Research On Quality Diagnosis Of Video Surveillance Based On Deep Learning

As the application scenarios of the video surveillance system become wider and wider,the scale of integration continues to increase,and the number continues to increase,the video surveillance system requires real-time online,and various failure problems occur in the system link due to various reasons during long-term operation.This affects the quality of the video image.In order to ensure that the front-end system link equipment works normally,video quality diagnosis research is urgently developed,becoming an important way to solve such problems.There are many deficiencies in the use of manual fault detection.Therefore,this paper discusses and researches the video surveillance quality diagnosis algorithm based on deep learning for various types of faults in video surveillance,and provides related technical support for the later monitoring equipment operation and maintenance work.The main work of the paper is as follows:The thesis first analyzes the composition of the monitoring system,studies the actual types of video monitoring failures and their causes,analyzes the current video monitoring quality evaluation methods,classifies and introduces them,and summarizes the commonly used algorithm characteristics and existing defects.In-depth investigation and analysis of video surveillance quality feature extraction and diagnosis index problems,and the appropriate video surveillance quality diagnosis standards are summarized for various video failure type feature extraction.Then,the theory and development of deep learning are introduced.The paper analyzes the importance of deep learning in the diagnosis process,studies the development and changes of CNN networks,and summarizes the advantages of CNN networks in the feature extraction process.After analysis and comparison,this article It adopts VGG network and Goog Le Net network with easy network expansion,simple structure and good performance for analysis.According to the requirements of deep learning on the equipment,the hardware and software environment required for the experiment were built.By expanding the TID2013 database,a more comprehensive data set suitable for deep learning has been collected.After data set preprocessing,the input of the network is obtained,a deep learning neural network framework is built,and the model parameters are adjusted.Conduct network training.Read the trained network model to complete the test.Finally,based on the deep learning video surveillance quality diagnosis model solution,the paper implements the VGG deep learning video surveillance quality diagnosis algorithm,uses it to complete the anomaly detection experiment,and uses Goog Le Net to realize the image diagnosis recognition rate experiment.The results show that the method adopted in this paper can obtain better diagnostic results than traditional algorithms.The method in this paper has the characteristics of high accuracy and strong applicability.Based on the research of video surveillance quality,this paper uses deep learning network to construct an effective video image quality evaluation and diagnosis model.Through thesis work,the expansion of suitable and effective data sets,the work also provides technical support for video quality diagnosis research,video surveillance service optimization and other aspects.