Diverse and diverse natural data, with very good generalization ability and robustness. [0004] At present, there are many kinds of algorithms that can realize the identification of typical people. The more traditional algorithm is to use selection search to select the candidate area first, then extract its ...

AI Technical Summary

Problems solved by technology

[0004] At present, there are many kinds of algorithms that can realize the identification of typical people. The more traditional algorithm is to use selection search to select the candidate area first, then extract its features, and finally use classifiers such as support vector machines to classify. Get some possible targets; but this traditional algorithm has a lot of disadvantages, one is because the selection search is very slow, which will consume a long time and increase the time cost; the other is because of classification such as support vector machines The classification result of the machine is not ideal, the diversity is poor and the robustness is not high

[0005] The neural network that appears today has strong feature extraction characteristics, so the target detection and person recognition has been put on the neural network to do it, and the subsequent fast regional convolutional neural network appeared to identify the identity of the typical person; but its recognition The speed, accuracy is still not high

Images