Research On Anomaly Recognition Of Urban Management Components Based On Deep Learning

Nowadays with the rapid development of economy,the process of urbanization is accelerating.The number of urban population is increasing,and the requirements for urban management are getting higher and higher.To manage cities more effective,most governments are putting cameras everywhere to monitor them.However,due to the increasingly frequent urban activities of people,the urban management component only relying on cameras can no longer satisfy the regulatory requirements.Manual monitoring of cameras is not only time-consuming,but also inefficient.So this article in view of the city management of some common components abnormal do recognition research decides to use the convolution in the deep learning neural network to some common components abnormal do processing.In this way,cameras can automatically identify some city steward parts abnormalities.System also can automatically report case,so that we can improve the efficiency of urban management,convenient urban management personnel working at the same time,greatly reduce labor time.The main work of this paper is as follows:(1)In this paper,five common cases such as well-lid,road damage,small advertisement,road occupation operation and road rubbish were collected through Internet and mobile phone shooting,and a training data set was made,with a total of5660 data samples.By comparing the two most popular identification models YOLOv3(You Only Look Once)and SSD(Single Shot Multi Box Detector),the monitoring model that is more suitable for the data set in this paper is selected.In this paper,through the convergence and convergence values of the model loss function,as well as the recognition accuracy,it is found that the SSD model is more suitable for the data set in this paper.(2)Further study on SSD model shows that the robustness of SSD model is not good enough.Due to the high position of the camera,the collection of some data will cause the target to become smaller in the figure,but the SSD model cannot identify it well.Therefore,this paper proposes a network replacement improvement for the SSD model,replacing the VGG network it depends on with the residual network RESNET-101,so as to avoid the network degradation problem and use a deeper network to train.(3)By comparing the model before and after the improvement,it is found that the residual network can avoid the problem of network degradation and increase the number of network layers.Experimental results show that the improved model is more accurate in identifying small targets.Finally,the improved model is applied to the digital city management system to realize the intelligent city management.