Deep learning-based human action recognition method

Abstract

The invention relates to the field of captcha, in particular to a deep learning-based human action recognition method. According to the method, a neural network comprising an LSTM is adopted to realize the judgement of the current page operation main body; construction of machine action samples is based on human action samples; 4 generation manners which comprise random generation are adopted; positive sample tracks are randomly segmented into a plurality of sub-segments and the segmented sub-segments are randomly spliced; scaling and disturbance are carried out on track parameters on the basis of positive samples; and such a negative sample generation manner is based on the positive samples, so that the negative samples have higher similarity with the positive samples, the obfuscation is stronger, the neural network trained by training samples has higher distinguishing ability. According to the method, the recognition of the current page operation main body on the basis of an LSTM network is firstly realized; and the method is particularly suitable for the business scenes of realizing verification of operation main bodies during graph dragging and judging the operation main bodies.

Description

technical field [0001] The invention relates to the captcha field, in particular to a human behavior recognition method based on deep learning. Background technique [0002] Nowadays, network technology is more and more developed, and there are more and more network applications, such as various websites, emails, blogs, e-government websites, etc., have become the necessities of our daily life. However, with the rapid development of the Internet, network security has become an increasingly prominent issue. In particular, network security attacks such as automatic registration and login of malicious programs, malicious flooding, and brute force cracking of accounts and passwords with specific programs. In order to avoid these occurrences, it is necessary to identify whether it is a person or a program that is currently registering or accessing the web page. The most common captcha (abbreviation for Completely Automated Public Turing Test to Tell Computers and Humans Apart) ...

AI Technical Summary

Problems solved by technology

The use of captcha network security defense methods based on human behavior has been widely used. When using such methods for network protection, it is first necessary to have the ability to recognize human behavior or machine behavior. However, based on traditional machine learning methods, human behavior The judgment accuracy of extracted features and reclassification is not high

Because some characteristics of human behavior are deep-level characteristics, it is difficult to extract such characteristics through artificial rules.

Images