Research And Implementation Of Internet Pornographic Image Recognition Algorithm Based On Deep Neural Network

With the rapid development of the Internet and multimedia technology,the issue of content security on the Internet is increasingly becoming serious.Many pornographic images spread rapidly on the Internet,infringing on the rights and interests of women and children and unconducive to the physical and mental health of teenagers.Even though Deep Learning technology has developed rapidly in the field of image recognition,limitations exist in the current network of pornographic image recognition technology,primarily due to the insufficient classification of pornographic image data sets.Pornographic images are hidden and contain various subjects and forms of expression.Based on the issues mentioned above,the current thesis proposes a pornographic image recognition algorithm based on a deep neural network by exploring the current Internet pornographic images.The main contents and contributions of the current thesis include:(1)Based on the open-source dataset and combined with the actual situation on the Internet,this thesis constructs a dataset of internet pornographic images for pornographic image algorithm research.Finally,an internet pornographic image dataset with 34920 images in the pornographic category,70396 images in the normal category,and 25391 images in the sexy category with 130707 images were collected.Based on the actual situation of the Internet,this dataset introduces more images with similar semantics to pornographic images,constructs sexy classification,and expands the image data under each classification by crawler technology.It specifically improves the expression ability of pornographic image datasets to further improve the classification ability of pornographic image recognition and model generalization ability.(2)A parallel structure depth neural network pornographic image recognition algorithm combining convolutional neural network and transformer model is proposed.In this parallel architecture,the convolution neural network branch outputs the local features,and the transformer model learns the global relationship of the local features.The interaction between the two ensures the interaction between the local features and the global representation.Compared with the previous model based on a deep neural network,this model is more suitable for the diversity of pornographic image data and has better model generalization ability and greater robustness.Based on the experimental test with the internet pornographic image dataset,the algorithm has a high accuracy and recall rate.Through comparative experiments,the proposed algorithm has better performance.Finally,to meet the practical application requirements of pornographic image recognition,an internet pornographic image recognition system is designed and implemented based on the algorithm proposed in the current thesis.The experiments prove the robustness and timeliness of the system.