A live pornographic image detection method based on semi-supervised learning in dense antagonistic networks

Abstract

The invention provides a live broadcast pornographic image detection method based on dense antagonistic network semi-supervised learning. Each layer is directly connected with the input live picture and loss function to improve the network density and reduce the phenomenon of gradient disappearance. This method enhances the expression ability of model features and improves the recognition accuracyof live broadcast pictures by generating a countermeasure network identification model. A semi-supervised learning model based on antagonistic network is constructed to learn some tagged data offline. By fitting the generated space of live images, the ability of identifying pornographic images with limited tagged information is improved to the maximum extent. The invention designs a circular progressive training database construction method, which greatly reduces the labor cost and improves the practical application value of the invention. The semi-supervised learning method designed by the invention not only reduces the intensity of manual labeling of training data to a certain extent, but also effectively improves the detection accuracy of pornographic images.

Description

technical field [0001] The invention relates to the processing and analysis of multimedia big data in the field of computer vision, in particular to a method for detecting live pornographic images based on semi-supervised learning of dense confrontation network, which belongs to the field of machine learning and machine vision. Background technique [0002] As a new channel for information dissemination, the current webcast platform has become a brand-new social media. Its characteristics of real-time, rapidity and no geographical restrictions are popular among the public, but this also brings new problems. The National Public Information Network Security Supervision stipulates that bad behaviors such as tattoos, pornography, vulgarity, violence, and dating are prohibited on live webcasts. Special rectification work will be carried out to strengthen the standardized management of the webcast platform. How to effectively manage webcasting so that it can disseminate informati...

AI Technical Summary

Problems solved by technology

[0006] In order to solve the problems of a large number of labeled samples, long training time and low classification accuracy existing in the existing image classification technology, the present invention provides a live pornographic image detection method based on dense adversarial network semi-supervised learning

Images