Anomaly Detection At The Front End Of Image Acquisition Based On Deep Learning

Today,Monitoring video has spread to every corners of our lives.With the large number of monitoring equipments being put into use,how to ensure that the huge surveillance systems are in normal working conditions has become a hot topic of research.Through the analysis and analysis of existing literature data,There are many factors in monitoring the quality of video images.There are human and environmental influences,and in the long-term use of the equipment,various hardware failures may occur.We can analyze the video image to find the fault condition of the equipment and improve the work efficiency of the maintenance personnel.Therefore,it is of practical significance to study and analyze the abnormal quality of the image.In the current monitoring video image quality detection method,Most of them require a large amount of previous image preprocessing such as feature extraction,and are susceptible to image scenes and natural environmental factors.Although abnormal images can be identified to some extent,there are many reasons that affect image quality due to manual intervention.Too many ways to increase The subjectivity of the subject has caused it to have too many limitations.The use of deep learning can effectively avoid these problems.The paper identifies abnormal images by constructing a deep learning model.The main research work includes,Firstly the analysis of basic model of deep learning、the tool caffe framework used and deep learning methods,through the analysis of the collected monitoring video images.There are two kinds of abnormalities of color cast and camera angle,and artificially collected normal images to simulate mosaic,sharpness,Gaussian noise,and salt-and-pepper noise anomaly types,and then classify them,not only adding to the experimental data sample The type of anomaly also increases the influence of complex environmental factors.The experimental data is normalized and the appropriate training set,verification set,and test set are set according to the needs of the experiment,so that a data model is established,followed by a deep structure neural network.Detailed comparative analysis,understanding of its structural principles,selection of appropriate methods and parameter optimization,and finally the image anomaly detection model were tested and the results were analyzed.The accuracy of the test indicates the feasibility of the method,and then with the traditional The machine learning algorithm is implemented The contrastive analysis analyzes the features of the abnormal color image and sharpness image in the experimental data set,and then uses it to train the SVM and BP neural network models.Through analysis of the model’s test results,we can see that the deep structure is convoluted.The neural network model has better accuracy,and is less affected by factors such as lighting and scenes,and it has better adaptability to complex environments and can provide reference for abnormal detection of surveillance video images.