Research On The Application Of Violent Terrorism Video Recognition Technology Based On Content

Anti terrorist audio and video work is an important part of network anti-terrorism,but with the continuous development of Internet technology and the continuous expansion of scale,the original work mode has been difficult to adapt to the current work requirements,there are low work efficiency,high human consumption,not in-depth work and other problems.In the current historical period,using the advantages of artificial intelligence technology to promote the transformation of anti terrorist audio and video work is both necessary and operational.In view of the prominent problems of inaccurate training set and unclear application scenarios in current academic research,this paper will start from the concept of top-level violence and terrorism video,discuss and summarize the content characteristics of violence and terrorism video,and use the violence and terrorism video file identified by judicial practice as the training set of video recognition technology;combining with relevant laws and regulations,it puts forward that each legal subject plays an important role in the fight against violence and terrorism audio and video Responsibilities and obligations to be undertaken,clear application scenarios and business requirements of anti terrorist audio and video work,and put forward suggestions for improving anti terrorist audio and video work.According to the video key frame extraction,image feature detection and other key technologies involved in the video recognition algorithm,this paper proposes the key frame extraction algorithm based on the average value of frame difference for the content feature of the video,and proposes the improved algorithm based on the loss function of YOLOv3 for the location and size feature of the logo of the video.(1)In this paper,we propose a video recognition method for different legal subjects.It is the first step of technology application to make clear the application scenarios and requirements.Only by making clear what is violent video and the current situation and problems of fighting against violent video,can we truly understand the implementation direction of content-based violent video recognition technology and achieve a targeted goal.In view of the common problems in academic research,such as inaccurate training sets and unclear application scenarios,this paper studies the definition and performance of terrorism,cyber terrorism and violent audio-video respectively,summarizes the content and characteristics of violent audio-video in combination with judicial practice,and summarizes the specific needs of relevant subjects in the fight against violent audio-video in accordance with relevant laws Finally,a video recognition method for violent terrorism is proposed,which is suitable for different legal subjects.(2)A key frame extraction algorithm based on the average value of frame difference is proposed.According to the content characteristics of violent terrorism video,the key frame is extracted by frame difference method,and a special key frame extraction algorithm based on the average value of frame difference is proposed,which can not only effectively reduce the detection consumption of violent terrorism features,improve the detection efficiency,but also take the key frame image extracted from violent terrorism video as the training sample Due to the lack of training samples,the experimental data prove that it can extract the key shots of the video efficiently and accurately.(3)An improved algorithm of YOLOv3 model based on loss function is proposed.This paper analyzes the influence of SSE,cross entropy and other loss functions on the training of YOLOv3 model.Aiming at the characteristics of relatively fixed size and location of the logo of the video,an improved algorithm of YOLOv3 model based on the improved loss function is proposed,and a special loss function type is designed for the training of the image of the terrorist.The effectiveness of the algorithm is verified by experiments.