Lightweight Deployment of Bridge Monitoring System Based on Edge Algorithm

Abstract Urban road congestion seriously hinders people's daily work and social development, making bridge infrastructure become the preferred way to solve the problem of jammed traffic. Based on the real‐time automatic detection of bridge traffic flow information by monitoring facilities, dynamic traffic evacuation can alleviate the pressure of bridge commuting and reduce the possibility of traffic accidents. In order to meet the deployment requirements, the embedded system must carry out a lightweight design of the deep learning model to improve the calculation accuracy and logical reasoning ability without reducing the processing quality and performance of the model. In this paper, we design a lightweight neural network model structure to integrate the relevant information of different levels of network connections. Taking into account the shortcomings of traditional data processing, using the characteristics of data fragmentation in the edge computing scene, and according to the multi‐scale external characteristics, we design the IoT model of the bridge monitoring system to improve the performance of the neural network model. We use cutting priority in the design to solve the problem of parameter feature dimension and reduce the large amount of calculation. Through experimental analysis, it is verified that the dynamic convolutional network can replace the conventional convolutional network image capture and processing capabilities. Using the current mainstream algorithm for comparative testing, it is proved that the performance of the proposed model algorithm is improved by 17. 2% and 19. 6%, respectively. The logic operation efficiency of the neural network model can be achieved in the initial stage of the lightweight deployment. Besides, the data integrity and the dynamic sparsity rate are higher than those of the current mainstream model algorithms. It is verified that the ability of network traffic anomaly detection can be increased to characterize the backbone network. It also adds identity mapping between the sub‐modules of the backbone network through residual network to prevent gradient disappearance/explosion, over‐fitting and network degradation, while accelerating the convergence speed of the model. For the normal operation of bridge traffic, it can avoid traffic accidents and congestion, and play the role of intelligent management of traffic and commuting rules. It also improves the operation efficiency, safety and stability of road and bridge facilities.