Research On Tibetan Flag Detection Algorithm Based On Convolutional Neural Network

With the rapid development of mobile Internet,self-media has become a key social platform in people's lives.This makes information sharing very convenient,but the security issues that come with it are also becoming increasingly prominent.In recent years,overseas independent proponents in Tibet have continuously disseminated illegal information through a new platform such as the self-media.The purpose of this paper is to check out pictures on the Internet platform that contain the Tibetan independence flag,which is essentially a target detection technology.Target detection has always been one of the research hotspots and difficulties in the field of computer vision.Common applications include face detection,pedestrian detection and object detection.The research content of this paper is the banner inspection.Based on the literature review both at home and abroad,combined with the characteristics of the banner detection task,we studied the Tibetan independence flag detection algorithm and obtained certain results.The main work of this article is as follows: Firstly,we have organized and built a picture data set containing Tibetan flags of different scales,styles,occlusion,deformation and illumination.Secondly,we propose a Tibet independence flag recognition algorithm based on convolutional neural network.Data set amplification and addition of the Dropout layer are used to prevent over-fitting;the 1x1 filter structure is designed to replace the full-connected layer to increase the network width and improve the network performance.The effects of various optimization algorithms on the classification effect are analyzed;the final experiment results show the convergence effect and classification performance of the network model.Thirdly,by analyzing the proportion distribution of the Tibetan independence flag in the test sample,we propose a regional nomination rule based on multi scale matching strategy to detect the banner of Tibetan independence,and combine the practical application requirements and shield the detected target flag by using the shielding mechanism.Different from the traditional multi-scale sliding window mechanism,this method needs to recognize 150 nominated regions at most to complete the detection task of any size picture.Eventually,the effectiveness of the detection algorithm is verified by experiments,and the final detection rate is over 90%.Finally,this paper compares the Tibetan independence flag detection algorithm based on the traditional shallow machine learning,and further validates the advantages of the convolutional neural network model in the Tibetan independence flag detection task by analyzing the distribution of the experimental results and the distribution of the positive samples in the training center.