| Nowadays,with the rapid development of computer and network technology,various related scientific and technological applications are increasingly penetrating into all aspects of people's lives.Safe and rapid information dissemination greatly enriches people's spare time life.Internet communication technology brings various social networks and self-Media to become important means for people to obtain information and get entertained.But some junk information is becoming more and more rampant.Tibetan independence activists take advantage of the abundant and convenient channels of information dissemination on the Internet to spread the reactionary information of national division on the Internet,affecting social stability and endangering the security of cyberspace.Detection of flags in pictures is the main content of this paper.On the basis of consulting relevant literature and laboratory work at home and abroad,this paper improves the related algorithms of Tibetan independence flag detection before,aiming at improving efficiency and performance.The main work is as follows: 1.Reconstructing an image data set of Tibetan independence banner,which contains banners of different sizes,partial occlusion,distortion and illumination conditions;2.Using Dilated Convolution and Depthwise Separable Convolution,the convolution neural network for detection is improved,and the previous 1x1 convolution is used.The convergence effect and network performance of different optimization algorithms are analyzed.3.Selective Search is used to nominate the image area and an improved region fusion method based on NMS is proposed for candidate regions merging and target labeling,which reduces the amount of computation in order to improve theperformance.Finally,by analyzing the experimental results,this paper further verifies that the improved convolutional neural network model proposed in this paper can play a great role in the Tibetan independence banner detection task. |
PDF Full Text Request