Research And Implementation Of Intelligent Detection Technology For Job Security

At present,there are many production safety accidents in our country,and the safety production situation is very serious.The unsafe behavior of construction personnel is the main factor for the occurrence of safety accidents.Research has proved that timely supervision of unsafe behaviors of construction site workers can not only improve the level of project management,but also reduce the incidence of safety accidents.At present,most of the supervision methods for the unsafe behavior of construction workers rely on manual supervision.This supervision method is not only labor-intensive,but also inefficient and incomplete.Therefore,the development of an intelligent supervision method is very important for safety production management.important.In view of this,this paper proposes a target detection algorithm based on deep learning to judge whether construction workers have unsafe behaviors in job security scenarios,and build a visual operation interface to improve the efficiency of construction site safety management.The research contents of this paper include the following:（1）By investigating the types of unsafe behaviors of workers in the construction scene,the unsafe behavior categories of intelligent detection in this paper are clarified,and the behavior categories are redefined.Finally,the establishment of the unsafe behavior data set of construction workers is completed.（2）This Thesis proposes optimization and improvement strategies for the YOLOv4 target detection algorithm from three aspects:network parameter optimization,multi-scale feature fusion network design,and network structure optimization.First,the a priori box size is re-clustered by the K-Means++algorithm;the multi-scale feature fusion network is used to predict the output of the worker’s unsafe behavior;CSPNet)structure replaces the original cross-stage local network to construct a new CSPDarknet53 network structure,and then uses the densely connected convolutional network（Dense Convolutional Network,Dense Net）to replace the traditional linear convolution kernel network structure;Finally,an improved YOLOv4 network model is obtained.（3）In order to verify the feasibility of the improved algorithm,This Thesis detects the actual construction video and analyzes the results.The experimental results show that the improved YOLOv4 model has improved both the average detection accuracy and detection frame rate.The average average precision（mAP）of the improved YOLOv4 algorithm is91.92%,and the average detection frame rate（FPS）of the improved YOLOv4 model reaches26.228f.s^-1,indicating that the improved YOLOv4 algorithm achieves the expected effect and meets the requirements of real-time detection tasks.（4）This Thesis has developed a visual intelligent detection platform,which can conduct real-time production safety monitoring and management of unsafe behaviors on construction sites,and achieve a convenient,efficient and intelligent supervision purpose.