Research On Sensitive Image Recognition Algorithm And System Implementation

Nowadays,people’s lifestyles have become more and more convenient due to the rapid development of network science and technology.At the same time,it has a lso provided lawbreakers with a more concealed and extensive way of propagation,making it easy for them to spread various information on the Internet such rumors and harmful information.These sensitive information are disseminated in the form of text,image and video,and image have become the main form of dissemination due to its have the characteristics of intuitiveness and convenience.In order to block sensitive information,many social network and platform recruit professional appraisers to deal with it.The main feature of manual blocking of sensitive information is its high accuracy,but it will greatly harm the mind and body of the appraiser.How to use more effective technical means to identify and get rid of sensitive image in the network has became an important research topic.Aiming at the problem of insufficient samples in sensitive data sets,a sensitive image recognition algorithm based on Y-sharing network is proposed.The Y-shared network solves the problem that the cross-stitch network uses different data sets for multiple tasks,and the performance of a task model is reduced compared to the single-task model.The multi-task Y-shared network uses Res Net34 as a single-task network model for pornographic image and horror image recognition tasks,and the two single-task network models are connected through Y-shared unit.The Y-shared unit applies the horror image feature and the pornographic image feature by linear combination to generate a new horror image feature to realize feature sharing.The experimental results show that the Y-shared network proposed in this thesis has a significant improvement in model performance compared to the single-task model on the horror image recognition task with insufficient data set samples.The recall rate increased by nearly 8.7%,and f1-score improved by nearly 4.2%.The model performance of the Y-shared network meet the target requirements of the project’s sensitive image recognition task.Based on the above research results,a sensitive image recognition system that can identify pornographic image and horror image is designed and implemented in.The system is mainly divided into three modules,namely the browser plug-in module,the sensitive image recognition module and the background management module.The browser plug-in module is the input module of the system.It is used to obtain all the image addresses of the user’s current web page,forward it to the background sensitive image recognition module,and determine whether to close the user’s current web pa ge according to the response result.The sensitive image recognition module is the core module of the system,which implements the sensitive image recognition algorithm based on the Y-shared network,and mainly recognizes the images uploaded by the plug-in module.The background management module is the main display module for system information.In this module,the administrator can clearly and intuitively overview the historical access status of the system and the classification status of the recognized images,as well as the blacklist collected by the system.