Detection Technology And Implementation Of Sensitive Elements Of Public Opinion Image Based On Instance Segmentation

The complex and changeable domestic and international situations have multiplied the pressure on public opinion work,and the rapid development of the Internet has brought new challenges to public opinion work.From the perspective of public opinion,it is very important to study the information contained in the image and identify the information sensitivity transmitted by the image.Images in social networks are complex and diverse,and the detection technology of public opinion elements in the field of public opinion is still immature.Therefore,the image noise of social network image is reduced,the social network public opinion image detection data set is constructed,and the existing image detection technology is combined to design and implement the public opinion element detection algorithm of social network image to ensure the security of social network.Aiming at the problem of large image noise in social networks and the lack of public opinion data set for instance segmentation.Design and implement the social network image noise reduction algorithm,and construct the social network public opinion image detection data set.Combining the coder-decoder idea of U-Net network and using the attention mechanism,the social network image denoising algorithm can improve the average precision by 0.2%in the application of instance segmentation.Aiming at the blurring problem of object boundary segmentation caused by low-resolution mask in instance segmentation technology.Improve and design the public opinion element detection algorithm based on instance segmentation,add a mask prediction branch on the target detection framework FCOS to complete the instance segmentation task,and improve the accuracy of the public opinion element detection model.Compared with SOLOV2,one of the most excellent instance segmentation algorithms at present,the average precision of the public opinion element detection algorithm model based on instance segmentation is improved by 0.1%.Aiming at the high cost of computing resources for the public opinion element detection model.Improve and design the public opinion element detection algorithm of social network image,use knowledge review and residual learning methods to train the model,and improve the average accuracy of 0.3%of the low complexity public opinion element detection model by learning the high complexity public opinion element detection model.The total time for image prediction results has been reduced by 37.91 seconds,and the prediction time for a single image has been reduced by 0.3 seconds.Finally,a social network image public opinion element detection system is designed and implemented.Functional testing was conducted on each module of the system,and the test results proved the usability of the system.The public opinion management department can quickly and effectively view the public opinion image through the visual system page,which is of great significance for guiding the trend of network public opinion and creating a good network public opinion environment.