Research And Implementation Of Logo Identification Technology For Terrorism Data On The Network

With the spread of harmful data on the Internet,it is an essential issue for the whole society how to effectively detect such harmful data and maintain the health and safety of the network.The detection framework with image processing technology is gradually formed through the research and analysis of harmful data.The research of this subject is based on the“Internet View Content Detection and Analysis Subsystem--GA”project of the 30^th Institute of China Electronics Technology Group,Chengdu 30Kaitian Compay.The project requires to extract the acquired images and video into the visual feature library,and then use the visual feature matching technology of feature library matching and deep learning to match,identify and analyze the images,videos and document content pushed by the content platform.Screen out the harmful content in the massive information,and provide effective clues for the subsystems of relevant departments for further correlation analysis.This article focuses on how to use existing image processing techniques to quickly identify the fearful logo in harmful data.In the Logo recognition module,the image classification model plays an important role,and support vector machine?SVM?is the most widely used classifier in image classification applications.Therefore,this thesis first analyze the SVM image classification model based on the traditional bag-of-words?BOW?model,and concludes the two defects of the traditional bag-of-words model:The characterization accuracy of visual words is not high and The flexibility is poor.This thesis improves the traditional bag-of-words model and the discriminative power of the bag-od-words model by combining with the contrast words and Hamming embedding technology.The latter bag-of-words model has been integrated into the SVM image classification model,and offer a SVM image classification model based on high discriminative bag-of-words model.The test results show that the classification accuracy and applicability are greatly improved.The traditional template matching model has the defects of high false alarm rate and slow recognition rate for the feared logo data.In this thesis,we combine the SVM image classification model based on the high discriminative bag-of-words model with the template matching model,which can provide a new template matching model scheme,and implementing the Logo recognition module.Finally,the module performance test and system test are performed on the above-mentioned Logo recognition module.The module performance test results show that:?1?The matching rate is basically not affected by the size of the template library,and the single recognition time is stable within 0.6s,which better solves the problem of slow template matching?2?The data being matched by template has a better classification effect through the SVM decision mechanism,which causes the false positive rate of non-terrorism data is maintained within 0.1%.According to the system test,the Logo recognition module given in this thesis can stably realize various requirements of the system.